Hyperspectral Image Classification using Deep Neural Network

Hyperspectral imaging (HSI) captures detailed spectral and spatial information across a wide range of wavelengths, offering a unique capability to differentiate and classify materials based on their spectral signatures. However, the high dimensionality of HSI data presents challenges in feature extraction and classification, particularly when dealing with limited labeled data. This research paper reviews various feature extraction techniques for HSI, with a focus on unsupervised methods due to their potential in managing unlabeled datasets. The paper examines the application of Convolutional Neural Networks (CNNs), specifically ResNet50 and U-Net architectures, for efficient feature extraction and classification. A comparative analysis of these models shows that while U-Net offers slightly better accuracy, ResNet50 provides more consistent performance with faster processing times. Through this analysis, the study highlights the trade-offs between accuracy and computational efficiency, offering insights for future advancements in HSI feature extraction. These results underscore the trade-offs between precision and speed, providing valuable insights for real-world applications such as agriculture, medical diagnostics, and environmental observation.