The Applied Research Of SAR Image Change Detection

The synthetic aperture radar(SAR)sensor is widely used to detect image changes due to the characteristic of operating in day and night and all-weather conditions.Change detection which distinguishes the changed regions in the same location at different periods,is mainly applied to environmental monitoring and disaster assessment,etc.The detection results will be very big deviation if we use the method of processing optical image directly due to the influence of various factors.It is a hot issue for many scholars to solve the influence of speckle in recent years.There are many valid methods to improve the accuracy of SAR image change detection,but the applicability is terrible and when we process different SAR image pairs with different models,the results are differ greatly.In particular,a single threshold cannot achieve optimal results,when the data sets have multiple target regions.Therefore,we consider the practical application demand in this article,carried out a series of experimental studies with different high resolution SAR image data sets to reduce the influence of speckle.First of all,the logarithmic function is commonly applied to transform multiplicative speckle noise into additive noise logarithmic ratio,but the local information is neglected.So Weighted logarithmic ratio method is used to improve accuracy and the arithmetic that is similar to the kernel function is used to enhance the difference.The experimental results show that the weighted logarithm ratio method is better than the classical mean logarithm ratio.The method based on logarithmic ratio can detect the valid results in most cases,but the main changed area will miss when we test with another data set,which proves the limitations of logarithmic ratio.According to the characteristic of logarithmic curve,the values in lower pixel level are enhanced,but also weakening the values in the high pixel level.For most data sets,the changed areas occur in the water,vegetation areas and so on which shows dark on the vision.When the changed areas occur on bright region such as buildings,we will lose changed area.In fact,methods based on logarithmic function are only aiming to the changed area at the appropriate low-pixel level and the value range at this low pixel level is not clear,so additional method is required.As a common and valid method to find changed area,the difference method has a higher false alarm ratio.In this paper,we choose the saliency extraction to improve method.The saliency extraction is original from optical image processing method,which is mainly used for obtaining visually evident area in the eyes.In most cases,it will enhance the high pixel level and weaken the low pixel level.Therefore,we propose the method that two detection routes with weighted logarithm ratio and saliency extraction method is carrying at the same time,and finally the detection results are effectively superimposed to obtain all changed areas.In order to further effectively detect the changed area,we introduce shearlet denoising and shearlet fusion by considering the sparse characteristics and multi-scale features of shearlet.It is necessary to use logarithmic operation because that shearlet denoising processes mainly additive noise.After denoising process,one route continues to use the weighted logarithm ratio method and the other route performs exponent operation before the use of saliency extraction.The difference images obtained from two routes are combined with shearlet fusion to obtain the actual change area to reduce the false alarm rate and reduce the disturbance.Image fusion combines the advantages of different methods in the extent,but the final detection accuracy can be affected when there is lots of noise original from the logarithmic difference map.The weighted threshold segmentation is used to improve method combined with weighted logarithmic ratio.To verify the validity of this method,this paper uses multiple actual SAR image data sets to verify the validity of the proposed methods.According to optical reference image and field validation of the statistics obtained respectively,we use detection rate(DR),false alarm rate(FAR)and overall accuracy(OA)to perform evaluation.The final experiment results are compared with the current mainstream algorithms to prove each part of the proposed improvements effectively and overall applicability is stronger.