“A new breakthrough has been made in the field of fine-grained image recognition. The research team proposes a fine-grained recognition network based on dual attention random selection of global context to address issues such as the easy neglect of small latent features and subtle differences in appearance. This network utilizes ConvNeXt as the backbone and introduces a dual attention random selection module to effectively focus on other potential small discriminative features. At the same time, combined with the global context attention module, deep semantic information is integrated into the middle layer, enhancing the ability to locate small features. In addition, innovative multi branch loss is proposed to guide the network to obtain diverse discriminative features by combining different branch features. On publicly available datasets such as Stanford cars, CUB-200-2011, FGVC Aircraft, and real-world vehicle dataset VMRURS, the network achieved high recognition accuracies of 95.2%, 92.1%, 94.0%, and 97.0%, respectively, outperforming other comparative methods and laying a solid foundation for the development of fine-grained image recognition.”