EFFICIENT SPATIAL–SPECTRAL REPRESENTATION LEARNING FOR THE CLASSIFICATION OF HYPERSPECTRAL IMAGES USING A 2D CNN FRAMEWORK
Abhishek Gupta, Pankaj Agarwal, Pramod Kumar Soni
Abstract: Recent progress in remote sensing technologies has facilitated to capture the information in hundreds of spectral bands called hyperspectral imagery (HSI). HSI contains a huge amount of information and can facilitate precise discrimination of material and land cover types and can be used in a wide range of applications. Although information extraction from HSI imagery is a tedious task due to its high dimensions and intermixing of pixels in information in different spectral bands. To mitigate all these issues in this work, a computationally efficient CNN framework, ESSL, is designed that effectively extracts spatial–spectral representations from HSI using patch-based learning without relying on computationally expensive 3D convolutions and enables hierarchical receptive field growth, allowing the network to capture both fine-grained textures and broader spatial contexts relevant to land-cover classification. The experiments are performed on well-known publicly available datasets and performance is evaluated on different evaluation metrics and it has been observed that the proposed method has obtained an OA of 99.96 ± 0.0003, 98.70 ± 0.0036 on the Pavia University and Indian Pines datasets, respectively.
Keywords: Hyperspectral Image; CNN; Landcover; Deep Learning; Environment; Classification
DOI: 10.24874/PES08.03.017
Recieved: 11.06.2026 Revised: 25.07.2026 Accepted: 30.08.2026
UDC:
Reads: 10 
