Fine Grain Classification of Dog Breeds Based on HERBS Method

In an AI system, performing a fine-grained task like sorting dog breeds requires detailed observation. Therefore, the task of classifying images into highly specific and detailed categories, including birds [WBW+ 11], dogs [KJYL11], and medical images [ZWH+ 20] is known as fine-grained visual classification (FGVC). An effective FGVC model is high-temperature refinement and background suppression (HERBS) [CKL23]. The HERBS model can be integrated into various backbone networks, so in this article we apply the HERBS model to identify dog breeds in images and introduce the HERBS model from a technical point of view, outlining our model setup. The performance of HERBS model in the task of dog breed classification is also illustrated. This paper explores the application of deep learning methods in fine-grained classification of dog breeds, and experiments prove the effectiveness of the implemented HERBS method, demonstrating its ability to outperform other well-known classification methods on various datasets.