Change Detection In High-resolution Remote Sensing Images By Integrating Multiple Feature

The spectral characteristics of ground objects can reflect the rich information on the categories and attributes of objects,and is widely used to detect land cover change.However,with the availability of increasingly high-resolution(HR)satellite images,a large amount of salt and pepper noise and the low accuracy of the detection of artificial objects frequently appear in methods based on a single spectral difference.In contrast,spatial features such as texture and structure are more stable and not affected by spectral differences,helping to identify artificial objects such as buildings and roads.Therefore,contextual semantic features,spatial features and other change information of high-resolution remote sensing images are also applied to change detection methods.Inspired by such research,in this thesis we propose the change detection algorithm to make a comprehensive application of various forms of information,to convert from a single detection method to a multi-method fusion,and to convert from the pixel level to the object level.In order to overcome the limitations of a single method and improve the applicability and effectiveness of the change detection algorithm in high-resolution remote sensing images,this dissertation takes three high-resolution remote sensing images with different spatial resolutions,different sensors,and different ground objects as data sources,and proposes the change detection algorithm integrating multiple information at the pixel level and the object level respectively.The details are as follows:(1)In view of the problem that most existing spectral feature extraction methods are based on a single operator and cannot obtain comprehensive spectral attribute change information,this thesis proposes a new spectral change information extraction method that combines three methods.By integrating the initial results of three spectral change detection methods(iteratively reweighted multivariate alteration detection,iterative slow feature analysis,and principal component analysis)to obtain accurate and comprehensive spectral change information after majority voting decision analysis.Taking the second dataset as an example,compared with traditional methods,the overall accuracy(OA)is increased by 0.517%～11.290%,the false alarm rate(FAR)is decreased by 0.003～0.123,the missed detection rate(MR)is decreased by0.011～0.126,and the kappa coefficient and F1 scores are increased by 0.022～0.296.The idea of integrating multiple methods can overcome the limitation of a single operator and the influence of false alarms,such as salt and pepper noise,as well as obtain the optimal spectral difference information.(2)At the pixel level,a single feature is difficult to reflect the comprehensive change information,so it cannot be effectively used for change detection.Therefore,we propose a change detection framework in which the spectral features,texture features and structural features are integrated to get the change information of objects.The framework has three components: Firstly,the spectral features are extracted by using the above spectral feature extraction method;Secondly,the texture feature set and structure feature set are respectively obtained by multi-scale grey level co-occurrence matrix and histogram of oriented gradient,which are used to obtain the change information of target objects in terms of texture feature and geometric feature;Finally,the spectral,texture,and structural features are analyzed and logical fused to get the initial results.Meanwhile,the mathematical morphology operation is performed as a post-processing on the initial results to get the final comprehensive change pattern.To evaluate the performance of the proposed method,three sets of high-resolution remote sensing images are selected to carry out change detection experiments,and two experiments are compared with other methods.Compared with the spectral change results,the FAR and MR decreased by 0.003～0.144,the OA increased by 0.528～3.039 %,and the kappa coefficient and F1 scores increased by 0.053～0.112.The results show that the proposed method can integrate the advantages of multiple features and improve the applicability and robustness of the change detection.(3)At the object level,detection noise appearing in the pixel-level results due to the abundant spatial details in the high-resolution images makes it difficult to acquire an accurate interpretation result.In this thesis,an object-oriented change detection approach is proposed which integrates spectral–spatial–saliency change information and fuzzy integral decision fusion for high-resolution remote sensing images with the purpose of eliminating the impact of detection noise.First,to reduce the influence of feature uncertainty,spectral feature change is generated by using the above method.The saliency change map of bi-temporal images is obtained with the co-saliency detection method to complement the insufficiency of image features.Secondly,six multi-scale grey-level co-occurrence matrix statistics and another spatial feature of the rolling guide filter are added,so the spatial change information is obtained by spatial feature set construction and the optimal feature selection strategy.Then,the first principal component image of the superimposed image is segmented by the fractal evolution net approach,and the optimal segmentation scale is calculated by the optimal scale determination strategy to obtain the image objects.Finally,different pixel-level image change information and the segmentation result are fused using the fuzzy integral decision theory to determine the object change probability.Three high-resolution remote sensing image datasets are selected to carry out change detection experiments,and three groups of comparative experimental analysis and four aspects of experimental results are discussed.The OA of the proposed algorithm is greater than 95%,the FAR is lower than 0.016,and the kappa coefficient and F1 scores are higher than 0.780.The results show that the proposed algorithm can overcome the influence of salt and pepper noise and obtain complete change objects with clear boundary.Spectral–spatial–saliency change information was found to play a major role in the change detection of high-resolution remote sensing images,and the fuzzy integral decision strategy was found to effectively obtain reliable changed objects to improve the accuracy and robustness of change detection.