Individual identification of endangered amphibians using deep learning and smartphone images: case study of the Japanese giant salamander (Andrias japonicus)

Individual identification of endangered amphibians using deep learning and smartphone images: case study of the Japanese giant salamander (Andrias japonicus) | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Individual identification of endangered amphibians using deep learning and smartphone images: case study of the Japanese giant salamander (Andrias japonicus) Kosuke Takaya, Yuki Taguchi, Takeshi Ise This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-2559407/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 27 Sep, 2023 Read the published version in Scientific Reports → Version 1 posted 11 You are reading this latest preprint version Abstract Information obtained via individual identification is invaluable for ecology and conservation. Physical tags, such as PIT tags and GPS, have been used for individual identification; however, their impact on animal behavior and survival rates is unclear and the tags may become lost. Although non-invasive methods that do not affect the target species (such as manual photoidentification) are available, these techniques utilize stripes and spots that are unique to the individual, which requires training, and applying them to large datasets is challenging. Many studies that have applied deep learning for identification have focused on species-level identification; however, few have addressed individual-level identification. In this study, we developed an image-based identification method with deep learning using the head spot of the Japanese giant salamander ( Andrias japonicus ), an endemic and endangered species in Japan. We trained and evaluated the dataset collected over two days from 11 individuals in captivity, including 7,075 images from the smartphone camera. Photographing was conducted three times a day at approximately 11:00 (morning), 15:00 (evening), and 18:00 (afternoon). As a result, individual identification by EfficientNet-V2 achieved 99.86% accuracy, 0.99 Kappa coefficient, and 0.99 F1 score. Performance was lower in the evening than in the morning or afternoon when trained and evaluated at each photographing time. This method does not require direct contact with the target species, and the effect on the animals is minimal; moreover, individual-level information can be obtained under natural conditions. In the future, smartphone images can be applied to citizen science surveys and individual-level big data collection, which is difficult to perform using current methods. Biological sciences/Ecology Biological sciences/Zoology Andrias japonicus computer vision deep learning individual identification photo identification Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Introduction Individual identification of wildlife provides fundamental information for ecological studies and conservation efforts [ 1 , 2 ] . For example, individual-based studies combined with mark-recapture methods offer estimates of population size, survival and reproduction rates, and immigration and emigration rates. In addition, health status indicators, such as weight and presence of parasites, and behavior variations are important in behavioral evolution and urban adaptation and can be revealed through individual identification. These findings can contribute to answering ecological and evolutionary questions and facilitating the conservation of endangered species [ 3 ] . Individual identification is essential for collecting information at the individual level. There are two main types of approaches: invasive and noninvasive. Invasive methods, such as blood and tissue sampling and the attachment of physical tags, GPS, and radio transmitters, have been adopted. However, capturing target species in the field is not easy, and concerns have been raised about the negative effects of tag attachment [ 4 – 7 ] . Additionally, research permits are required for certain animals, although obtaining such permits is particularly difficult for endangered species. Furthermore, tags rarely last long and tag loss can sometimes occur [ 8 ] , which represents a challenge when applied to long-lived organisms. Although these issues can be addressed by non-invasive methods, such as genetic analysis using DNA in feces and hairs, collecting samples from aquatic organisms, such as amphibians, is not easy because feces diffuse in the water. In addition, DNA analysis is expensive and requires fresh samples. Biometric identification techniques, such as manual photoidentification, are inexpensive and can identify individuals without harming animals. These methods have several advantages, such as no animal capture, no tags lost, and no effect on animal behavior, because individual-specific patterns, such as stripes and spots, can be utilized for identification, which is beneficial for the study of endangered species. For example, individual identification of cetaceans [ 9 ] , sea lions [ 10 ] , lions [ 11 ] , polar bears [ 12 ] , African elephants [ 13 ] , and sea turtles [ 8 , 14 ] . Such methods are alternatives to invasive methods, although they cannot be applied when natural markings are absent [ 15 ] . However, as the number of individuals increases, image classification requires more time and effort; therefore, processing large datasets is difficult. In recent years, computer vision has attracted attention as a method of overcoming the challenges of manual photoidentification. Deep learning, such as convolutional neural networks (CNNs), is a new approach for automatically extracting features from large amounts of data. Its implementation in various fields is rapidly advancing with improvements in computing power, such as Graphics Processing Units (GPUs). This technique, which has been used for human facial recognition, was first applied for animal identification in 2014 [ 16 ] . The target species for individual identification are mainly mammals [ 17 – 20 ] , although the method has also been applied to birds and reptiles, with large research bias observed according to taxonomic group [ 8 , 21 ] . On the other hand, pattern recognition was used to identify amphibian individuals [ 22 ] ; previous studies have not applied image recognition with deep learning. Amphibian populations are declining globally, with 41% of amphibians listed as threatened by extinction on the IUCN Red List [ 23 ] . Therefore, the application of deep learning is needed in this taxon which is a high conservation priority [ 24 , 25 ] . The Japanese giant salamander ( Andrias japonicus ) is one of the world's largest amphibians and endemic species and reaches 150 cm in total length, and it is distributed in the up-streams and middle rivers of western Japan [ 26 ] . This species has primitive morphological features similar to those of fossil species and is known to have a life span of over 60 years. Their diet is carnivorous, including fish and crabs, and they are top predators in the stream ecosystem. Although this species is protected by law, its population has declined because of habitat modifications and fragmentation [ 27 ] . Therefore, it is listed as a vulnerable species by the IUCN and Ministry of the Environment's Red List. Furthermore, hybridization of this species with the Chinese giant salamander ( Andrias davidianus ) has become a problem that requires immediate conservation efforts. Currently, PIT tags were mainly used to identify individuals of this species; however, capture is necessary to insert tags. Before PIT tags became popular, spot patterns on their bodies were mainly used for identification by experts capable of identifying individuals using the unique spots. In this study, we aimed to identify individuals of the Japanese giant salamander via deep learning using spot patterns captured by smartphone. Individual identification from images is a non-contact, non-destructive, and low-cost method that can be applied to other species. Because actual conservation practices are expensive, inexpensive identification methods will provide conservation opportunities for more species. In particular, charismatic species for conservation can easily attract people's attention, and they are mainly mammals; however, few amphibian species have been treated before [ 28 , 29 ] , which increases the difficulty of obtaining financial support for conservation. Therefore, inexpensive identification can significantly contribute to the conservation of amphibians. Materials & Methods Ethics declarations Japanese giant salamanders are protected by the Law for the Protection of Cultural Properties as a “National Natural Monument”. Therefore, this study was approved by the Hiroshima City Asa Zoological Park, which has permission from the Agency for Cultural Affairs and is categorized as a non-invasive study. Natural Markings The markers used for individual identification should be permanent, distinctive of the class of animals, and universally displayed throughout the population, and they should also be measurable using a recording device [ 30 ] . In addition, a suitable measurement region for the target species must be determined. The Japanese giant salamander has spot patterns all over its body, and the head and tail spots have been used to identify individuals [ 31 ] . A study showed that about 8 years continuous identification could be conducted by the body pattern of the species [ 32 ] . Therefore, the head of this species was selected as the measurement region and spots on the head were selected as the variable for measurement because these spots were clearly observed and easily photographed by the camera. Image Acquisition Eleven individuals kept at the Conservation breeding facility of Japanese giant salamanders in Hiroshima City Asa Zoological Park were used in this study (Supplementary Fig. 1). This facility has been used for researching and breeding Japanese giant salamanders successfully in captivity in Japan since 1971. A smartphone (iPhone11 equipped with a 12-megapixel camera) was used for image acquisition. Obtaining optimal images on land was difficult because of the body surface reflection and active movement of the individuals. Therefore, we photographed salamanders in water using a camera above the water. The distance between the salamander's head and the camera was approximately 60 cm. Because reflections on the water surface were a problem with this method, the photographer held an umbrella to suppress reflections on the water surface and conducted shooting under the umbrella (Supplementary Fig. 2). Spots were photographed on video and converted to images at ten frames per second using the Free Video to JPG Converter software program, and these images were used as training and test images. Photographing was performed on August 20–21, 2022. Each video recording lasted approximately 30 s, and shooting was conducted three times a day at approximately 11:00, 15:00, and 18:00, which are defined as morning, afternoon, and evening, respectively. The images converted