Research on Object Detection Technique in High Resolution Remote Sensing Images Based on U-Net

Scene segmentation is an important method to implement fine-grained object detection in remote sensing image. The method based on fully convolution neural network is the hotspot in the field. In this paper, a hilly convolution neural network model based on U-Net model which achieves state of art result on biomedical segmentation application is proposed. Experiments on DSTL dataset show that the proposed model produces accurate pixel-wise classifications for remote sensing image in an end-to-end way.