An Attention Network for Remote Sensing Image Classification Integrating Mamba_Attention and Lightweight Convolutions

We targets the trade-off in remote sensing image classification between insufficient exploitation of small features and high computational cost, and proposes a framework that balances accuracy, efficiency, and interpretability. Specifically, we introduce Mamba_Attention, an attention module built upon structured state space modeling, to capture long-range dependencies and global context in remote sensing imagery with lower computational overhead. Meanwhile, we incorporate depthwise separable convolution (DS Conv) into a ResNet backbone, replacing key 3×3 convolutional modules to significantly reduce parameters and computation, thereby accelerating inference. Extensive experiments on multiple remote sensing classification benchmarks demonstrate that the proposed method achieves superior classification performance while improving computational efficiency and enhancing interpretability.