from the video taken on August 20 and 21 were used for training and testing, respectively. The framework employed in this study is illustrated in Fig. 1 . The head of the salamander in the image was automatically detected using YOLOv5 [ 33 ] . Annotation data were then created using the LabelImg annotation tool [ 34 ] . First, an image of the Japanese giant salamander was loaded using this software (Supplementary Fig. 3). Next, a rectangle was created to include only the head of the salamander. Finally, the rectangles were labeled "head" and the output format was the YOLOv5 format to train the model. After detecting the Japanese giant salamander head in this model, OpenCV was used to crop the head image at a size of 60% for the training and test images (Fig. 2 ; Table 1). Although the head image detected by YOLOv5 may contain a background, only an image of the spots was created by cropping at 60%. The training and test images were resized to 224 × 224 pixels because the size varied from image to image. Image Augmentation Augmentation is performed to prevent overfitting. The orientation of the individual in the image differs because it moves during the shooting; therefore, rotation and crop processing were added to identify salamanders, regardless of the direction of the measurement region. In addition, brightness, Gaussian noise, color jitter, and saturation processing were performed because the light conditions in the image were not uniform. The augmentation process was applied to each parameter with a probability of 50%. For example, applying rotation and cropping will result in the following three patterns of images: (1) both processes are applied, (2) either process is applied or not, and (3) neither process is applied. This process produces various training images. The augmented images were then randomly separated into training sets (70%) and validation sets (30%). Efficientnet-V2 In this study, EfficientNet-V2, an improved version of EfficientNet, was used for classification. EfficientNet is a convolutional neural network that achieves more efficient performance by uniformly scaling up the depth, width, and resolution while scaling down the model instead of arbitrarily scaling these factors, as observed in conventional practice [ 35 ] . For example, the ResNet architecture is scaled up by adding more layers to improve accuracy. However, this approach results in increased computational complexity and vanishing gradient problems. EfficientNet addresses this issue by exploring the relationship between increases in each dimension using a compound coefficient. In addition, unlike other CNN models, it uses a new activation function called Swish instead of a Rectifier Linear Unit (ReLU). EfficientNet-V2 is an improved version of EfficientNet with better training speed and parameter efficiency [ 36 ] . The EfficientNet-V2 model employs a neural architecture search (NAS) to optimize model accuracy, size, and training speed. In this case, the EfficientNetV2-B0 model was used as the network, and fine-tuning was performed using a pretrained model with the Imagenet21k dataset. Fine-tuning uses the weights of the trained model and can thus achieve high accuracy with a small number of training images. The number of epochs was set to 50, and the batch size was set to 32 for training. Adam was used as the optimizer, and the dropout was set to 0.3. In this study, early stopping was also employed to prevent overfitting and automatic termination was performed when the validation loss did not improve by more than 0.001 for five consecutive epochs, and the weights when the validation loss was the best were used for the best model. These analyses were performed using the NVIDIA DGX Station A100. Evaluation Metrics The overall accuracy, Cohen's Kappa coefficient, and macro average F1 score were used for evaluation. Although the overall accuracy is widely used in assessments, proper evaluations become difficult in cases of imbalanced data. Therefore, we also used Cohen's Kappa coefficient and the macro average F1 score, which can be used to assess unbalanced data. The models were evaluated at different periods: morning, afternoon, and evening. For example, test images taken in the morning were used to assess the models trained based on the morning images; in addition, mixed models that contained images from all periods (morning, afternoon, and evening) were created and verified. Results Morning model The results of the morning model are shown in Fig. 3 and Table 2. All individuals were correctly identified with an accuracy of 97.04%, a Kappa coefficient of 0.97, and an F1 score of 0.98. The identification results for each individual are presented in a confusion matrix. The vertical axis represents the ground truth, the horizontal axis represents the class predicted by the model, and each number represents an individual number. The number in each cell represents the number of identified images, and the color of each cell indicates the percentage of images per ground truth. For example, light blue indicates a ratio of 0.0, indicating that no image was classified as that cell. In contrast, dark blue indicates a ratio of 1.0, meaning that all ground-truth images were classified to that cell. The morning model misclassified individual No. 1 as No. 10 in 20/107 (18.69%) images and individual No. 3 as No. 10 in 8/102 (7.84%) images. In addition, individual No. 9 images were misclassified as No. 11 by 2/105 (1.90%) images. Afternoon Model The results of the afternoon model are shown in Fig. 4 and Table 2. All individuals were correctly identified with an accuracy of 94.36%, a Kappa coefficient of 0.94, and an F1 score of 0.92. The model misclassified 58/105 images (55.24%) of the individual No. 6, 36/105 images as No. 1, and 22/105 images as No. 9 (Fig. 4 ). In addition, individual No. 3 were misclassified as No. 8 by 3/86 images (3.49%). Evening Model The results of the evening model are shown in Fig. 5 and Table 2. Compared to the morning and afternoon models, the accuracy was lower, with an accuracy of 86.86%, Kappa coefficient of 0.85, and F1 score of 0.98. The evening model misclassified 124/147 images (84.35%) of individual No. 11 (Fig. 6 ), 122/147 images (82.99%) as No. 9, and 2/147 images (1.36%) as No. 8. In addition, 36/187 images (19.25%) of No. 8 individuals were misclassified, 35/187 images as No. 3, and 1/187 images as No. 4. In addition, 3/103 images (2.91%) of individual No. 5 were misidentified as No. 8 and 1/103 images (0.97%) of individual No. 2 were misclassified as No. 4. Mixed Model The mixed-model results are shown in Fig. 7 and Table 2. The accuracy was 99.86%, Kappa coefficient was 0.99, and F1 score was 0.99. Although there were some misidentifications in the images of individuals No. 2, No. 6, and No. 8, the model correctly identified almost all individuals. Discussion This study examines a new image-based identification method for endangered amphibians using deep learning. Most wildlife studies employ artificial tag attachments to identify individuals by capturing animals [ 14 ] . However, physical markings have several issues, such as the stress associated with capture and tag attachment and the impact of the marker itself. In contrast, non-invasive methods, such as photoidentification, have a lower impact on animals, although they also present certain disadvantages. For example, photographic matching requires identification skills and cannot be applied to large datasets because of its labor-intensive nature and human errors. Although deep learning can overcome these challenges, it has only been applied to a few taxa, which mainly include mammals. Thus, research on amphibians using deep learning is lacking; however, such work is required for conservation. Our study demonstrates the effectiveness of a new identification method using deep learning for one of the world's largest amphibians that is currently threatened with extinction. The head spot was suitable for individual identification, and an accuracy of 99.86% was achieved by applying EfficientNet-V2 to smartphone images without conventional feature extraction. The high performance obtained with smartphone images suggests that combining this method with citizen science could contribute to amphibian conservation; moreover, our approach could be applied to other amphibian species at a low cost. The accuracy of this mixed model was 99.86%, the Kappa coefficient was 0.99, and the F1 score was 0.99. Previous studies using deep learning to conduct individual identification have shown accuracies of 92.5% for chimpanzees [ 20 ] , 96.3% for pandas [ 18 ] , and 83.9% for brown bears [ 19 ] . One reason for the high accuracy in this study was the high similarity between the training and test images. While previous studies showed a significant variation in the date and location of training images, the training and test images used in this study were from only two days. Furthermore, the high accuracy was probably due to the lower variation in the images because the salamanders did not move much during shooting and the clear imagery obtained for the spots in water in a captive environment. When photographing salamanders on land, individuals move relatively fast and their body surfaces reflect light. These problems can be gentled in water. Because markings for individual identification can be recorded under natural conditions when water surface reflections are suppressed, photographing underwater individuals is recommended for aquatic amphibians. Another factor contributing to the high accuracy was that the images contained only the spots used for analysis. In fact, the accuracy was reduced when images were used without cropping, especially in the evening model (Supplementary Table 1). The performance of the evening model was lower than that of the morning and afternoon models, which may be due to the quality of the images resulting from the light conditions. In particular, individual No. 11 in the evening model was misclassified in 124/147 (84.35%) images. These images could not be classified correctly because the spots were not clearly photographed (Fig. 6 ). In addition, confusion may occur because the results of the feature extraction by AI are similar. Furthermore, the fact that only the head was used for analysis may have contributed to the limited information in the images. In the case of the expert, observers identified individuals based on the characteristic pattern of spots on the whole body, not just the head, to isolate slight differences between individuals. Therefore, verifying how performance varies depending on the areas for individual identification is necessary. The selection of the measurement region for identification is important and has a significant effect on the accuracy of the model. For example, in the case of seals, the accuracy was 59% for fur-based identification [ 37 ] but improved to 88% for face-based identification [ 17 ] . Arzoumanian et al. [ 38 ] achieved more than 90% pair image matching using flank (front dorsal region) spot