Adaptive Land Cover Change Detection Based On Multiple Features And Class Probability Fusion

Land cover change information plays an increasingly important role in global sustainable development,climate change research,ecological and environmental assessment,disaster risk monitoring and territorial spatial planning,etc.Accurate and efficient acquisition of land cover change information has become a hot issue to be solved urgently.With the development of aerial remote sensing,space remote sensing and other earth observation technologies,change detection methods based on remote sensing images have become an effective means to obtain land cover change information.In recent years,domestic and foreign researchers have proposed a large number of remote sensing image change detection methods.However,the current methods still have problems such as insufficient utilization of remote sensing image feature information,low adaptive degree of threshold selection methods and change type information being affected by error accumulation,resulting in the change detection results not accurately reflecting the dynamic changes of land cover.Therefore,this thesis presents an in-depth analysis of the practical problems faced by land cover change detection,and proposes a multi-feature fusion change magnitude map calculation method based on spectral enhancement,an adaptive change threshold selection method based on land cover class probability and spatial neighborhood information,and a change type determination method based on the knowledge base of spectral changes in typical land cover types.The main research contents are as follows.(1)A multi-feature fusion change magnitude map calculation method based on spectral enhancementChange magnitude calculation is the process of directly comparing the spectral,texture and shape features of remote sensing images from different time periods to obtain change magnitude maps of remote sensing images.This thesis presents an in-depth survey of the current research status and development trend of change magnitude map calculation.Aiming at the problem of insufficient utilization on remote sensing image feature information,a multi-feature fusion change magnitude map calculation method based on spectral enhancement is proposed.The innovation of the method lies in the introduction of spectral enhancement features into the multi-feature fused change magnitude maps.The change magnitude map based on spectral enhancement features is calculated by integrating spectral value features and spectral shape features through wavelet fusion.Then,the texture features and shape features of remote sensing images are considered to complement the spectral features and improve the distinguishability of the changed and unchanged regions.The integrated change magnitude map is obtained by calculating adaptive weights through information entropy,thus improving the accuracy of change detection results.The experimental results show that the detection results of the proposed method are better than those of the conventional method.(2)An adaptive change threshold selection method based on land cover class probability and spatial neighborhood informationChange region generation is a process of dividing change magnitude maps into change regions and unchanged regions by selecting suitable thresholds for change magnitude maps using change threshold selection methods.This thesis presents a systematic analysis of the current research status and development trend of change threshold selection.Aiming at the problem of low adaptive degree on threshold selection method,an adaptive change threshold selection method based on land cover class probability and spatial neighborhood information is proposed.The innovation of the method lies in the integration of the land cover type and spatial neighborhood information into the adaptive change threshold selection.The Bayesian criterion is used to calculate the class probabilities of the change magnitude in each land cover type to integrate the information of land cover types.Then,the spatial neighborhood information of class probabilities is obtained according to the bilateral filtering method to construct a spatial surface based on the land cover type and spatial neighborhood information.The final threshold value is determined by the degree of difference between the spatial surface and the change magnitude map,thus minimizing the possibility of missed and false alarms.The experimental results show that the accuracy of the detection results in the proposed method is higher than the accuracy of the detection results in the traditional method.(3)A change type determination method based on the knowledge base of spectral changes in typical land cover typesChange type determination is the process of obtaining further information on land cover change types "from to" based on the generation of change regions.This thesis presents a corresponding study on the current status and development trend of change type determination.Aiming at the problem of change type information being affected by error accumulation,a change type determination method based on the knowledge base of spectral changes in typical land cover types is proposed.The innovation of the method lies in the consideration of the knowledge base of spectral change information in typical land cover type changes into the change type identification.The knowledge base of spectral change information in typical land cover type changes is constructed from the perspectives of spectral mean change,spectral angle change and thematic index change.Then,knowledge matching is carried out based on the previous moment's remote sensing image classification data and the knowledge base with the spectral change information of remote sensing images in the change regions.The minimum value of the matching result is selected as the judgement criterion to obtain the final land cover change type of the change regions,thus reducing the impact of error accumulation on the change type determination.The experimental results show that the proposed method can not only identify change type information,but also reduce false identification.