Spatial Resolution and Noise impact on Multi Band Hyperspectral Image Processing

Abstract

Hyperspectral imaging has become a vital tool for analysing various applications covering environmental monitoring and agricultural sectors. However the effectiveness of hyperspectral image processing can be significantly affected by spatial resolution and noise; pertaining to inaccurate classification outcomes. This study investigates the impact of these two factors (spatial resolution and noise) on multi-band hyperspectral image processing. This study specifically employs the ResNet architecture, a deep learning model known for its effectiveness in image classification tasks. The study highlights the effects of spatial resolution on the performance of ResNet by showcasing that lower spatial resolutions can result in a loss of critical spectral and spatial details. This loss leads to challenges in feature extraction and classification accuracy of trained learning models. Through a series of experiments, we demonstrate how variations in spatial resolution impact the ability of ResNet to accurately classify hyperspectral data. Secondly the effect of Noise on ResNet model reveals that while ResNet exhibits robustness against certain noise types, its performance deteriorates under high-noise conditions when specifically spatial resolution increases. To address this issue, we propose a pre-processing techniques aimed at noise reduction, assessing their effectiveness in enhancing the classification accuracy of ResNet. Overall, our study highlights the intricate relationship between spatial resolution, noise, and machine learning performance in hyperspectral image processing.