patterns for the identification of whale sharks but reported that image matching failed when photographs were obtained at oblique angles of more than 30°. This study demonstrated the benefits of using deep learning to identify individuals; however, certain challenges should be noted. First, individual identification was conducted in captivity, which facilitated uniform shooting conditions. In situations where images were captured in the wild, the background, light conditions, and direction of the target species differ, which leads to high image variations, thereby affecting the identification accuracy. Prior studies have shown that the accuracy is lower when individual identification is conducted in the field than in captivity, with a reported 92% accuracy for chimpanzees in captivity and 77% accuracy in the wild [ 39 ] . In the future, individual identification using this method should be verified for practical application in the field. Second, this study did not use data obtained from different years; therefore, verification of long-term individual identification is needed. In particular, long-term monitoring is important for surveying long-lived target species and for their conservation. Finally, relatively large adult giant salamanders with a total length of over 50 cm generally showed a little change in their spots. However, few studies have examined changes in Japanese giant salamander spotted patterns over their lifetime, except for Tochimoto [ 32 ] , which showed continuous identification by body pattern for eight years. For other species, regions that do not change over long periods should be used for individual identification. For example, Bauwens et al. [ 40 ] analyzed images of over 900 European adders ( Vipera berus ) over 12 years and found that head-scale features did not change and were useful markers. Therefore, future studies are needed to determine the effects of aging and weight change on the Japanese giant salamander spot visibility. The approach implemented in this study, which combines image and deep learning, can identify target species inexpensively without the need for capture. Moreover, although conventional machine learning requires the design of species-specific algorithms for feature extraction, which is challenging to implement [ 16 , 41 ] , highly accurate individual identification was achieved in our study using images alone. This simplicity is a major advantage, and we believe that similar methods can be applied to other species of amphibians. The findings of this study can contribute to the conservation of amphibians, which are threatened with extinction worldwide. In addition, our research demonstrated the feasibility of individual identification using smartphone images. Smartphones are widely available at a low cost; therefore, applying this methodology will significantly advance conservation via combination with citizen science. For example, whale shark research has accumulated over 43,000 images with the help of 3,400 researchers and citizen scientists, and over 3,800 individuals have been identified [ 42 ] . Such a large database can prevent the illegal release of individuals and trade in wildlife [ 43 ] . For example, Japanese giant salamanders can go away on land from the streams by heavy rain, and sightings of such animals has attracted attention on social networking sites, such as Twitter. In this case, individual identification from the images may assist in determining and releasing the original habitat. Furthermore, identification of individuals may be possible after illegal capture. For example, a case of human transportation of the species over 100 km was revealed from Hyogo to Shiga Prefecture in 2022 by its PIT tag. The purpose of its transportation is unknown; however, this species is prohibited from unauthorized capture and movement by law. A citizen science survey can prevent illegal trade and the release of giant salamanders by constructing a database that allows for the matching of individuals. Moreover, ecological information such as age at maturity and migration patterns can be obtained by collecting images over a long period. Understanding life history is essential for conservation, particularly for long-lived species. Combining images with deep learning enables inexpensive long-term monitoring and offers new opportunities to contribute to ecology. Conclusion In this study, we applied deep learning to identify endangered amphibian, the Japanese giant salamander. Our study shows that deep learning-based individual identification, which was previously concentrated on mammals, is feasible for amphibians and that high performance can be achieved using smartphone images. Individuals could be identified with high accuracy by clearly photographing the head spots with suppressed reflection on the water surface. This method provides stress-free shooting of the target species in natural conditions without interfering with their movements. Image-based individual identification is noninvasive and inexpensive and can automatically process large datasets when combined with deep learning. In addition, considering the widespread popularity of smartphones, the application of this method will contribute to the conservation of species with natural markings. Declarations Acknowledgments We are very grateful to Hiroshima City Asa Zoological Park for their cooperation in our research on Japanese giant salamanders. With permission from the Agency for Cultural Affairs, the Asa Zoological Park is researching and breeding giant salamanders, a special natural monument. Author contributions Kosuke Takaya: Conceptualization, data collection, data analysis, interpretation, and preparation of the first original manuscript Yuki Taguchi: Conceptualization, data collection, interpretation, and suggestions for the original manuscript Takeshi Ise: Guided all steps of the analysis and manuscript preparation. Data availability statement All data that support the findings of this study are included within the article (and any supplementary files). Competing interests The authors declare no conflicts of interest associated with this manuscript. References Alberts, S. C. Social influences on survival and reproduction: Insights from a long-term study of wild baboons. J. Anim. Ecol. 88, 47–66 (2019). Festa-Bianchet, M., Côté, S. D., Hamel, S. & Pelletier, F. Long‐term studies of bighorn sheep and mountain goats reveal fitness costs of reproduction. J. Anim. Ecol. 88, 1118–1133 (2019). Clutton-Brock, T. & Sheldon, B. C. Individuals and populations: The role of long-term, individual-based studies of animals in ecology and evolutionary biology. Trends Ecol. Evol. 25, 562–573 (2010). Burley, N., Krantzberg, G. & Radman, P. Influence of colour-banding on the conspecific preferences of zebra finches. Anim. Behav. 30, 444–455 (1982). Funk, W. C., Donnelly, M. A. & Lips, K. R. Alternative views of amphibian toe-clipping. Nature 433, 193–193 (2005). Gauthier-Clerc, M. et al. Long–term effects of flipper bands on penguins. Proc. R. Soc. Lond. B 271 Supplement 6(suppl_6), S423-S426 (2004). Moorhouse, T. P. & MacDonald, D. W. Indirect negative impacts of radio-collaring: Sex ratio variation in water voles. J. Appl. Ecol. 42, 91–98 (2005). Carter, S. J. B., Bell, I. P., Miller, J. J. & Gash, P. P. Automated marine turtle photograph identification using artificial neural networks, with application to green turtles. J. Exp. Mar. Biol. Ecol. 452, 105–110 (2014). Bichell, L. M. V., Krzyszczyk, E., Patterson, E. M. & Mann, J. The reliability of pigment pattern-based identification of wild bottlenose dolphins. Mar. Mamm. Sci. 34, 113–124 (2018). Osterrieder, S. K., Salgado Kent, C., Anderson, C. J., Parnum, I. M. & Robinson, R. W. Whisker spot patterns: A noninvasive method of individual identification of Australian sea lions (Neophoca cinerea). J. Mammal. 96, 988–997 (2015). Pennycuick, C. J. & Rudnai, J. A method of identifying individual lions Panthera leo with an analysis of the reliability of identification. J. Zool. 160, 497–508 (1970). Anderson, C. J. R., Roth, J. D. & Waterman, J. M. Can whisker spot patterns be used to identify individual polar bears? J. Zool. 273, 333–339 (2007). Chui, S. Y. S. & Karczmarski, L. Everyone matters: Identification with facial wrinkles allows more accurate inference of elephant social dynamics. Mamm. Biol., 1–22 (2022). Schofield, G., Katselidis, K. A., Dimopoulos, P. & Pantis, J. D. Investigating the viability of photo-identification as an objective tool to study endangered sea turtle populations. J. Exp. Mar. Biol. Ecol. 360, 103–108 (2008). Dorning, J. & Harris, S. The challenges of recognising individuals with few distinguishing features: Identifying red foxes Vulpes vulpes from camera-trap photos. PLOS ONE 14, e0216531 (2019). Schneider, S., Taylor, G. W., Linquist, S. & Kremer, S. C. Past, present and future approaches using computer vision for animal re-identification from camera trap data. Methods Ecol. Evol. 10, 461–470 (2019). Birenbaum, Z. et al. SEALNET: Facial recognition software for ecological studies of harbor seals. Ecol. Evol. 12, e8851 (2022). Chen, P. et al. A study on giant panda recognition based on images of a large proportion of captive pandas. Ecol. Evol. 10, 3561–3573 (2020). Clapham, M., Miller, E., Nguyen, M. & Darimont, C. T. Automated facial recognition for wildlife that lack unique markings: A deep learning approach for brown bears. Ecol. Evol. 10, 12883–12892 (2020). Schofield, D. et al. Chimpanzee face recognition from videos in the wild using deep learning. Sci. Adv. 5, eaaw0736 (2019). Ferreira, A. C. et al. Deep learning-based methods for individual recognition in small birds. Methods Ecol. Evol. 11, 1072–1085 (2020). Gamble, L., Ravela, S. & McGarigal, K. Multi-scale features for identifying individuals in large biological databases: An application of pattern recognition technology to the marbled salamander Ambystoma opacum. J. Appl. Ecol. 45, 170–180 (2008). IUCN. https://www.iucnredlist.org/ , (2022). Cordier, J. M. et al. A global assessment of amphibian and reptile responses to land-use changes. Biol. Conserv. 253, 108863 (2021). Nori, J. et al. Amphibian conservation, land-use changes and protected areas: A global overview. Biol. Conserv. 191, 367–374 (2015). Matsui, M. & Hayashi, T. Genetic uniformity in the Japanese giant salamander, Andrias japonicus . Copeia. 1, 232–235 (1992). Taguchi, Y. & Natuhara, Y. Requirements for small agricultural dams to allow the Japanese giant salamander ( Andrias japonicus ) to move upstream. Japanese J. Conserv. Ecol. 14, 165–172 (2009). Bennett, J. R., Maloney, R. & Possingham, H. P. Biodiversity gains from efficient use of private sponsorship for flagship species conservation. Proc. Biol. Sci. 282, 20142693 (2015). Clucas, B., McHugh, K. & Caro, T. Flagship species on covers of US conservation and nature magazines. Biodivers. Conserv. 