Mamba-CNN: Remote Sensing Image Scene Classification Network with Cross-fusion of Long Sequence and Short Sequence Features

Recently the state space models (SSMs) with efficient hardware-aware designs, namely the Mamba deep learning model, has shown great potential in long sequence modeling. Therefore, we propose a Mamba model suitable for remote sensing image scene classification, aiming to reduce computational complexity to linear while retaining the efficient modeling ability of long sequences. In order to enhance the adaptability of our model in processing remote sensing image, we introduce HWC-Mamba to achieve 1D selective scanning in a three-dimensional image space with a global receptive field. In addition, to compensate for the shortcomings of Mamba in effectively capturing short distance features, we have introduced Convolutional Neural Networks (CNN) to enhance the model's feature extraction ability for remote sensing image by integrating Mamba and CNN features. Experiments on two public remote sensing image datasets (AID and UCM) have demonstrated the feasibility and effectiveness of the proposed method for remote sensing image scene classification. Our method can greatly reduce the manpower and material costs of remote sensing image scene classification tasks.