Research on small target detection algorithm based on improved yolov3

As a popular target detection algorithm, yolov3 has high detection speed and detection accuracy. However, compared with other target detection algorithms, yolov3 has poor performance in small target recognition, with higher missed detection rates and higher false detection rates. In response to the above problems, this paper draws on the idea of adaptive spatial feature fusion (ASFF), and makes full use of the semantic information of high-level features and the fine-grained, edge, and texture features of low-level features to fuse high-level feature maps and shallow feature maps. Secondly, the K-means clustering algorithm is used to generate a more suitable anchor frame, which is conducive to further improving the recognition accuracy of the model, and finally the network structure is further improved. Through training and recognition in the NWPU vhr-10 dataset, the mAP of the improved yolov3 algorithm is 79.89%, and the mAP of the unimproved yolov3 algorithm is 78.26%. After the improvement, the mAP of 1.63% is increased, and the improvement can solve the problem of low target recognition rate of yolov3 to a certain extent.