Aiming at the issues of insufficient perception of edge details and inadequate interaction between channel and spatial information in residual image classification networks, this paper proposes a Mach band mechanism-based image classification network (MBMNet). Built upon the ResNet-34 architecture, the MBMNet introduces a dynamic Mach band module (DMM) and a Machband attention module (MBM). Specifically, the DMM utilizes depthwise convolutions and the Laplacian operator for edge detection, employing a learnable scaling factor to strike a dynamic balance between edge enhancement and suppression, thereby intensifying the network's perception of critical edges. The MBM constructs a channel enhancement branch (CEB) and a spatial enhancement branch (SEB) across their respective dimensions, embedding the DMM into both. The CEB extracts key channel information via mixed pooling and depthwise separable convolutions, synergizing with the DMM to reinforce intra-channel edge and structural variations. Concurrently, the SEB extracts local spatial textures and structural features through multi-layer depthwise separable convolutions, leveraging the DMM to amplify spatial edge responses in critical regions. Finally, a feature fusion module (FFM) integrates the outputs from both branches to facilitate the collaborative modeling and information exchange of channel and spatial features, further employing residual fusion to preserve semantic consistency and ensure stable feature propagation. Experimental results demonstrate that the MBMNet effectively bolsters the perception of edges and structural details while promoting synergistic channel-spatial interactions, thereby significantly improving the accuracy and robustness of image classification.