Target Extraction from Remote Sensing Image Based on Scene Recognition and Target Segmentation

Extracting dense and different sizes targets from large-scale remote sensing images is a challenging task. This paper proposes a remote sensing image target extraction method based on scene recognition and target segmentation. The method recognizes images having targets first and then extracts targets via segmentation, both implements using deep network models. Firstly, cropping large-scale remote sensing images into smaller images, and classifying scenarios by whether they contain targets or not. Next, a full-resolution neural network target segmentation model with multi-source input is constructed. In the segmentation model, feature resolution retaining, and feature fusion together with data exchange mechanism lead to better feature extraction for different sizes targets and overcome the problem of gradient vanishing. Experiments for building extraction on two remote sensing data sets show that the proposed method obtains better results than the comparable deep neural network models in accuracy, and does better in targets integrity and edges smoothness.