Remote Sensing Image Detection Based on the Improved YOLOv5
Aiming at the problems of background complexity and scale change in existing remote sensing image tar-get detection, an improved remote sensing image target detection algorithm based on YOLOv5 model was proposed. Firstly, the Mosaic image is used to enhance the reconstructed dataset to improve the training effect and robustness of the model. Secondly, the SE attention mechanism is added to the backbone network of YOLOv5s, so that the im-proved model can capture the target feature information more accurately. Finally, BiFPN is used to replace the FPN+PAN structure in the original model, so that the model can carry out feature fusion at different scales, and reduce the shallow information loss in the detection process. The experimental results show that compared with the original model, the average precision, accuracy rate and recalling rate of the improved model are promoted and the im-proved model has stronger feature extraction ability and higher detection efficiency for remote sensing image target detection, which verifies the effectiveness of the improved method.
万方数据
维普
