Video SAR Moving Target Detection Algorithm Based on Improved Faster R-CNN
In view of the scarcity of video SAR data for deep learning and the problems of missing de-tection and false alarm existing in moving target detection algorithm,based on the deep learning dataset produced by the real video SAR data from Sandia National Laboratory,and an improved Faster R-CNN video SAR moving target detection algorithm is proposed.The algorithm takes the intercepted ResNet50 as the feature extraction network,uses K-means and genetic algorithm to adaptively calculate the size of anchor box,and adds the S-curve enhancement method to enhance the image contrast information in the data preprocessing.The experiment verifies that the proposed method can significantly improve the moving target detection rate and speed,with AP and F1 score improved by more than 10 points,which effectively reduces the false alarm and missed detection,and the overall performance is better than that of the one-stage algorithms SSD and RetinaNet.
video SARmoving target detectionFaster R-CNNimage enhancementK-meansge-netic algorithm
万方数据
维普
