Aiming at the problem of insufficient generalization ability of single model method in fine-grained image classification, a dy-namic weighted multi-model fusion method for fine-grained image classification is proposed in this paper. This method uses the network model based on the attention mechanism as the sub-model participating in the fusion, and at the same time, during the model training process, a weight adaptive adjustment algorithm is proposed. According to the actual performance of the sub model in each training, the algorithm can adaptively adjust its weight value to ensure that the whole model reaches the optimal state. The experimental results show that, compared with the traditional single-model method, the method in this paper improves the classification effect while the mod-el performance is more stable, and it performs well in complex background classification tasks, with stronger practical significance.
维普
万方数据