Improved target detection algorithm for remote sensing images with YOLOV7

On the task of remote sensing image target detection, aerial images have detection difficulties in terms of small data volume, small imaging, and few pixel features, while traditional target detection algorithms have poor real-time and accuracy, and high false and missed detection rates for small targets. Targeting the aforementioned issues, a remote sensing target detection algorithm based on the improved YOLOV7 is proposed. First, on the basis of the traditional YOLOV7 algorithm, the BiFormer attention mechanism is added to the YOLOV7 backbone network, and BiFormer proposes a new dynamic sparse attention method through two-layer routing to achieve more flexible computational allocation with content-awareness, which improves the algorithm's feature extraction capability; second, Wise IoU is adopted as the bounding frame regression loss function instead of the original CIoU, which improves the convergence speed of the algorithm. The experimental results on the DOTA dataset show that the accuracy of the improved YOLOV7 algorithm reaches 76.8%, and the mAP is improved by 1.7%. To a certain extent, the detection accuracy of YOLOV7 algorithm on remote sensing images is improved.