Research Of Hyperspectral Target Detection Using Unmixing Technology

Hyperspectral pixel unmixing and hyperspectral target detection are two important branches of hyperspectral remote sensing technology.Hyperspectral image not only contains abundant spectral information,but also contains spatial information.The endmember set and the abundance information of each endmember can be obtained by the hyperspectral image unmixing algorithm.By using the information obtained from the pixel unmixing,a target detection algorithm based on pixel unmixing is proposed in this thesis.Meanwhile spatial information is fused into target detection algorithm.Main tasks include:(1)This thesis summarizes the research status of the spectral imaging technology and target detection,and analyzes a variety of classical algorithms of the hyperspectral target detection and pixel unmixing.(2)A hyperspectral target detection algorithm based on pixel unmixing is proposed.Traditional hyperspectral target detection algorithm focuses on the information of spectral.It is easy to make the detection results affected by the noise which only uses spectral as a prior information.More prior information should be added to the target detection.The endmember set and the abundance information of each endmember can be obtained by the hyperspectral image unmixing algorithm.By using the information obtained from the pixel unmixing,a target detection algorithm based on pixel unmixing is proposed and the improved algorithm is named new nonnegative constrained least squares(N-NCLS).The N-NCLS takes advantage of the N-FINDR to obtain the endmember set and uses spectral angle mapping(SAM)to get the target endmember and uses NCLS to obtain the abundance of the target endmember.Besides,the N-NCLS uses particle swarm optimization(PSO)to optimize the threshold of abundance images.(3)A hyperspectral target detection algorithm based on 4-TV(Total Variation)is proposed.The spatial information of hyperspectral images is fused into target detection algorithm.The target detection is transformed into a convex optimization problem.In this thesis,a detailed derivation is made for the solution of this convex optimization.Experimental results show the effectiveness of the improved algorithm.