YOLOv3_ReSAM: A Small-Target Detection Method

Small targets in long-distance aerial photography have the problems of small size and blurry appearance, traditional object detection algorithms face great challenges field small-object detection. With collection massive data information age, been gradually replaced by deep learning an advantage. In this paper, YOLOV3-Tiny backbone network is augmented using pyramid structure image features to achieve multi-level feature fusion prediction. order eliminate loss spatial hierarchical caused pooling operations convolution processes multi-scale multi-layer structures, a attention mechanism based on residual proposed. At same time, idea reinforcement introduced guide bounding box regression basis rough positioning native boundary strategy, variable IoU calculation method used as evaluation index reward function, model proposed for fine adjustment. The VisDrone2019 set was selected experimental support. Experimental results show that mAP value improved 33.15%, which 11.07% higher than model, accuracy 23.74%.