Hyperspectral Imagery Change Detection Based On Characteristic Subspace Theory

Change detection is a hot topic of remote sensing research field.It is the basis of understanding the interaction between human and natural environment in time and accurately.The remote sensing image is the main source of data for change detection.In recent years,a new generation of satellites equipped with hyperspectral sensors is changing the viewing angle of the Earth's surface with its high spectral resolution,it provides new possibilities for surface change detection.Due to the limitation of spatial resolution,there are a lot of mixed pixels in hyperspectral images.In order to make full use of high-dimensional spectral information,.spectral unmixing is a problem that we need to study.The change detection based on hyperspectral data has become an important research direction in the field of remote sensing.Related research has made significant progress,but there are still the following deficiencies:First,the detection changes results after the classification are limited by the classification accuracy,change detection accuracy is not high.Second,the methods failed to make full use of the image spectrum information.Third,current change detection algorithm does not provide the detection result of the change type.In this thesis,main contents and innovations of the thesis are summarized as follows:(1)In this thesis,an endmembers extraction algorithm based on particle constrained spatial optimization is proposed.Particle Swarm Optimization is an optimization algorithm based on continuous space.Because the number of endmembers is small and endmembers is scattered in the hyperspectral image,the search space of the PSO is scattered.The traditional PSO algorithm exist the weaknesses of being sensitive to initial value,low convergence,easy to fall into the local optimum.To solve this problem,the thesis selects the pixels with high pure pixel index as the preselected pixels,and sorts the preselected pixels.Finally,the sorting pixels are taken as the searching space of PSO,to reduce the search space and improve the efficiency of the algorithm.Experimental results show that this algorithm has better result than other algorithm in simulating and AVIRIS images.(2)We propose a feature subspace learning algorithm suitable for hyperspectral image change detection.Firstly,the difference matrix is obtained by performing the difference calculation on the two images,and then the pixel with large spectral vector value is used as the target.The background space is constructed by the corresponding pixel and spectral feature.Secondly,Hyperspectral images were spectrum-resolved,and the thresholds were used to separate the different types of objects included in the projected images.This thesis makes full use of the fine spectral information of hyperspectral image,and uses the method of first projection reclassification to overcome the problem that the traditional classification change detection method is susceptible to the accuracy of classification and the high error rate.It provides the type of change,Effectively detect the changes of different objects in two phases.