A Lightweight and Fine-Grained Feature Fusion Network for Remote Sensing Scene Classification

Deep learning (DL) has become a hot topic in the research field of high-resolution remote sensing image (HRRSI) scene classification in recent years. Although traditional convolutional neural network (CNN) has achieved excellent achievements in scene classification, with the increasing depth and size of deep learning methods, its application in mobile and embedded vision applications has brought significant challenges. We present a novel lightweight and fine-grained feature fusion network for classification with low model parameters and high accuracy to address this issue. This paper uses MobileNetV2 as the backbone, and the multi-scale features are extracted in a more fine-grained way. The receptive field was extended by intergroup fusion depthwise convolution (IFDW Conv). We also use squeeze and excitation (SE)-block to extract dependence between channels. Therefore, we call our model SE-IFDWNet. Ablation experiments verify the effectiveness of IFDW Conv and SE-block proposed in our network. Experimental results on three standard HRRSI datasets show that SE-IFDWNet has advantages in classification accuracy and efficiency compared with the current state-of-the-art methods.