Comparative Study of the Performance of SqueezeNet and GoogLeNet CNN Models for Identification of Kazakhstan Potatoes Varieties

With the growth of potato production in Kazakhstan, the development and implementation of digital technologies that improve grading productivity is a very relevant issue. A comparative analysis of two types of classical deep neural network models used for classification of color im-ages of potatoes from the Kazakhstan region is presented in the article. Ten representative varie-ties of Kazakhstani potatoes were selected as objects of study: Alians, Alians mini, Astana, Astana mini, Edem, Edem mini, Nerli, Nerli mini, Zhanaisan and Zhanaisan mini. Two convolutional neural network (CNN) architectures - GoogLeNet and SqueezeNet - were fine-tuned using trans-fer learning with three different optimization methods. The convolutional neural networks SqeezeNet and GoogLeNet were used to identify the variety of the captured potato images. The comparison between the achieved classification accuracy with the two neural networks was made using standard evaluation metrics, such as accuracy, precision and recall, supplemented by con-fusion matrix analysis to reveal potential misclassifications. The analysis of the metrics showed that both neural network architectures are applicable for developing automated systems for iden-tifying the correspondence of tubers with specific varietal characteristics and sorting them in con-trolled laboratory image acquisition conditions. When analyzing the results for the ten studied va-rieties, several varieties were identified for which high recognition accuracy was obtained (Astana, Zhanaisan and Zhansyan mini), those that were not identified very well (Alliance, Alliance mini, Astana mini, Edem) and one variety that was poorly recognized - the Nerli variety. For the Astana and Zhanaisan varieties, accuracy rates exceeding 97% were achieved, making the models suitable for use in digital potato tuber sorting systems. For the Nerli and Alians varieties, further network training on a larger sample, including a wider range of color variations, is recommended.