17, 1517–1528 (2008). Kühl, H. S. & Burghardt, T. Animal biometrics: Quantifying and detecting phenotypic appearance. Trends Ecol. Evol. 28, 432–441 (2013). Kobara. J. (1985). Japanese Giant Salamander. (Doubutsu sha, book in Japanese). Tochimoto, T., Ecological studies on the Japanese giant salamander, Andrias japonicus , in the Ichi River in Hyogo Prefecture (1) Marking of Animals for recognition. Jour. Jpn. Assoc. Zoo. Aqu . 31, 112–116 (1990). Redmon, J., Divvala, S., Girshick, R. & Farhadi, A., (2016). You only look once: Unified, real-time object detection in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (779–788). Tzutalin. LabelImg. GitHub . https://github.com/tzutalin/labelImg , (2015). Tan, M. & Le, Q., (2019). Efficientnet: Rethinking model scaling for convolutional neural networks in International conference on machine learning (6105–6114). PMLR. Tan, M. & Le, Q., (2021). Efficientnetv2: Smaller models and faster training in International Conference on Machine Learning (10096–10106). PMLR. Cunningham, L. Using computer-assisted photo-identification and capture-recapture techniques to monitor the conservation status of harbour seals (Phoca vitulina). Aquat. Mamm. 35, 319–329 (2009). Arzoumanian, Z., Holmberg, J. & Norman, B. An astronomical pattern-matching algorithm for computer‐aided identification of whale sharks Rhincodon typus. J. Appl. Ecol. 42, 999–1011 (2005). Freytag, A. et al. Chimpanzee faces in the wild: Log-euclidean CNNs for predicting identities and attributes of primates in Ger. conference on pattern recognition (51–63) (Springer, Cham, 2016). Bauwens, D., Claus, K. & Mergeay, J. Genotyping validates photo-identification by the head scale pattern in a large population of the European adder (Vipera berus). Ecol. Evol. 8, 2985–2992 (2018). Bolger, D. T., Morrison, T. A., Vance, B., Lee, D. & Farid, H. A computer-assisted system for photographic mark–recapture analysis. Methods Ecol. Evol. 3, 813–822 (2012). Holmberg, J., Norman, B. & Arzoumanian, Z. Estimating population size, structure, and residency time for whale sharks Rhincodon typus through collaborative photo-identification. Endang. Species Res. 7, 39–53 (2009). Hiby, L. et al. A tiger cannot change its stripes: Using a three-dimensional model to match images of living tigers and tiger skins. Biol. Lett. 5, 383–386 (2009). Tables Table 1 Dataset summary. Date, time, and purpose of the images in each individual. Table 2. Identification results and comparison of each model. Additional Declarations No competing interests reported. Supplementary Files 230205Supplementaryinformation.pdf Cite Share Download PDF Status: Published Journal Publication published 27 Sep, 2023 Read the published version in Scientific Reports → Version 1 posted Editorial decision: Major revision 27 Apr, 2023 Reviews received at journal 26 Apr, 2023 Reviewers agreed at journal 19 Apr, 2023 Reviews received at journal 11 Mar, 2023 Reviewers agreed at journal 01 Mar, 2023 Reviewers agreed at journal 28 Feb, 2023 Reviewers invited by journal 28 Feb, 2023 Editor assigned by journal 24 Feb, 2023 Editor invited by journal 24 Feb, 2023 Submission checks completed at journal 24 Feb, 2023 First submitted to journal 07 Feb, 2023 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-2559407","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":178573978,"identity":"78b3a662-c6c9-490d-a9da-87c3612bb13e","order_by":0,"name":"Kosuke Takaya","email":"data:image/png;base64,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","orcid":"","institution":"Kyoto University","correspondingAuthor":true,"submittingAuthor":false,"prefix":"","firstName":"Kosuke","middleName":"","lastName":"Takaya","suffix":""},{"id":178573980,"identity":"88173676-1555-410f-a063-bfb424bf9c22","order_by":1,"name":"Yuki Taguchi","email":"","orcid":"","institution":"Asa Zoo","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Yuki","middleName":"","lastName":"Taguchi","suffix":""},{"id":178573982,"identity":"a94b6b18-ca4a-4950-8dbe-9db8c45155af","order_by":2,"name":"Takeshi Ise","email":"","orcid":"","institution":"Kyoto University","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Takeshi","middleName":"","lastName":"Ise","suffix":""}],"badges":[],"createdAt":"2023-02-07 09:14:23","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-2559407/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-2559407/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s41598-023-40814-1","type":"published","date":"2023-09-27T15:01:37+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":33585658,"identity":"2af1611e-b943-4e0f-807e-bdeef04ddbbd","added_by":"auto","created_at":"2023-02-28 22:51:09","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":340260,"visible":true,"origin":"","legend":"\u003cp\u003eFramework of the classification model for identifying Japanese giant salamanders using smartphone photographs.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-2559407/v1/5738ce41728d26e3b6898f37.png"},{"id":33585659,"identity":"e353af54-595b-4c6c-89a7-984893316779","added_by":"auto","created_at":"2023-02-28 22:51:09","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":397391,"visible":true,"origin":"","legend":"\u003cp\u003eHead of a Japanese giant salamander detected by Yolov5 (red box); the image also includes the background. Cropping the head image to 60% of its original size (white box) produced an image that only included spots.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-2559407/v1/540a9ee238aa99eaa901d6b3.png"},{"id":33586348,"identity":"51997b01-a353-4986-9c9e-f477df6d3583","added_by":"auto","created_at":"2023-02-28 22:59:09","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":67347,"visible":true,"origin":"","legend":"\u003cp\u003eConfusion matrix of morning model. The vertical axis shows the ground truth, and the horizontal axis shows the model's prediction results. The numbers on the axes indicate the individual numbers of the Japanese giant salamanders; the number in each cell indicates the number of classified images; and the color of each cell indicates the percentage of images in each class.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-2559407/v1/5be6fc1d3419a94b932e02ec.png"},{"id":33586347,"identity":"076818b8-256a-4783-b56f-e04b50f41477","added_by":"auto","created_at":"2023-02-28 22:59:09","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":67173,"visible":true,"origin":"","legend":"\u003cp\u003eConfusion matrix of the afternoon model.\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-2559407/v1/7913c5f003437c0ef89e9f80.png"},{"id":33585662,"identity":"1a7a0e2a-0e0d-41c6-a09c-f23739c03bc5","added_by":"auto","created_at":"2023-02-28 22:51:09","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":68780,"visible":true,"origin":"","legend":"\u003cp\u003eConfusion matrix of the evening model.\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-2559407/v1/3e87e681d3d851fc64735a70.png"},{"id":33585664,"identity":"b0382f73-63a0-456a-a2ae-1c0daffbd0ff","added_by":"auto","created_at":"2023-02-28 22:51:09","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":946786,"visible":true,"origin":"","legend":"\u003cp\u003eMisclassifications of the evening model, including 124/147 images (84.35%) of individual No. 11, 122/147 images (82.99%) as individual No. 9, and 2/147 images (1.36%) as individual No. 8. A shows the spots for each individual, and B shows the test image of each individual.\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-2559407/v1/e9124d7fe7655bfa4dc8289e.png"},{"id":33585660,"identity":"e90133b7-c6ea-4291-84f6-6b3441df7402","added_by":"auto","created_at":"2023-02-28 22:51:09","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":74120,"visible":true,"origin":"","legend":"\u003cp\u003eConfusion matrix of the mixed model.\u003c/p\u003e","description":"","filename":"7.png","url":"https://assets-eu.researchsquare.com/files/rs-2559407/v1/70e0841dad333e0be597635c.png"},{"id":43974508,"identity":"d231426c-3602-4ed3-9409-0d0f1752a3fa","added_by":"auto","created_at":"2023-10-02 15:08:11","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2180178,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-2559407/v1/8a13dbab-7472-441e-9b5b-eb45839eae6c.pdf"},{"id":33585665,"identity":"b1bf75d0-fcd7-4c1e-840f-1639f850b9e1","added_by":"auto","created_at":"2023-02-28 22:51:09","extension":"pdf","order_by":9,"title":"","display":"","copyAsset":false,"role":"supplement","size":871776,"visible":true,"origin":"","legend":"","description":"","filename":"230205Supplementaryinformation.pdf","url":"https://assets-eu.researchsquare.com/files/rs-2559407/v1/c6f5ff7335ba79fc95469d1c.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Individual identification of endangered amphibians using deep learning and smartphone images: case study of the Japanese giant salamander (Andrias japonicus)","fulltext":[{"header":"Introduction","content":"\u003cp\u003eIndividual identification of wildlife provides fundamental information for ecological studies and conservation efforts\u003csup\u003e[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]\u003c/sup\u003e. For example, individual-based studies combined with mark-recapture methods offer estimates of population size, survival and reproduction rates, and immigration and emigration rates. In addition, health status indicators, such as weight and presence of parasites, and behavior variations are important in behavioral evolution and urban adaptation and can be revealed through individual identification. These findings can contribute to answering ecological and evolutionary questions and facilitating the conservation of endangered species \u003csup\u003e[\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eIndividual identification is essential for collecting information at the individual level. There are two main types of approaches: invasive and noninvasive. Invasive methods, such as blood and tissue sampling and the attachment of physical tags, GPS, and radio transmitters, have been adopted. However, capturing target species in the field is not easy, and concerns have been raised about the negative effects of tag attachment\u003csup\u003e[\u003cspan additionalcitationids=\"CR5 CR6\" citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]\u003c/sup\u003e. Additionally, research permits are required for certain animals, although obtaining such permits is particularly difficult for endangered species. Furthermore, tags rarely last long and tag loss can sometimes occur \u003csup\u003e[\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]\u003c/sup\u003e, which represents a challenge when applied to long-lived organisms. Although these issues can be addressed by non-invasive methods, such as genetic analysis using DNA in feces and hairs, collecting samples from aquatic organisms, such as amphibians, is not easy because feces diffuse in the water. In addition, DNA analysis is expensive and requires fresh samples.