Categorizing hyperspectral imagery using convolutional neural networks for land cover analysis

Abstract

Categorizing hyperspectral imagery (HSI) is crucial in various remote sensing applications, including environmental monitoring, agriculture, and urban planning. Recently, numerous approaches have emerged, with convolutional neural network (CNN)-based algorithms demonstrating remarkable performance in HSI classification due to their ability to learn complex spatial-spectral features. However, these algorithms often require significant computational resources and storage capacity, which can be limiting in practical applications. In this study, we propose a novel CNN architecture tailored for HSI classification within the spectral domain, focusing on optimizing computational efficiency without compromising accuracy. The architecture leverages advanced spectral feature extraction techniques to enhance classification performance. Experimental evaluations on multiple benchmark hyperspectral datasets reveal that the proposed approach not only improves classification accuracy but also achieves a superior balance between performance and computational demand compared to traditional methods like K-nearest neighbors (KNN) and other deep learning-based techniques. Our results demonstrate the potential of the proposed CNN model in advancing the field of HSI classification, offering a viable solution for real-world applications with constrained computational resources.