Cropland Recognition Based on Collaborative Spatial Attention and Edge Detection for Multi-Source Remote Sensing Data

Real-time and accurate cropland recognition is conducive to scientific and rational planting management and utilization of agricultural resources. In recent years, deep learning methods have made great progress in recognizing cropland. However, there are still many challenges in experimenting with complex areas based on a single data source. In response to the above, this paper proposes a cropland recognition method based on cooperative spatial attention and edge detection with multi-source remote sensing data. Specifically, the edge extraction module obtains the gradient information of the image and extracts the edge features of the image. Then, through the spatial attention module, the global features are retained while the local texture information is highlighted. It is worth saying that ASPP ensures that features are extracted at the same time without losing information at the lowest resolution. We conducted experiments on the study area of Chengdu Plain, and the results showed that the OA, mIoU and F1 scores of USEA-Net reached 85.21%, 75.08% and 83.1%, respectively, which verified its effectiveness and superiority.