FRMT: RMT with frequency feature learning framework for land cover classification

The classification of land cover is of great significance in the management and sustainable development of land resources. In recent years, deep learning methods have made remarkable progress in the field of remote sensing image processing. However, due to the complex difference in structure and scale of surface objects in remote sensing images, most of the current methods are prone to the problem of unclear boundary extraction, and the classification results of different scale surface objects are poor. In this letter, we proposed FRMT by introducing frequency domain information into RMT framework. A frequency feature learning module (FFLM) is proposed. The frequency domain features are decomposed into high frequency and low frequency features, and the frequency features are learned by spiking convolution to strengthen the model's learning of boundary and different scale features. We performed experimental validation on the very challenging multi-class remote sensing dataset GID-15 and Huawei Ascend Cup competition dataset, and the experimental results show that our model achieves the best results compared to five other recently proposed models, which mIoU achieves 77.91% and 71.85%, respectively.