Multi-view learning for hyperspectral image classification: An overview

Hyperspectral images (HSI) are obtained from hyperspectral imaging sensors to capture the object’s information in hundreds of spectral bands. However, how to make full advantage of spatial and spectral information from a large number of spectral bands to improve the performance of HSI classification remains an open question. Many HSI classification works have recently been reported by employing multi-view learning (MVL) algorithms and have achieved promising results. Generally, MVL based HSI classification can be divided into three steps, i.e. (1) multi-view construction, (2) interactivity enhanced, and (3) multi-view fusion. This paper presents a review of MVL methods in HSI classification based on the general steps of MVL. Specifically, multi-view construction builds various representations from the raw HSI data as different views to adapt to an MVL setup. Secondly, interactivity enhanced aims to interact with different view features, so that the current view contains information from other views and to achieve a pre-fusion effect. Finally, multi-view fusion uses different fusion methods to combine multiple views and classify HSIs using complementary information between the views. In addition, we analyzed and discussed separately representative approaches in each step and their characteristics, and introduced some of the most advanced work. Overall, this survey aims to provide an insightful overview of developments in MVL in HSI classification and help researchers identify its future trends.