\u003c/p\u003e \u003cp\u003eBiometric identification techniques, such as manual photoidentification, are inexpensive and can identify individuals without harming animals. These methods have several advantages, such as no animal capture, no tags lost, and no effect on animal behavior, because individual-specific patterns, such as stripes and spots, can be utilized for identification, which is beneficial for the study of endangered species. For example, individual identification of cetaceans\u003csup\u003e[\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]\u003c/sup\u003e, sea lions\u003csup\u003e[\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]\u003c/sup\u003e, lions\u003csup\u003e[\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]\u003c/sup\u003e, polar bears\u003csup\u003e[\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]\u003c/sup\u003e, African elephants\u003csup\u003e[\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]\u003c/sup\u003e, and sea turtles\u003csup\u003e[\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]\u003c/sup\u003e. Such methods are alternatives to invasive methods, although they cannot be applied when natural markings are absent\u003csup\u003e[\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]\u003c/sup\u003e. However, as the number of individuals increases, image classification requires more time and effort; therefore, processing large datasets is difficult.\u003c/p\u003e \u003cp\u003eIn recent years, computer vision has attracted attention as a method of overcoming the challenges of manual photoidentification. Deep learning, such as convolutional neural networks (CNNs), is a new approach for automatically extracting features from large amounts of data. Its implementation in various fields is rapidly advancing with improvements in computing power, such as Graphics Processing Units (GPUs). This technique, which has been used for human facial recognition, was first applied for animal identification in 2014\u003csup\u003e[\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]\u003c/sup\u003e. The target species for individual identification are mainly mammals\u003csup\u003e[\u003cspan additionalcitationids=\"CR18 CR19\" citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]\u003c/sup\u003e, although the method has also been applied to birds and reptiles, with large research bias observed according to taxonomic group\u003csup\u003e[\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]\u003c/sup\u003e. On the other hand, pattern recognition was used to identify amphibian individuals\u003csup\u003e[\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]\u003c/sup\u003e; previous studies have not applied image recognition with deep learning. Amphibian populations are declining globally, with 41% of amphibians listed as threatened by extinction on the IUCN Red List\u003csup\u003e[\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]\u003c/sup\u003e. Therefore, the application of deep learning is needed in this taxon which is a high conservation priority \u003csup\u003e[\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eThe Japanese giant salamander (\u003cem\u003eAndrias japonicus\u003c/em\u003e) is one of the world's largest amphibians and endemic species and reaches 150 cm in total length, and it is distributed in the up-streams and middle rivers of western Japan\u003csup\u003e[\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]\u003c/sup\u003e. This species has primitive morphological features similar to those of fossil species and is known to have a life span of over 60 years. Their diet is carnivorous, including fish and crabs, and they are top predators in the stream ecosystem. Although this species is protected by law, its population has declined because of habitat modifications and fragmentation\u003csup\u003e[\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]\u003c/sup\u003e. Therefore, it is listed as a vulnerable species by the IUCN and Ministry of the Environment's Red List. Furthermore, hybridization of this species with the Chinese giant salamander (\u003cem\u003eAndrias davidianus\u003c/em\u003e) has become a problem that requires immediate conservation efforts. Currently, PIT tags were mainly used to identify individuals of this species; however, capture is necessary to insert tags. Before PIT tags became popular, spot patterns on their bodies were mainly used for identification by experts capable of identifying individuals using the unique spots.\u003c/p\u003e \u003cp\u003eIn this study, we aimed to identify individuals of the Japanese giant salamander via deep learning using spot patterns captured by smartphone. Individual identification from images is a non-contact, non-destructive, and low-cost method that can be applied to other species. Because actual conservation practices are expensive, inexpensive identification methods will provide conservation opportunities for more species. In particular, charismatic species for conservation can easily attract people's attention, and they are mainly mammals; however, few amphibian species have been treated before\u003csup\u003e[\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]\u003c/sup\u003e, which increases the difficulty of obtaining financial support for conservation. Therefore, inexpensive identification can significantly contribute to the conservation of amphibians.\u003c/p\u003e"},{"header":"Materials \u0026 Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eEthics declarations\u003c/h2\u003e \u003cp\u003eJapanese giant salamanders are protected by the Law for the Protection of Cultural Properties as a \u0026ldquo;National Natural Monument\u0026rdquo;. Therefore, this study was approved by the Hiroshima City Asa Zoological Park, which has permission from the Agency for Cultural Affairs and is categorized as a non-invasive study.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eNatural Markings\u003c/h3\u003e\n\u003cp\u003eThe markers used for individual identification should be permanent, distinctive of the class of animals, and universally displayed throughout the population, and they should also be measurable using a recording device\u003csup\u003e[\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]\u003c/sup\u003e. In addition, a suitable measurement region for the target species must be determined. The Japanese giant salamander has spot patterns all over its body, and the head and tail spots have been used to identify individuals\u003csup\u003e[\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]\u003c/sup\u003e. A study showed that about 8 years continuous identification could be conducted by the body pattern of the species\u003csup\u003e[\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]\u003c/sup\u003e. Therefore, the head of this species was selected as the measurement region and spots on the head were selected as the variable for measurement because these spots were clearly observed and easily photographed by the camera.\u003c/p\u003e\n\u003ch3\u003eImage Acquisition\u003c/h3\u003e\n\u003cp\u003eEleven individuals kept at the Conservation breeding facility of Japanese giant salamanders in Hiroshima City Asa Zoological Park were used in this study (Supplementary Fig.\u0026nbsp;1). This facility has been used for researching and breeding Japanese giant salamanders successfully in captivity in Japan since 1971. A smartphone (iPhone11 equipped with a 12-megapixel camera) was used for image acquisition. Obtaining optimal images on land was difficult because of the body surface reflection and active movement of the individuals. Therefore, we photographed salamanders in water using a camera above the water. The distance between the salamander's head and the camera was approximately 60 cm. Because reflections on the water surface were a problem with this method, the photographer held an umbrella to suppress reflections on the water surface and conducted shooting under the umbrella (Supplementary Fig.\u0026nbsp;2). Spots were photographed on video and converted to images at ten frames per second using the Free Video to JPG Converter software program, and these images were used as training and test images. Photographing was performed on August 20\u0026ndash;21, 2022. Each video recording lasted approximately 30 s, and shooting was conducted three times a day at approximately 11:00, 15:00, and 18:00, which are defined as morning, afternoon, and evening, respectively. The images converted from the video taken on August 20 and 21 were used for training and testing, respectively.\u003c/p\u003e \u003cp\u003eThe framework employed in this study is illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e1\u003c/span\u003e. The head of the salamander in the image was automatically detected using YOLOv5\u003csup\u003e[\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]\u003c/sup\u003e. Annotation data were then created using the LabelImg annotation tool\u003csup\u003e[\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]\u003c/sup\u003e. First, an image of the Japanese giant salamander was loaded using this software (Supplementary Fig.\u0026nbsp;3). Next, a rectangle was created to include only the head of the salamander. Finally, the rectangles were labeled \"head\" and the output format was the YOLOv5 format to train the model. After detecting the Japanese giant salamander head in this model, OpenCV was used to crop the head image at a size of 60% for the training and test images (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e2\u003c/span\u003e; Table\u0026nbsp;1). Although the head image detected by YOLOv5 may contain a background, only an image of the spots was created by cropping at 60%. The training and test images were resized to 224 \u0026times; 224 pixels because the size varied from image to image.\u003c/p\u003e\n\u003ch3\u003eImage Augmentation\u003c/h3\u003e\n\u003cp\u003eAugmentation is performed to prevent overfitting. The orientation of the individual in the image differs because it moves during the shooting; therefore, rotation and crop processing were added to identify salamanders, regardless of the direction of the measurement region. In addition, brightness, Gaussian noise, color jitter, and saturation processing were performed because the light conditions in the image were not uniform. The augmentation process was applied to each parameter with a probability of 50%. For example, applying rotation and cropping will result in the following three patterns of images: (1) both processes are applied, (2) either process is applied or not, and (3) neither process is applied. This process produces various training images. The augmented images were then randomly separated into training sets (70%) and validation sets (30%).