基于改进YOLOv7的遥感图像目标检测算法OA

In response to the challenges posed by complex backgrounds,significant differences in target scales,and limited feature infor-mation for small targets in optical remote sensing images,an improved remote sensing target detection model based on YOLOv7 is pro-posed.The main improvements include the introduction of efficient multi-scale attention(EMA)in the backbone network to preserve multi-scale information in the feature map.The SPPCSPC module incorporates SPPF pooling structure and a large selective kernel at-tention mechanism(LSK).Additionally,the maximum pooling layer in the MP module is replaced by the space-to-depth layer(SPD)to reduce the loss of fine-grained feature information,thereby enhancing the capability to identify and locate small targets in complex back-grounds.The chosen loss function is the MPDIoU loss function,providing more accurate regression results and improving the model's convergence speed.Experimental results demonstrate that the proposed algorithm achieves an average precision mean average precision(mAP)of 95%on the NWPU VHR-10 dataset,representing a 3.2%improvement compared to the original model and effectively enhan-cing the detection accuracy of remote sensing targets.