基于双重感知注意力与特征度量协同的小样本遥感图像目标检测方法OA

Remote sensing image object detection has broad application prospects in many fields.However,under the background of few-shot,it faces challenges such as insufficient feature extraction,poor positioning accuracy,and prone to classification errors.To address these issues,firstly,a dualperceptual attention module is constructed.Its background attenuation attention effectively suppresses background interference,and the spatial perceptual attention guides the network to focus on the key information for target positioning,helping the Region Proposal Network(RPN)generate better region proposal boxes,reducing the probability of target omission,and improving the performance of few-shot target positioning.At the same time,introspective metric learning is introduced,it prompts the model to deeply explore the similarities and differences between samples during training,strengthening feature learning and understanding,and improving classification accuracy.Finally,a few-shot object detection model based on fine-tuning transfer learning is designed and verified on the remote sensing dataset NWPU VHR-10.Experimental results show that compared with the benchmark algorithm,the proposed algorithm achieves significant improvements in average precision.