The Research Of SAR Image Change Detection Based On Difference Image Fusion And Difference Image Denoising

Synthetic Aperture Radar(SAR),which has the special ability to penetrate clouds and to capture images with high resolution under all-weather and all-time,becomes one of the most important methods in earth observation and military detection fields.As one of the principal technology of SAR image analysis and interpretation,SAR image change detection has strong commercial and military value and has become a hot spot in the world.It is well known that the performance of SARimage change detection is mainly relied on the quality of the difference image.However,difference image is often corrupted by the inherentmultiplicative-speckle noise in SAR image which leads to the result that the change detection becomes more difficult.So In this thesis,we mainly focus on the fusion and noise reduction of the difference image.The main work is as follows:1.A SAR image change detection method based on local energy adaptive weights for difference image fusion was proposed.First,the probabilistic-patch-based algorithm was adopted for speckle noise reduction of the two multitemporal SAR images,and then the mean ratio operator and the log ratio operator were applied to generate two kinds of difference maps.Second,the two kinds of difference maps were combined with the local energy adaptive weights which were generated from the mean ratio difference image to form the fused image.The local energy adaptive weights which help to makes use of the advantages of two kinds of difference maps can be adaptively adjusted according to different characteristics of difference map.Finally,the k-means clustering algorithm with k=2 was used to cluster the fused image into changed area and unchanged area.Experimental results obtained on three real SAR image data sets confirmedthe advantages ofproposed algorithm which achieved good experimental results both in robustness and detection accuracy.2.A method of SAR image change detection based on adaptive weights for difference image fusion and adaptive threshold for noise reductionin high frequency sub-bands is proposed.First,a neighborhood logarithmic ratio difference map and a mean ratio difference map are generated with two input images,and then each difference map was decomposed into 1 low-frequency sub-band and 3 high-frequency sub-bands with wavelet transform.Second,an adaptive fusion weights were adopted to fuse the low-frequency sub-bands,and the principle of minimum local energy was adopted to fuse the high-frequency sub-bands.In order to improve the quality of the fused difference map,an adaptive threshold denoising method was adopted for speckle noise reduction on fused high-frequency sub-bands,and then the fused difference map was generated by wavelet inverse transform.Finally,the k-means clustering was used to cluster the fused image into two clusters to obtain the final change detection result.This method could not only effectively remove the multiplicative-speckle noise but also keep the details of the image.So the method is robust to multiplicative speckle noise and has high detection accuracy.The effectiveness of the method was proved by the experimental results.3.A method based on difference image fusion and noise reduction using nonsubsampled contourlet transform(NSCT)combined with salient information was proposed.First,mean ratio,logarithmic ratio and neighborhood log ratio difference images were constructed with two input imagesrespectively,andthe saliency image was extracted by log ratio difference image.Second,the mean ratio and neighborhood log ratio difference images were decomposed by NSCT method which produces low-pass sub-band and directional sub-bands.In order to highlight the change region of the fused difference image,the low-pass sub-band of the neighborhood log ratio difference image was restricted by the saliency image.At the same time,the directional sub-bands were selectively denoised by saliency image in different scales,and then the principle of minimum local energy which aims at suppressing the background area of the fused difference image was used to fuse the directional sub-bands.Finally,NSCT inverse transform was used to obtain the fused difference image,and the change detection map was generated by k-means clustering method.Experimental results show that the method can get better edge and detail information as well as higher detection accuracy.