Deep Learning-Driven Adaptive Band Selection for Efficient Hyperspectral Image Analysis

Analysis of hyperspectral images (HSIs) can be highly accurate, however conventional band selection techniques are frequently computationally costly and do not thoroughly examine the spectral-spatial relationship required for accurate interpretation. Existing hand-crafted approaches, such as filter-based, wrapper-based, and hybrid approaches, have drawbacks such as an excessive computing load and an inability to account for spatial correlations. These issues are addressed by the proposed Deep Learning-Based Adaptive Band Selection (DLABS), which uses Convolutional Neural Networks (CNNs) to automate band selection. In addition to providing observable improvements in classification accuracy and savings in processing time and memory usage, the DL-ABS system learns the intricate spectralspatial dependency relationship. The proposed method achieves 92.5%, 94.3%, and 95.1% accuracy for datasets such as Indian Pines, Pavia University, and Salinas, respectively, with reduced memory and computation time. In order to illustrate the advantages of the proposed approach over traditional HSI classification techniques, experimental data are presented. The proposed system's efficiency and scalability for real-time hyperspectral imaging use.