Small Target Detection for UAV Aerial Images Based on ECW-YOLO

The feature information of small targets is not obvious in the aerial images of unmanned aerial vehicles (UAVs), which leads to insufficient feature information and low accuracy for small target detection. To address the problem, ECW-YOLO algorithm is proposed. First, C3K2-EMBC module is designed to extract efficient feature information. Secondly, a complex multidirectional and multi-scale feature pyramid network (CMMPN) is given to achieve interactive fusion of multi-scale features. Finally, the WIOUv3 loss function is used to avoid the negative impact of low-quality samples during training process and improve the localization performance of the target detection algorithm. Experimental results on the VisDrone2019 dataset show that the algorithm has excellent accuracy and robustness in small target detection of UAV aerial images.