Multi scale PCA-HOG optimized attention mechanism for optical and SAR image matching
For the problem of low image matching accuracy caused by nonlinear radiation differences and strong noise interference in optical and SAR images,this paper proposes a remote sensing image matching method that combines PCA enhanced multi-scale HOG features with attention mechanism.In order to overcome the nonlinear grayscale distortion between images,directional gradient histograms are used to construct image structural features.To avoid strong noise interference in the image,principal component analysis algorithm is used to enhance the local gradient direction of the multi-scale directional gradient histogram,ensuring accurate extraction of image structural features under noise interference Besides,we aggregated the spatial position and contextual information of feature points through an attention mechanism graph neural network,then solved the optimal allocation matrix of the aggregated feature vectors,and determined matching points based on thresholds.Experiments were conducted using four sets of optical and SAR images,and the results showed that this method achieved the highest number of correct matching points,the highest matching accuracy,the smallest root mean square error of matching points,and reduced matching time.
万方数据
维普
