Drone Detection Method Based on MobileViT and CA-PANet

Aiming at the problems of large amount model parameters and false missing detections multi-scale drone targets, we present a novel detection method, YOLOv4-MCA, based on lightweight MobileViT Coordinate Attention. The proposed approach is improved according to framework YOLOv4. Firstly, use an as feature extraction backbone network, which can fully extract local global representations object reduce model’s complexity. Secondly, adopt Attention improve PANet obtain attention called CA-PANet, more positional information promote fusion with low- high-dimensional features. Thirdly, utilize K-means++ method optimize anchor box efficiency. At last, construct dataset conduct performance experiment Mosaic data augmentation method. experimental results show that mAP reaches 92.81%, FPS 40 f/s, number only 13.47 M, better than mainstream algorithms achieves high accuracy for targets using low parameters.