Feature Line Embedding Based on Support Vector Machine for Hyperspectral Image Classification

In this paper, a novel feature line embedding (FLE) algorithm based on support vector machine (SVM), referred to as SVMFLE, is proposed for dimension reduction (DR) and improving the performance of generative adversarial network (GAN) in hyperspectral image (HSI) classification. The GAN has successfully shown high discriminative capability many applications. However, owing traditional linear-based principal component analysis (PCA) pre-processing step cannot effectively obtain nonlinear information; overcome problem, (SVMFLE) was proposed. SVMFLE DR scheme implemented through two stages. first scatter matrix calculation stage, FLE within-class matrix, between-scatter vector-based between-class are obtained. Then second weight determination training sample dispersion indices versus SVM-based calculated determine best matrices final transformation matrix. Since reduced space obtained by much more representative than that using conventional schemes, HSI classification higher. effectiveness with or nearest neighbor (NN) classifiers evaluated comparing them state-of-the-art methods three benchmark datasets. According experimental results, NN higher schemes indices. Accuracies 96.3%, 89.2%, 87.0% were Salinas, Pavia University, Indian Pines Site datasets, respectively. Similarly, classifier also achieves 89.8%, 86.0%, 76.2% accuracy rates these