Fine-grained Object Recognition based on Multi-Scale Destruction and Construction Learning

With the development of image processing technology, how to effectively use computer vision technology for image classification has become an important research branch in the field of computer vision. Traditional image classification is to identify coarse-grained categories, such as using computer vision technology to distinguish meta-categories such as "dog", "grape" and "bird". In many practical applications, images need to be classified at a fine-grained level, such as distinguishing which subcategory the image belongs to in the "dog" category: "husky", "Samoyed" and "Alaska". Due to the characteristics of large intra-class differences and small inter-class differences in fine-grained image classification, fine-grained image classification has become a challenging research. This paper proposes multi-scale input (MS_DCL) and multi-scale detail boosting (MSDB_DCL) destruction and construction learning methods, so that the network can learn object and component-level discriminative information for fine-grained image classification. The experimental results show that MS_DCL is finer than other existing ones. The granular image classification method has higher recognition accuracy, and the test accuracy rate can reach 87.4% on the CUB200-2011 dataset.