A U-Net Based Approach for High-Accuracy Land Use Land Cover Classification in Hyperspectral Remote Sensing

Deep learning has been demonstrated to have significant potential in the classification of hyperspectral images (HSI). Hyperspectral imaging has gained more recognition in recent years in the area of computer vision research. Its potential to handle remote sensing-related issues, particularly those related to the agricultural sector, has led to its rising popularity. Due to the significant spectral band redundancy, the small number of training samples, and the non-linear relationship between spatial position and spectral bands, HSI classification is a challenging task. Therefore, we propose machine and deep learning-based models to classify the land features with the highest accuracy. Effective bands have been discovered by applying principal component analysis (PCA) to minimize the dimensionality of hyperspectral images. In this work, we evaluate the land use and land cover (LULC) classification efficiency of three different algorithms like Support Vector Machine (SVM), Spectral Angle Mapper (SAM) and U-Convolutional Neural Network (U-net).Using Hyperion images, we demonstrate and evaluate the findings from each method. In this work, we apply deep convolutional neural networks for the classification of high-quality remote sensing images. Semantic image segmentation is used for U-Net frameworks. In the Nagpur district, we map the existence or lack of vegetation and agricultural land using the U-net neural network architecture for Hyperion images.