Cat Breed Classification with YOLOv11 and Optimized Training

Abstract

Accurate identification of cat breeds poses a significant challenge due to subtle inter-breed differences and intra-breed variability. This study leverages YOLOv11, the latest version of the YOLO family, to address these challenges through advanced deep-learning techniques. By training on a dataset consisting of five distinct cat breeds (Persian, Maine Coon, Siamese, Pallas's Cat, and Bengal), the model demonstrates exceptional capability in identifying nuanced breed-specific features. Data augmentation techniques were employed to enhance the dataset's diversity, while various optimization algorithms (Adam, Adamax, NAdam, AdamW, RAdam, RMSProp, and SGD) were evaluated to optimize the performance of the model. Experimental results showed that RAdam and SGD emerged as the top-performing optimizers, achieving an average recall of 96.8%, precision of 97.2%, and mAP50 of 98.1%, significantly outperforming other optimization methods. In contrast, RMSProp exhibited the lowest performance, particularly in terms of precision and mean Average Precision (mAP50). Additionally, data augmentation techniques were applied to enhance the diversity of the dataset, improving the robustness of the model. These findings highlight the effectiveness of YOLOv11 in cat breed classification, with potential applications in pet identification, animal conservation, and veterinary diagnostics.