\u003c/p\u003e\n\u003ch3\u003eEfficientnet-V2\u003c/h3\u003e\n\u003cp\u003eIn this study, EfficientNet-V2, an improved version of EfficientNet, was used for classification. EfficientNet is a convolutional neural network that achieves more efficient performance by uniformly scaling up the depth, width, and resolution while scaling down the model instead of arbitrarily scaling these factors, as observed in conventional practice\u003csup\u003e[\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]\u003c/sup\u003e. For example, the ResNet architecture is scaled up by adding more layers to improve accuracy. However, this approach results in increased computational complexity and vanishing gradient problems. EfficientNet addresses this issue by exploring the relationship between increases in each dimension using a compound coefficient. In addition, unlike other CNN models, it uses a new activation function called Swish instead of a Rectifier Linear Unit (ReLU). EfficientNet-V2 is an improved version of EfficientNet with better training speed and parameter efficiency\u003csup\u003e[\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e]\u003c/sup\u003e. The EfficientNet-V2 model employs a neural architecture search (NAS) to optimize model accuracy, size, and training speed. In this case, the EfficientNetV2-B0 model was used as the network, and fine-tuning was performed using a pretrained model with the Imagenet21k dataset. Fine-tuning uses the weights of the trained model and can thus achieve high accuracy with a small number of training images. The number of epochs was set to 50, and the batch size was set to 32 for training. Adam was used as the optimizer, and the dropout was set to 0.3. In this study, early stopping was also employed to prevent overfitting and automatic termination was performed when the validation loss did not improve by more than 0.001 for five consecutive epochs, and the weights when the validation loss was the best were used for the best model. These analyses were performed using the NVIDIA DGX Station A100.\u003c/p\u003e\n\u003ch3\u003eEvaluation Metrics\u003c/h3\u003e\n\u003cp\u003eThe overall accuracy, Cohen's Kappa coefficient, and macro average F1 score were used for evaluation.\u003c/p\u003e \u003cp\u003eAlthough the overall accuracy is widely used in assessments, proper evaluations become difficult in cases of imbalanced data. Therefore, we also used Cohen's Kappa coefficient and the macro average F1 score, which can be used to assess unbalanced data. The models were evaluated at different periods: morning, afternoon, and evening. For example, test images taken in the morning were used to assess the models trained based on the morning images; in addition, mixed models that contained images from all periods (morning, afternoon, and evening) were created and verified.\u003c/p\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003eMorning model\u003c/h2\u003e \u003cp\u003eThe results of the morning model are shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e3\u003c/span\u003e and Table\u0026nbsp;2. All individuals were correctly identified with an accuracy of 97.04%, a Kappa coefficient of 0.97, and an F1 score of 0.98. The identification results for each individual are presented in a confusion matrix. The vertical axis represents the ground truth, the horizontal axis represents the class predicted by the model, and each number represents an individual number. The number in each cell represents the number of identified images, and the color of each cell indicates the percentage of images per ground truth. For example, light blue indicates a ratio of 0.0, indicating that no image was classified as that cell. In contrast, dark blue indicates a ratio of 1.0, meaning that all ground-truth images were classified to that cell. The morning model misclassified individual No. 1 as No. 10 in 20/107 (18.69%) images and individual No. 3 as No. 10 in 8/102 (7.84%) images. In addition, individual No. 9 images were misclassified as No. 11 by 2/105 (1.90%) images.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eAfternoon Model\u003c/h3\u003e\n\u003cp\u003eThe results of the afternoon model are shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e4\u003c/span\u003e and Table\u0026nbsp;2. All individuals were correctly identified with an accuracy of 94.36%, a Kappa coefficient of 0.94, and an F1 score of 0.92. The model misclassified 58/105 images (55.24%) of the individual No. 6, 36/105 images as No. 1, and 22/105 images as No. 9 (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e4\u003c/span\u003e). In addition, individual No. 3 were misclassified as No. 8 by 3/86 images (3.49%).\u003c/p\u003e\n\u003ch3\u003eEvening Model\u003c/h3\u003e\n\u003cp\u003eThe results of the evening model are shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e5\u003c/span\u003e and Table\u0026nbsp;2. Compared to the morning and afternoon models, the accuracy was lower, with an accuracy of 86.86%, Kappa coefficient of 0.85, and F1 score of 0.98. The evening model misclassified 124/147 images (84.35%) of individual No. 11 (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e6\u003c/span\u003e), 122/147 images (82.99%) as No. 9, and 2/147 images (1.36%) as No. 8. In addition, 36/187 images (19.25%) of No. 8 individuals were misclassified, 35/187 images as No. 3, and 1/187 images as No. 4. In addition, 3/103 images (2.91%) of individual No. 5 were misidentified as No. 8 and 1/103 images (0.97%) of individual No. 2 were misclassified as No. 4.\u003c/p\u003e\n\u003ch3\u003eMixed Model\u003c/h3\u003e\n\u003cp\u003eThe mixed-model results are shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e7\u003c/span\u003e and Table\u0026nbsp;2. The accuracy was 99.86%, Kappa coefficient was 0.99, and F1 score was 0.99. Although there were some misidentifications in the images of individuals No. 2, No. 6, and No. 8, the model correctly identified almost all individuals.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis study examines a new image-based identification method for endangered amphibians using deep learning. Most wildlife studies employ artificial tag attachments to identify individuals by capturing animals\u003csup\u003e[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]\u003c/sup\u003e. However, physical markings have several issues, such as the stress associated with capture and tag attachment and the impact of the marker itself. In contrast, non-invasive methods, such as photoidentification, have a lower impact on animals, although they also present certain disadvantages. For example, photographic matching requires identification skills and cannot be applied to large datasets because of its labor-intensive nature and human errors. Although deep learning can overcome these challenges, it has only been applied to a few taxa, which mainly include mammals. Thus, research on amphibians using deep learning is lacking; however, such work is required for conservation. Our study demonstrates the effectiveness of a new identification method using deep learning for one of the world's largest amphibians that is currently threatened with extinction. The head spot was suitable for individual identification, and an accuracy of 99.86% was achieved by applying EfficientNet-V2 to smartphone images without conventional feature extraction. The high performance obtained with smartphone images suggests that combining this method with citizen science could contribute to amphibian conservation; moreover, our approach could be applied to other amphibian species at a low cost.\u003c/p\u003e \u003cp\u003eThe accuracy of this mixed model was 99.86%, the Kappa coefficient was 0.99, and the F1 score was 0.99. Previous studies using deep learning to conduct individual identification have shown accuracies of 92.5% for chimpanzees\u003csup\u003e[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]\u003c/sup\u003e, 96.3% for pandas\u003csup\u003e[\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]\u003c/sup\u003e, and 83.9% for brown bears\u003csup\u003e[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]\u003c/sup\u003e. One reason for the high accuracy in this study was the high similarity between the training and test images. While previous studies showed a significant variation in the date and location of training images, the training and test images used in this study were from only two days. Furthermore, the high accuracy was probably due to the lower variation in the images because the salamanders did not move much during shooting and the clear imagery obtained for the spots in water in a captive environment. When photographing salamanders on land, individuals move relatively fast and their body surfaces reflect light. These problems can be gentled in water. Because markings for individual identification can be recorded under natural conditions when water surface reflections are suppressed, photographing underwater individuals is recommended for aquatic amphibians. Another factor contributing to the high accuracy was that the images contained only the spots used for analysis. In fact, the accuracy was reduced when images were used without cropping, especially in the evening model (Supplementary Table\u0026nbsp;1).\u003c/p\u003e \u003cp\u003eThe performance of the evening model was lower than that of the morning and afternoon models, which may be due to the quality of the images resulting from the light conditions. In particular, individual No. 11 in the evening model was misclassified in 124/147 (84.35%) images. These images could not be classified correctly because the spots were not clearly photographed (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e6\u003c/span\u003e). In addition, confusion may occur because the results of the feature extraction by AI are similar. Furthermore, the fact that only the head was used for analysis may have contributed to the limited information in the images. In the case of the expert, observers identified individuals based on the characteristic pattern of spots on the whole body, not just the head, to isolate slight differences between individuals. Therefore, verifying how performance varies depending on the areas for individual identification is necessary. The selection of the measurement region for identification is important and has a significant effect on the accuracy of the model. For example, in the case of seals, the accuracy was 59% for fur-based identification\u003csup\u003e[\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e]\u003c/sup\u003e but improved to 88% for face-based identification\u003csup\u003e[\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]\u003c/sup\u003e. Arzoumanian \u003cem\u003eet al.\u003c/em\u003e\u003csup\u003e[\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e]\u003c/sup\u003e achieved more than 90% pair image matching using flank (front dorsal region) spot patterns for the identification of whale sharks but reported that image matching failed when photographs were obtained at oblique angles of more than 30\u0026deg;.