Remote Sensing Image Object Detection Method Based On Improved YOLOv3

Abstract In order to solve the problem that irregular targets and dense are difficult be detected in optical remote sensing images, this paper improved YOLOV3 model Firstly, further combine feature information of different scales, PaNet structure is introduced into FPN part original YOLOv3, obtained effective layer continued extracted for a round feature. The not only up-sampled achieve fusion, but also down-sampled again enhanced fusion SimOTA method dynamically match positive samples set sample numbers targets, which improves speed algorithm, reduces extra hyperparameters Experimental verification using richer DOIR data sets shows detection ability algorithm significantly improved. Compared with its mAP by 15.1 points, among accuracy small most.