Improved YOLOv7 UAV image small target detection algorithm

In order to improve the detection efficiency of small targets in UAV images in traffic systems, a small target detection algorithm for UAV images in traffic systems based on improved YOLOv7 is proposed. In the feature fusion network, the small target detection layer is added to extract feature information of different scales to improve the perception ability and positioning accuracy of small targets. Deep separable convolution ( DSConv ) is introduced into the feature extraction and feature fusion network of the model, and feature extraction is performed on each channel without introducing additional parameters, which reduces the computational complexity and memory occupation of the network and enhances the ability of the network to extract target features. The nearest upsampling module of the original model is replaced by the lightweight operator Carafe, which expands the receptive field of the model and improves the quality and accuracy of the feature map. In the feature extraction and feature fusion connection part of the model, the CBAM hybrid attention mechanism is introduced to dynamically adjust the channel and spatial information in the feature map, and improve the model 's ability to capture the important features of small targets in the traffic system under complex background. The experimental results show that compared with the original algorithm, the improved YOLOv7 algorithm improves the mAP by 4.5 percentage points and the recall rate by 5 %. Compared with the current mainstream algorithms, the improved algorithm can effectively improve the detection accuracy of small targets in UAV images in traffic systems, and significantly improve the false detection and missed detection of small target images in complex backgrounds.