Dual-attention metric network for few-shot fine-grained image classification

In fine-grained tasks, subtle differences in images are often reflected in the channel and spatial dimensions. However, existing classification methods typically overlook the effective fusion of cross-channel and spatial information, resulting in poor performance when handling complex details. To overcome the challenge of insufficient extraction of information, we propose the Dual-Attention Metic (DAM) network, which incorporates both the CAM and the SAM. CAM preserves cross-dimensional information through 3D reshaping and enhances the dependencies between channel and spatial dimensions via a Multi-Layer Perceptron (MLP). SAM utilizes two convolutional layers to fuse spatial information, focusing more effectively on spatial features. Furthermore, experiments on four datasets demonstrate that our method outperforms existing methods, achieving substantial improvements in performance.