SAR Image Change Detection Based On Sparse Learning And Saliency Detection

Synthetic aperture radar(SAR)is one of the most important ways to observe the earth at present.SAR can work all day due to its advantages of imaging,making it become the focus of research in various countries.SAR image change detection has solved many problems in practical applications,such as natural disasters,disaster assessment,geographic information update,military monitoring scheduling application,which makes it become a research hotspot.In this thesis,three new algorithms are proposed based on the theory of SAR image change detection.A method of SAR image change based on cascade dictionary sparse learning is proposed.Firstly,the log-ratio image and the subtraction image are obtained on two SAR images at different time in the same region by log-ratio operation and subtraction operation respectively.The dictionaries are learned from the two obtained images,and then cascade the two dictionaries.Then,the sparse coefficients are obtained by using the cascade dictionary.Sparse reconstruction is used to obtain two reconstructed images.The two reconstructed images are fused to obtain a difference map.Finally,the results are obtained by clustering algorithm.SAR image change detection based on dictionary sparse learning has long been studied,however the construction of the dictionary has always been a problem,if the dictionary is too large,it's easy to bring error detection,otherwise the omission might increase greatly.So we use the cascade dictionary method for change detection.Experiments carried out on four sets of image data show that the proposed method can effectively suppress speckle noise and improve the performance of the detection.A SAR image change detection algorithm based on saliency fusion is proposed.Firstly,the image is filtered,and then the global saliency image and the local saliency image are constructed respectively.Then the global saliency image and the local saliency image are merged to obtain the difference image based on the saliency detection.Finally,result is obtained by clustering algorithm.The saliency detection can accurately locate the change area,but it is not clear enough for the change of the edge pixel information.This method can effectively judge the pixels which are difficult to distinguish in the change detection,and can get better results in the four sets of experimental data.A SAR image change detection algorithm based on local saliency and stacked sparse auto-encoder is proposed.Firstly,we use the method of local saliency to select the samples,and train the stacked sparse auto-encoder,and then all test samples are input in the stacked network to test,finally the result is obtained by clustering algorithm.The sparse auto-encoder can automatically learn the characteristics of the image,and has a better ability to express the edge information,so it has strong ability to express the complex relationship.The experimental results show that the proposed method has better performance in solving the problem of change detection,and has better control in both the error detection and miss detection.