Classification of Landsat 8 Images Using Convolutional Neural Network Based on Minimum Noise Fraction Transform

Abstract

The use of remote sensing methods has transformed environmental management and regional planning by allowing the identification of items or phenomena on the Earth's surface. However, noise in picture data remains a chronic difficulty in this discipline, compromising spatial resolution and object detection accuracy. The purpose of this study is to improve the classification accuracy of Landsat 8 pictures by developing a Convolutional Neural Network (CNN) based on the Minimum Noise Fraction (MNF) transform. The goal is to evaluate MNF's efficacy in compressing and organizing multispectral images, hence reducing the influence of noise on picture categorization. The MNF transform is used to Landsat 8 image data to remove noisy bands before adopting CNN as a supervised classification approach. The current study takes use of CNN's inherent benefits in dealing with high-dimensional data, learning complicated representations, and automatically extracting key features from pictures, while simultaneously evaluating MNF's efficiency in increasing image quality. The findings show that using MNF as a preprocessing step produces images with improved quality and organization. Subsequent classification using CNN obtained an astounding accuracy of 97.41%, with a great representation of the study region and varied land use categories, highlighting the synergy between MNF and CNN in improving classification performance. The article suggests that combining MNF transform with CNN enhances classification accuracy of Landsat 8 pictures, with positive implications for developments in environmental monitoring, land use mapping, and remote sensing technologies.