Research On Multi-feature Synthesis High-resolution Remote Sensing Image Change Detection Based On Object-oriented

With constant development of science and information technology,the natural surface landscape and use forms,which are closely related to human society,are changing continuously,Detecting those changes timely can help people to realize the relationship between human and earth correctly,utilize and manage natural resource efficiently.So change detection plays an indispensable role in the sustainable development of society and economy.Due to the large-scale and large-area features,remote sensing skill has become a hot research method on change detection.Comparing with the traditional low-resolution remote sensing images change detection,high-resolution images specify the detail of objects on land surface,thereby increasing their capacity on dividing.But the rising of images' resolution also brings some problems to change detection,for example,how to quickly extract the change information and satisfy the degree of automation and the need for precision is still unsolvable.In this paper,we do some research on change detection of high-resolution remote sensing images.In the method,object-oriented multi-scale segmentation has been used to obtain the object in high-resolution images at first,then we extract features from the vector images of the image objects,taking into account their spectrum,shape,texture features,and the neighborhood features,which is taken as the focus of research.An object-oriented multi-feature synthesis for high-resolution remote sensing image change detection is also proposed.Firstly,the change detection experiment based on spatial neighborhood feature is carried out by QuickBird image.The presented neighborhood feature method is compared with both of the traditional difference method and the classification method.the results that our method has higher detection accuracy.Moreover,through the multi-feature importance analysis,many features,such as the spectral mean,standard deviation,area,brightness value,length,width,maximum difference,and neighborhood characteristics of the object spatial position are selected to construct an one-dimensional feature vector,the correlation coefficient of the general eigenvector of the image object before and after the phase is calculated,and the corresponding threshold is used to determine whether the corresponding pattern has changed.In the above automatic change detection,to ensure a relatively low leakage rate,the threshold is set to be more stringent,resulting in a higher false detection rate,so we need to remove pseudo-change information.First of all,unsupervised clustering method,ISODATA,is used to cope with the result of automatic change detection.After the clustering of the changes in the detection results,we will produce n similar characteristics of the characteristics of the category,which containing a lot of pseudo-change information,but the distribution rules of these pseudo-information can be found out.With the aid of human interpretation and judgment,the computer is able to find the pseudo-change information to match the results of the true clustering.After a certain number of verifications,it can be determined that the nature of a particular category or a certain category of the n categories is consistent with the characteristics shown by the pseudo-changing region,then in this case we can proceed with this category,and then quickly extract these pseudo-change information.Compared to the artificial one by one check the spot,this approach is batch procedure,its efficiency has been greatly improved and the accuracy has also been guaranteed.In order to validate the presented method,the change detection experiment was selected as the research area in Lianyuan City,Hunan Province.The images are gotten in the summer of 2015 and 2016 respectively.The spatial resolution is 1m and 2m.There is no cloud interference,the contours of the features are clear,the features are obvious and the imaging conditions are good.The automatic detection accuracy is 94.55%,and the Kappa coefficient is 59.26%,the missed rate is 3.57%,and the precision on pseudo-change is 54.55%.After the semi-automatic human-computer interaction,the overall detection accuracy is 98.76%,which is 4.21% higher than the previous 94.55%;Kappa coefficient is 86.34%,which is 27.08% higher than the previous 59.26%;the missed rate is 9.29% Compared with 3.57% before,it rises 5.72%;false rate is 16.45%,compared to the previous 54.55% down 38.10%.The comprehensive evaluation of the test results has been significantly improved.It concludes that this method has a certain significant value in the detection of remote sensing image changes in large scale range.It may play an important part on providing some theoretical support and decision analysis for the country to formulate and implement relevant development strategies and plans,optimizing the pattern of land and space development and optimizing the allocation of various land resources,promoting ecological and environmental protection,and building a resource-saving and environment-friendly society.