A Systematic Review: Remote Sensed Hyperspectral Image Segmentation and Caption Generation Using Deep Learning Methods

Hyperspectral images (HSIs) exhibit a high-dimensional nature, capturing data across numerous wavelengths in the electromagnetic spectrum, often spanning thousands of bands. It has found widespread applications in various real-life scenarios due to its ability to leverage the rich spectral information contained within each pixel. Deep Learning (DL) schemes offer a huge variety of chances to resolve traditional imaging tasks and also for approaching various simulating issues in the spatial-spectral region. This review work provides a systematic review of the relevant existing techniques based on HSI segmentation and image captioning. Initially, other DL methods like, Deep Belief Network (DBN), Convolutional Neural Network (CNN), Autoencoders, Fully Convolutional Neural Network (FCNN), UNet, and Graph Convolutional Network (GCN) are discussed. Secondly, a significant computer vision problem that has recently evolved is image captioning, which tries to automatically produce English explanations of an input image. Therefore, image captioning has garnered growing interest within the realm of remote sensing. This survey summarizes the relevant methods and concentrates on the feature extraction-based methods and attention mechanism-based techniques, which plays a significant role in image caption generation tasks. Finally, it provides the research gaps and its appropriate solution at the end of each survey.