OSEA, a deep learning-based bird classification tool, with pre-trained model, mobile and command line applications

This is a Preprint and has not been peer reviewed. This is version 1 of this Preprint. You must log in to post a comment. There are no comments or no comments have been made public for this article. This is a Preprint and has not been peer reviewed. This is version 1 of this Preprint. Add a Comment You must log in to post a comment. Comments There are no comments or no comments have been made public for this article. In response to the challenges of traditional biodiversity monitoring methods, we introduce OSEA (Open Species Estimation for Avians), a multi-platform, offline tool for bird species identification. Designed to recognize over 10,000 bird species, OSEA includes both a mobile application and a command-line interface (CLI), facilitating efficient bird species identification. The mobile app, developed using Flutter, offers cross-platform compatibility and integrates a pre-trained ResNet34 model. The CLI, suited for batch processing, allows users to process images offline, with geographic filtering capabilities for enhanced accuracy. OSEA utilizes the DongNiao International Birds 10000 Dataset (DIB-10K), which undergoes rigorous data cleaning to ensure quality and accuracy. The system's core model leverages the MetaFGNet architecture, trained on high-performance computing resources to achieve 90.8% accuracy on the training set and 87.6% accuracy on the validation set. Additionally, the mobile app and CLI incorporate species distribution data for efficient geographical filter. OSEA addresses significant challenges in biodiversity research, including the time-intensive nature of manual species identification and the limited availability of offline tools for large-scale image analysis. By offering an accessible, offline solution, OSEA empowers amateur birders, educators, and conservationists, particularly in regions with limited internet access. Furthermore, the tool’s compatibility with custom models allows flexibility for broader wildlife applications beyond birds. In conclusion, OSEA offers a practical, scalable, and user-friendly solution for bird species identification, contributing to the acceleration of biodiversity studies and conservation efforts. Future developments will focus on expanding the dataset, optimizing performance, and incorporating more underrepresented species, further enhancing the tool’s robustness and global applicability. https://doi.org/10.32942/X2FP6T Biodiversity, Bioinformatics, Life Sciences, Ornithology bird identification, biodiversity, command-line interface, Deep Learning, image classification, mobile application, biodiversity, command-line interface, Deep learning, image classification, Mobile Application Published: 2025-01-07 10:16 Conflict of interest statement: The author has no conflict of interests. Data and Code Availability Statement: Code for data processing and CLI software: https://github.com/sun-jiao/osea Code for mobile application: https://github.com/sun-jiao/osea_mobile The modified MetaFGNet code to adapt to the data structure of DIB-10K: https://github.com/sun-jiao/MetaFGNet Some binary files are not included in git, they can be found in the release section. Language: English