Certain investigations on hyperspectral image classification using deep learning techniques

Abstract

Hyperspectral Remote Sensing is the acquisition of images across the visible, near-infrared, mid-infrared and thermal infrared portions of the electromagnetic spectrum in several narrow, contiguous spectral bands. Hyperspectral images (HSI) furnish ample spectral information to identify and distinguish spectrally unique materials and also provides the potential for more accurate and detailed information extraction than possible with any other type of remotely sensed data. Due to its high discriminative ability, it plays an important role in applications such as precision agriculture, land-use monitoring, water resource management, mining, space exploration, change detection, defense, and environmental monitoring. HSI classification is the most vibrant area of research in the hyperspectral community and has drawn vast attention in the remote sensing field. Due to the high dimensionality of the data, inadequate datasets, big data, transfer learning and the limited availability of training samples, HSI presents significant challenges for classification. In order to overcome the aforementioned issues, various Deep Learning (DL) based architectures are being developed, presenting great potential in HSI data interpretation and classification. As new DL techniques emerge in recent years, classification of remotely sensed images with DL have achieved significant breakthrough in this area. The significant highlights of the DL techniques are: DL outperforms other techniques if the data size is large. DL is highly versatile in the data types supported. In particular, DL takes advantage of HSI data in spectral and spatial domains, both separately and in a coupled fashion. newline