\u003c/p\u003e \u003cp\u003eThis study demonstrated the benefits of using deep learning to identify individuals; however, certain challenges should be noted. First, individual identification was conducted in captivity, which facilitated uniform shooting conditions. In situations where images were captured in the wild, the background, light conditions, and direction of the target species differ, which leads to high image variations, thereby affecting the identification accuracy. Prior studies have shown that the accuracy is lower when individual identification is conducted in the field than in captivity, with a reported 92% accuracy for chimpanzees in captivity and 77% accuracy in the wild\u003csup\u003e[\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e]\u003c/sup\u003e. In the future, individual identification using this method should be verified for practical application in the field. Second, this study did not use data obtained from different years; therefore, verification of long-term individual identification is needed. In particular, long-term monitoring is important for surveying long-lived target species and for their conservation. Finally, relatively large adult giant salamanders with a total length of over 50 cm generally showed a little change in their spots. However, few studies have examined changes in Japanese giant salamander spotted patterns over their lifetime, except for Tochimoto\u003csup\u003e[\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]\u003c/sup\u003e, which showed continuous identification by body pattern for eight years. For other species, regions that do not change over long periods should be used for individual identification. For example, Bauwens \u003cem\u003eet al.\u003c/em\u003e\u003csup\u003e[\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e]\u003c/sup\u003e analyzed images of over 900 European adders (\u003cem\u003eVipera berus\u003c/em\u003e) over 12 years and found that head-scale features did not change and were useful markers. Therefore, future studies are needed to determine the effects of aging and weight change on the Japanese giant salamander spot visibility.\u003c/p\u003e \u003cp\u003eThe approach implemented in this study, which combines image and deep learning, can identify target species inexpensively without the need for capture. Moreover, although conventional machine learning requires the design of species-specific algorithms for feature extraction, which is challenging to implement\u003csup\u003e[\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e, \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e]\u003c/sup\u003e, highly accurate individual identification was achieved in our study using images alone. This simplicity is a major advantage, and we believe that similar methods can be applied to other species of amphibians. The findings of this study can contribute to the conservation of amphibians, which are threatened with extinction worldwide. In addition, our research demonstrated the feasibility of individual identification using smartphone images. Smartphones are widely available at a low cost; therefore, applying this methodology will significantly advance conservation via combination with citizen science. For example, whale shark research has accumulated over 43,000 images with the help of 3,400 researchers and citizen scientists, and over 3,800 individuals have been identified\u003csup\u003e[\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e]\u003c/sup\u003e. Such a large database can prevent the illegal release of individuals and trade in wildlife\u003csup\u003e[\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e]\u003c/sup\u003e. For example, Japanese giant salamanders can go away on land from the streams by heavy rain, and sightings of such animals has attracted attention on social networking sites, such as Twitter. In this case, individual identification from the images may assist in determining and releasing the original habitat. Furthermore, identification of individuals may be possible after illegal capture. For example, a case of human transportation of the species over 100 km was revealed from Hyogo to Shiga Prefecture in 2022 by its PIT tag. The purpose of its transportation is unknown; however, this species is prohibited from unauthorized capture and movement by law. A citizen science survey can prevent illegal trade and the release of giant salamanders by constructing a database that allows for the matching of individuals. Moreover, ecological information such as age at maturity and migration patterns can be obtained by collecting images over a long period. Understanding life history is essential for conservation, particularly for long-lived species. Combining images with deep learning enables inexpensive long-term monitoring and offers new opportunities to contribute to ecology.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eIn this study, we applied deep learning to identify endangered amphibian, the Japanese giant salamander. Our study shows that deep learning-based individual identification, which was previously concentrated on mammals, is feasible for amphibians and that high performance can be achieved using smartphone images. Individuals could be identified with high accuracy by clearly photographing the head spots with suppressed reflection on the water surface. This method provides stress-free shooting of the target species in natural conditions without interfering with their movements. Image-based individual identification is noninvasive and inexpensive and can automatically process large datasets when combined with deep learning. In addition, considering the widespread popularity of smartphones, the application of this method will contribute to the conservation of species with natural markings.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgments\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe are very grateful to Hiroshima City Asa Zoological Park for their cooperation in our research on Japanese giant salamanders. With permission from the Agency for Cultural Affairs, the Asa Zoological Park is researching and breeding giant salamanders, a special natural monument.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor contributions\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eKosuke Takaya: Conceptualization, data collection, data analysis, interpretation, and preparation of the first original manuscript\u003c/p\u003e\n\u003cp\u003eYuki Taguchi: Conceptualization, data collection, interpretation, and suggestions for the original manuscript\u003c/p\u003e\n\u003cp\u003eTakeshi Ise: Guided all steps of the analysis and manuscript preparation.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData availability statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll data that support the findings of this study are included within the article (and any supplementary files).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no conflicts of interest associated with this manuscript.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eAlberts, S. C. Social influences on survival and reproduction: Insights from a long-term study of wild baboons. J. Anim. Ecol. 88, 47\u0026ndash;66 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFesta-Bianchet, M., C\u0026ocirc;t\u0026eacute;, S. D., Hamel, S. \u0026amp; Pelletier, F. Long‐term studies of bighorn sheep and mountain goats reveal fitness costs of reproduction. J. Anim. Ecol. 88, 1118\u0026ndash;1133 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eClutton-Brock, T. \u0026amp; Sheldon, B. C. Individuals and populations: The role of long-term, individual-based studies of animals in ecology and evolutionary biology. Trends Ecol. Evol. 25, 562\u0026ndash;573 (2010).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBurley, N., Krantzberg, G. \u0026amp; Radman, P. Influence of colour-banding on the conspecific preferences of zebra finches. Anim. Behav. 30, 444\u0026ndash;455 (1982).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFunk, W. C., Donnelly, M. A. \u0026amp; Lips, K. R. Alternative views of amphibian toe-clipping. Nature 433, 193\u0026ndash;193 (2005).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGauthier-Clerc, M. \u003cem\u003eet al.\u003c/em\u003e Long\u0026ndash;term effects of flipper bands on penguins. \u003cem\u003eProc. R. Soc. Lond. B\u003c/em\u003e 271 Supplement 6(suppl_6), S423-S426 (2004).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMoorhouse, T. P. \u0026amp; MacDonald, D. W. Indirect negative impacts of radio-collaring: Sex ratio variation in water voles. J. Appl. Ecol. 42, 91\u0026ndash;98 (2005).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCarter, S. J. B., Bell, I. P., Miller, J. J. \u0026amp; Gash, P. P. Automated marine turtle photograph identification using artificial neural networks, with application to green turtles. J. Exp. Mar. Biol. Ecol. 452, 105\u0026ndash;110 (2014).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBichell, L. M. V., Krzyszczyk, E., Patterson, E. M. \u0026amp; Mann, J. The reliability of pigment pattern-based identification of wild bottlenose dolphins. Mar. Mamm. Sci. 34, 113\u0026ndash;124 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOsterrieder, S. K., Salgado Kent, C., Anderson, C. J., Parnum, I. M. \u0026amp; Robinson, R. W. Whisker spot patterns: A noninvasive method of individual identification of Australian sea lions (Neophoca cinerea). J. Mammal. 96, 988\u0026ndash;997 (2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePennycuick, C. J. \u0026amp; Rudnai, J. A method of identifying individual lions Panthera leo with an analysis of the reliability of identification. J. Zool. 160, 497\u0026ndash;508 (1970).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAnderson, C. J. R., Roth, J. D. \u0026amp; Waterman, J. M. Can whisker spot patterns be used to identify individual polar bears? J. Zool. 273, 333\u0026ndash;339 (2007).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChui, S. Y. S. \u0026amp; Karczmarski, L. Everyone matters: Identification with facial wrinkles allows more accurate inference of elephant social dynamics. Mamm. Biol., 1\u0026ndash;22 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSchofield, G., Katselidis, K. A., Dimopoulos, P. \u0026amp; Pantis, J. D. Investigating the viability of photo-identification as an objective tool to study endangered sea turtle populations. J. Exp. Mar. Biol. Ecol. 360, 103\u0026ndash;108 (2008).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDorning, J. \u0026amp; Harris, S. The challenges of recognising individuals with few distinguishing features: Identifying red foxes Vulpes vulpes from camera-trap photos. PLOS ONE 14, e0216531 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSchneider, S., Taylor, G. W., Linquist, S. \u0026amp; Kremer, S. C. Past, present and future approaches using computer vision for animal re-identification from camera trap data. Methods Ecol. Evol. 10, 461\u0026ndash;470 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBirenbaum, Z. \u003cem\u003eet al.\u003c/em\u003e SEALNET: Facial recognition software for ecological studies of harbor seals. Ecol. Evol. 12, e8851 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen, P. \u003cem\u003eet al.\u003c/em\u003e A study on giant panda recognition based on images of a large proportion of captive pandas. Ecol. Evol. 10, 3561\u0026ndash;3573 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eClapham, M., Miller, E., Nguyen, M. \u0026amp; Darimont, C. T. Automated facial recognition for wildlife that lack unique markings: A deep learning approach for brown bears. Ecol. Evol. 10, 12883\u0026ndash;12892 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSchofield, D. \u003cem\u003eet al.