L2,0-norm regularization based feature selection for very high resolution remote sensing images

摘要: This paper presents a l 2,0 -norm regularization based feature selection method to analyze very high resolution remote sensing imagery. The tackles the problem on 2,1 objective function and 2, 0 equality constraint. constrained optimization is solved by an efficient algorithm augmented Lagrangian figure out stable local solution. Though should handle non-convex non-smooth problem, it outperforms approximate convex counterparts state-of-art methods in light of classification accuracies 1-NN SVM classifiers. experimental results demonstrate effectiveness presented selecting features with great generalization capabilities.