A Novel Hyperspectral Unmixing Method based on Least Squares Twin Support Vector Machines

In hyperspectral images, endmembers characterizing one class of ground object may vary due to illumination, weathering, slight variations the materials. This phenomenon is called intra-class endmember variability which important factors affecting performance unmixing. However, often ignored in unmixing, causes a decrease accuracy How deal with focus. To address this problem, we propose novel unmixing method based on Least Squares Twin Support Vector Machines (ULSTWSVM). ULSTWSVM uses multiple training samples (endmembers) model pure class, takes into account At same time, obtains abundances by calculating distances from mixed pixels classification hyperplanes, simple and efficient. mainly comprises three steps: (1) obtain two non-parallel hyperplanes solving quadratic programming problems (QPPs) least squares sense, (2) calculate (3) normalize convert them abundances. Experimental results both synthetic real data show that proposed outperforms methods used for comparison.