U-SHAPE FUSION FEATURE BASED OBJECT DETECTION METHOD FOR REMOTE SENSING IMAGES

Due to the particularity of remote sensing image, such as wide field of vision, small target, how to quickly JP3and accurately detect targets in remote sensing images is still a challenging problem. A new method based on improved YOLOv3, U-YOLO, is presented. The selection method of anchor box was improved, and the problem of unbalanced selection of pre-selection box was solved. A U-shaped feature extraction module was proposed to extract deeper features and improve the detection effect. JPThe area factor applied to the loss function was put forward, which improved the difficulty of small target detection. The experiments were conducted on the NWPU VHR-10 dataset and RSOD dataset. Experimental results show that this method is 0.079 and 0.065 higher than the original YOLOv3 in the two groups of experiments, respectively.