Change Detection For High-resolution SAR Images Based On NSCT SPP Net

Synthetic Aperture Radar(SAR)which has all-time,all-weather detection and reconnaissance tracking capabilities,and is widely used in defense construction and national economy.Currently,as an important part of SAR image processing,SAR image change detection technology plays an important role in environmental detection,urban construction,and disaster assessment.This paper combines deep learning theory and proposes three innovative methods in the framework of SAR image change detection method based on the region of interest proposed by the team:1.A method for detecting SAR image change based on Spatial Pyramid Pooling Network(SPP Net)is proposed.The method uses SPP Net to achieve region of interest of twotemporal SAR images.Since SPP Net is a supervised deep learning network,the initial change detection result is obtained by combining DBN and FCM,and the initial change detection result is used as a pseudo-labels of SPP Net.The acquired region of interest is subjected to FCM clustering algorithm to complete the determination of change and unchanged classes,to achieve the final change detection.The SPP Net-based region of interest detection method can detect the region of interest in two-temporal SAR images well,and a good detection result of the region of interest can improve the final change detection effect.The algorithm performs experiments on five sets of data sets and has achieved good results.2.A non-subsampled contourlet transform(NSCT)and global-local SPP Net based SAR image change detection method is proposed..NSCT and global-local SPP Net consists of NSCT global SPP Net and NSCT local SPP Net.The NSCT global SPP Net achieves the acquisition of the region of interest,and the NSCT local SPP Net detects the change of the region after the region of interest is acquired.This method effectively preserves the edge contour information of the image,and has better detection ability for the changed regions in the high resolution SAR image.Experimental results show that the algorithm has achieved good results on five sets of data sets.3.A detection method for SAR image change based on multi-layer feature pyramid pooling network is proposed.The multi-layer feature pyramid pooling network not only includes the bottom-up connection of the convolutional operation in the traditional convolutional network,but also joins the horizontal connection after the feature layers of each level are pooled after the SPP layer.The multi-layer feature pyramid pooling network overcomes the fact that the network structure in the previous chapter only used the last deep features,did not combine the deep features with the shallow features,and realized the syndication of semantic information at each layer to improve the detection accuracy.