Research on Target Detection Algorithm Based on Aerial Video

Abstract

For the Yolov5 target detection algorithm, firstly, the DPFEM network is proposed to replace the original BottleneckCSP and C3 network structure for the problem of inadequate feature extraction for small targets, secondly, the MAFFEM module is proposed to alleviate the conflict of feature fusion due to the fusion conflicts brought by the different scales of the feature maps during the feature fusion, finally, the training The results show that the improved Yolov5 algorithm mAP0.5 and mAP0.5:0.95 are improved by 2.2% and 1.7%, respectively, and have certain application potential.