Document worth reading: “Deep Learning for Hyperspectral Image Classification: An Overview”

Hyperspectral image (HSI) classification has change right into a scorching matter inside the self-discipline of distant sensing. In regular, the superior traits of hyperspectral data make the proper classification of such data troublesome for typical machine learning methods. In addition, hyperspectral imaging usually affords with an inherently nonlinear relation between the captured spectral knowledge and the corresponding provides. In newest years, deep learning has been acknowledged as a strong feature-extraction gadget to efficiently take care of nonlinear points and broadly utilized in various image processing duties. Motivated by these worthwhile features, deep learning has moreover been launched to classify HSIs and demonstrated good effectivity. This survey paper presents a scientific analysis of deep learning-based HSI classification literatures and compares plenty of strategies for this matter. Specifically, we first summarize the first challenges of HSI classification which might’t be efficiently overcome by typical machine learning methods, and as well as introduce the advantages of deep learning to take care of these points. Then, we assemble a framework which divides the corresponding works into spectral-feature networks, spatial-feature networks, and spectral-spatial-feature networks to systematically analysis the most recent achievements in deep learning-based HSI classification. In addition, considering the reality that obtainable teaching samples inside the distant sensing self-discipline are usually very restricted and training deep networks require quite a few samples, we embrace some strategies to boost classification effectivity, which can current some pointers for future analysis on this matter. Finally, plenty of advisor deep learning-based classification methods are carried out on precise HSIs in our experiments. Deep Learning for Hyperspectral Image Classification: An Overview