Small target detection based on image enhancement in remote sensing images

Abstract

At present, the target detection algorithms represented by deep learning, such as: You only look once (YOLO), Single Shot MultiBox Detector (SSD) and other single-stage target detection algorithms have demonstrated high detection rate and stable detection effect However, when this type of detection algorithm detects certain types of targets in remote sensing images, because the relative and absolute sizes of these targets are very small, the problems of low detection rate and high false detection rate will occur. Traditional detection methods cannot meet the requirements of remote sensing. Image inspection needs. To solve this problem, this paper proposes a small target detection method based on image enhancement method. Firstly, the image overlap and block method are used to increase the relative size of the small target and retain the information of the small target. Secondly, the image after each block is The block uses an image enhancement method based on SRGAN to improve the resolution of the small target, which is used to improve the sub-table rate of the small target and enhance its feature and context information. Then based on the YOLOv4 method, the enhanced image block is detected. Finally, based on the NWPU VHR-10 data set, the effectiveness of the method is verified in the experimental results of remote sensing image based on image enhancement of small target detection.