Based on Multi-Feature Information Attention Fusion for Multi-Modal Remote Sensing Image Semantic Segmentation

Abstract

Semantic segmentation of remote sensing images are widely used in land census and agriculture. The scenes in remote sensing images are complex, easily affected by season, such as farmland. Besides, the size of the target in remote sensing image is different, the shape is irregular, and there is often the problem of missing detection, so the multi-source data information is directly fed into the neural network, resulting in fuzzy segmentation boundary, which is difficult to achieve fine segmentation. To solve this problem, we propose a Dual-way Feature attention Fusion Network (DFFNet), which consists of two branches, optical remote sensing image branch and elevation feature branch. The optical remote sensing image branch uses the spatial relationship module to learn and infer the global relationship between any two spatial positions or feature maps and then extracts the multi-level features of the image by capturing more context information and pyramid attention mechanism. The elevation feature branch strengthens the classification. Based on the boundary information, the remote sensing image segmentation is realized. Experiment results on ISPRS Vaihingen image dataset demonstrate the effectiveness of the proposed method.