Hyperspectral Target Detection Using Long Short-Term Memory and Spectral Angle Mapper

Abstract

Hyperspectral images are obtained by dividing the electromagnetic spectrum into hundreds of narrow bands. Thanks to this feature, hyperspectral imaging is successful in distinguishing surface materials and is frequently used in target detection. In this study, long short-term memory and spectral angle mapper are used to detect targets in images obtained from the VNIR sensor. Deep neural networks require annotated data related to each target and also background classes for target detection in hyperspectral images. In this study, the background objects are eliminated by using the spectral angle mapper as a kind of filter, and the long short-term memory is only trained on the candidate target signatures. Therefore, data annotation activities are carried out only for candidate target classes and data annotation cost is reduced. In addition, the experimental results show that the long short-term memory model, which is trained on signatures collected from 30 meter heights, detects targets successfully independently of height.