一种改进的YOLOv5s航拍车辆检测算法OA北大核心CSTPCD

To solve the problem of small vehicle target detection in aerial images,a vehicle detection algorithm based on improved YOLOv5s from the aerial perspective is proposed.The unused shallow feature information is further fused with other deep feature information to compose a new detection layer for small target detection to enhance the detection capability of small targets.The CSP module is combined with the space-to-depth(SPD)module to form the SPD-CSP module,which replaces the downsampling operation of the original network and reduces the loss of practical information of small targets during feature extraction.Furthermore,the efficient channel attention(ECA)module,a channel attention mechanism,is introduced into the Backbone part.To do so,the network will pay more attention to the vital information in the feature graph and reduce the interference of irrelevant information by adaptively adjusting the weight coefficients of different feature channels.The experimental results show that the proposed algorithm improves the mean average precision PmAP0.5by 6.4%on the VisDrone dataset compared to the YOLOv5s network,and the detection speed FPS reaches 65 frames per second,which enables real-time and accurate detection of aerial vehicles.