Research on Multi-Feature Triplet Attention Based Sharpness-Aware Classification Network

In addressing the issue of convolutional neural networks struggling to distinguish important features in image feature extraction and aiming to enhance the model’s generalization capabilities, this paper introduces a Sharpness-Aware Image Classification Network based on Multi-Feature Triplet Attention and designs a Multi-Feature Triplet Convolution Attention model to obtain channel feature weights and spatial feature weights from feature maps, allowing these features to interact and enabling the network to effectively discern crucial features. Additionally, it incorporates the Sharpness-Aware Minimization algorithm to seek neighborhood parameters with consistently low loss values, minimizing the loss and enhancing the model’s generalization capabilities. Experiments conducted on the CIFAR-10 and SVHN datasets demonstrate a significant improvement in classification accuracy compared to traditional neural networks.