Fast semi-supervised learning with anchor graph for large hyperspectral images

Abstract

As the labeled samples of hyperspectral image (HSI) are very scarce and labeling sample costs too much time and is expensive, semi-supervised learning (SSL) has an important application in hyperspectral image (HSI) classification. Among SSL approaches, graph-based SSL (GSSL) model has recently attracted much attention. However, most GSSL methods still can not deal with the large HSI as their high computational complexity. In this letter, we propose a novel approach, called fast semi-supervised learning with anchor graph (FSSLAG) to solve the large HSI classification problem. In the proposed FSSLAG algorithm, the anchor graph, which is parameter-free, naturally sparse and scale invariant, is first constructed. Then the label of samples can be inferred through the graph. The computational complexity of FSSLAG can be reduced to O(ndm), which is a significant improvement compared with traditional graph-based SSL methods that need O(n3), where n, d and m are the number of samples, features and anchors, respectively. Several experiments have demonstrated the effectiveness and efficiency of FSSLAG in terms of computational speed and classification accuracy.