\u003c/em\u003e Chimpanzee face recognition from videos in the wild using deep learning. Sci. Adv. 5, eaaw0736 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFerreira, A. C. \u003cem\u003eet al.\u003c/em\u003e Deep learning-based methods for individual recognition in small birds. Methods Ecol. Evol. 11, 1072\u0026ndash;1085 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGamble, L., Ravela, S. \u0026amp; McGarigal, K. Multi-scale features for identifying individuals in large biological databases: An application of pattern recognition technology to the marbled salamander Ambystoma opacum. J. Appl. Ecol. 45, 170\u0026ndash;180 (2008).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIUCN. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.iucnredlist.org/\u003c/span\u003e\u003cspan address=\"https://www.iucnredlist.org/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e, (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCordier, J. M. \u003cem\u003eet al.\u003c/em\u003e A global assessment of amphibian and reptile responses to land-use changes. Biol. Conserv. 253, 108863 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNori, J. \u003cem\u003eet al.\u003c/em\u003e Amphibian conservation, land-use changes and protected areas: A global overview. Biol. Conserv. 191, 367\u0026ndash;374 (2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMatsui, M. \u0026amp; Hayashi, T. Genetic uniformity in the Japanese giant salamander, \u003cem\u003eAndrias japonicus\u003c/em\u003e. Copeia. 1, 232\u0026ndash;235 (1992).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTaguchi, Y. \u0026amp; Natuhara, Y. Requirements for small agricultural dams to allow the Japanese giant salamander (\u003cem\u003eAndrias japonicus\u003c/em\u003e) to move upstream. Japanese J. Conserv. Ecol. 14, 165\u0026ndash;172 (2009).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBennett, J. R., Maloney, R. \u0026amp; Possingham, H. P. Biodiversity gains from efficient use of private sponsorship for flagship species conservation. \u003cem\u003eProc. Biol. Sci.\u003c/em\u003e 282, 20142693 (2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eClucas, B., McHugh, K. \u0026amp; Caro, T. Flagship species on covers of US conservation and nature magazines. Biodivers. Conserv. 17, 1517\u0026ndash;1528 (2008).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eK\u0026uuml;hl, H. S. \u0026amp; Burghardt, T. Animal biometrics: Quantifying and detecting phenotypic appearance. Trends Ecol. Evol. 28, 432\u0026ndash;441 (2013).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKobara. J. (1985). Japanese Giant Salamander. (Doubutsu sha, book in Japanese).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTochimoto, T., Ecological studies on the Japanese giant salamander, \u003cem\u003eAndrias japonicus\u003c/em\u003e, in the Ichi River in Hyogo Prefecture (1) Marking of Animals for recognition. \u003cem\u003eJour. Jpn. Assoc. Zoo. Aqu\u003c/em\u003e. 31, 112\u0026ndash;116 (1990).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRedmon, J., Divvala, S., Girshick, R. \u0026amp; Farhadi, A., (2016). You only look once: Unified, real-time object detection in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (779\u0026ndash;788).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTzutalin. LabelImg. \u003cem\u003eGitHub\u003c/em\u003e. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://github.com/tzutalin/labelImg\u003c/span\u003e\u003cspan address=\"https://github.com/tzutalin/labelImg\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e, (2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTan, M. \u0026amp; Le, Q., (2019). Efficientnet: Rethinking model scaling for convolutional neural networks in International conference on machine learning (6105\u0026ndash;6114). PMLR.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTan, M. \u0026amp; Le, Q., (2021). Efficientnetv2: Smaller models and faster training in International Conference on Machine Learning (10096\u0026ndash;10106). PMLR.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCunningham, L. Using computer-assisted photo-identification and capture-recapture techniques to monitor the conservation status of harbour seals (Phoca vitulina). Aquat. Mamm. 35, 319\u0026ndash;329 (2009).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eArzoumanian, Z., Holmberg, J. \u0026amp; Norman, B. An astronomical pattern-matching algorithm for computer‐aided identification of whale sharks Rhincodon typus. J. Appl. Ecol. 42, 999\u0026ndash;1011 (2005).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFreytag, A. \u003cem\u003eet al.\u003c/em\u003e Chimpanzee faces in the wild: Log-euclidean CNNs for predicting identities and attributes of primates in Ger. conference on pattern recognition (51\u0026ndash;63) (Springer, Cham, 2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBauwens, D., Claus, K. \u0026amp; Mergeay, J. Genotyping validates photo-identification by the head scale pattern in a large population of the European adder (Vipera berus). Ecol. Evol. 8, 2985\u0026ndash;2992 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBolger, D. T., Morrison, T. A., Vance, B., Lee, D. \u0026amp; Farid, H. A computer-assisted system for photographic mark\u0026ndash;recapture analysis. Methods Ecol. Evol. 3, 813\u0026ndash;822 (2012).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHolmberg, J., Norman, B. \u0026amp; Arzoumanian, Z. Estimating population size, structure, and residency time for whale sharks Rhincodon typus through collaborative photo-identification. Endang. Species Res. 7, 39\u0026ndash;53 (2009).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHiby, L. \u003cem\u003eet al.\u003c/em\u003e A tiger cannot change its stripes: Using a three-dimensional model to match images of living tigers and tiger skins. Biol. Lett. 5, 383\u0026ndash;386 (2009).\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003eTable 1 Dataset summary. Date, time, and purpose of the images in each individual.\u003c/p\u003e\n\u003cp\u003e\u003cimg width=\"527\" src=\"data:image/png;base64,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\" alt=\"image\"\u003e\u003c/td\u003e\n \u003c/p\u003e\n \n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTable \u0026nbsp;2. Identification results and comparison of each model.\u003c/p\u003e\n\u003cp\u003e\u003cimg src=\"data:image/png;base64,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\" alt=\"image\"\u003e\u0026nbsp;\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Andrias japonicus, computer vision, deep learning, individual identification, photo identification ","lastPublishedDoi":"10.21203/rs.3.rs-2559407/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-2559407/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eInformation obtained via individual identification is invaluable for ecology and conservation. Physical tags, such as PIT tags and GPS, have been used for individual identification; however, their impact on animal behavior and survival rates is unclear and the tags may become lost. Although non-invasive methods that do not affect the target species (such as manual photoidentification) are available, these techniques utilize stripes and spots that are unique to the individual, which requires training, and applying them to large datasets is challenging. Many studies that have applied deep learning for identification have focused on species-level identification; however, few have addressed individual-level identification. In this study, we developed an image-based identification method with deep learning using the head spot of the Japanese giant salamander (\u003cem\u003eAndrias japonicus\u003c/em\u003e), an endemic and endangered species in Japan. We trained and evaluated the dataset collected over two days from 11 individuals in captivity, including 7,075 images from the smartphone camera. Photographing was conducted three times a day at approximately 11:00 (morning), 15:00 (evening), and 18:00 (afternoon). As a result, individual identification by EfficientNet-V2 achieved 99.86% accuracy, 0.99 Kappa coefficient, and 0.99 F1 score. Performance was lower in the evening than in the morning or afternoon when trained and evaluated at each photographing time. This method does not require direct contact with the target species, and the effect on the animals is minimal; moreover, individual-level information can be obtained under natural conditions. In the future, smartphone images can be applied to citizen science surveys and individual-level big data collection, which is difficult to perform using current methods.\u003c/p\u003e","manuscriptTitle":"Individual identification of endangered amphibians using deep learning and smartphone images: case study of the Japanese giant salamander (Andrias japonicus)","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2023-02-28 22:51:04","doi":"10.21203/rs.3.rs-2559407/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Major revision","date":"2023-04-27T16:09:34+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2023-04-26T09:09:13+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"98c8993c-414e-4270-ba99-3e2c7a7c4ede","date":"2023-04-19T06:56:41+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2023-03-11T09:46:49+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"02e0092f-195e-4fbc-a1d8-467b6fab1d0f","date":"2023-03-01T08:38:50+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"0466be08-b782-4402-8639-d07aaaa4e8f1","date":"2023-03-01T04:50:42+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2023-02-28T13:24:47+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2023-02-24T08:24:46+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2023-02-24T05:40:13+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2023-02-24T05:37:27+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2023-02-07T09:10:56+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"02dd3671-ee55-436b-b8f4-6c5485ecc5f5","owner":[],"postedDate":"February 28th, 2023","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":19451386,"name":"Biological sciences/Ecology"},{"id":19451387,"name":"Biological sciences/Zoology"}],"tags":[],"updatedAt":"2023-10-02T15:04:32+00:00","versionOfRecord":{"articleIdentity":"rs-2559407","link":"https://doi.org/10.1038/s41598-023-40814-1","journal":{"identity":"scientific-reports","isVorOnly":false,"title":"Scientific Reports"},"publishedOn":"2023-09-27 15:01:37","publishedOnDateReadable":"September 27th, 2023"},"versionCreatedAt":"2023-02-28 22:51:04","video":"","vorDoi":"10.1038/s41598-023-40814-1","vorDoiUrl":"https://doi.org/10.1038/s41598-023-40814-1","workflowStages":[]},"version":"v1","identity":"rs-2559407","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-2559407","identity":"rs-2559407","version":["v1"]},"buildId":"-HB7Z8yhvgn0wM9Nzuekk","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}