MAENet: Multiple Attention Encoder–Decoder Network for Farmland Segmentation of Remote Sensing Images

Abstract

With the rapid development of computer vision, semantic segmentation as an important part of the technology has made some achievements in different applications. However, in the farmland segmentation scenario of remote sensing images, the capability of common semantic segmentation methods in restoring the farmland edge and identifying narrow farmland ridges needs to be improved. Therefore, in this letter a semantic segmentation method–multiple attention encoder–decoder network (MAENet)–for farmland segmentation is proposed. The design of a dual-pooling efficient channel attention (DPECA) module and its embedment in the backbone to improve the efficiency of feature extraction is described; secondly, a dual-feature attention (DFA) module is proposed to extract contextual information of high-level features; finally, a global-guidance information upsample (GIU) module is added to the decoder to reduce the influence of redundant information on feature fusion. We use three self-made farmland image datasets representing UAV data to train MAENet and compare them with other methods. The results show that the performances of segmentation and generalization of MAENet are improved compared with other methods. The MIoU and Kappa coefficient in the farmland multi-classification test set can reach 93.74% and 96.74%.