Depthwise Separable Convolution Based Lightweight HSRRS Image Classification Method.

With the development of satellite remote sensing technology, the classification and recognition of high spatial resolution remote sensing (HSRRS) images are widely used in many fields. Convolutional neural network (CNN), as a method with superior performance, has achieved amazing results in the HSRSS classification task. However, the common CNN model with excellent performance generally has a large model size and the slow computation speed. In order to solve this problem, Depthwise separable (DS) convolution is applied to design a lightweight CNN to achieve efficient HSRSS classification. The simulation results show that the DS-CNN with the slight performance loss has realized the compression and acceleration of the original CNN model.