Vehicle Target Detection in Remote Sensing Images Based on Improved YOLOv5s
Remote sensing images are widely used in the field of target detection.In the vehicle target detection,due to the complex back-ground of remote sensing images themselves and the small and dense vehicle target,the existing target detection methods often face the problems of poor detection effect,missing detection and misdetection.In view of this,we propose an improved YOLOv5s target detection method.A self-attention mechanism with position information encoding is introduced to strengthen the fusion of feature information within the same scale.MobileViT is employed at the higher level of the network to handle more abstract features and capture complex contextual information efficiently,which further enhances the recognition ability of small targets and solves the problem of small target miss-detection.A lightweight up-sampling operator is applied to reduce the model complexity.In addition,the ShapeIoU loss function is applied to make the model pay more attention to the shape and scale of the boundary frame in order to improve the detection accuracy.A complete experiment is conducted on the remote sensing vehicle data set-COWC.It is showed that all indicators in the model have been improved,with the accuracy increased by 0.3 percentage points,the recall increased by 1 percentage point,and the mAP increased by 0.8 percentage points.
remote sensing imagesvehicle target detectionYOLOv5sself-attention mechanismlightweight upsampling operatorloss function
万方数据
维普
