Enhancing Hyperspectral Image Classification Performance Using Adaptive Spectral Band Selection and Clustering-based Mutual Information (ASBS-CMI)

Hyperspectral imaging is an emerging tool in remote sensing applications for acquiring detailed spectral data. This paper introduces the Adaptive Spectral Band Selection and Clustering-based Mutual Information (ASBS-CMI) approach, employing adaptive spectral band selection through clustering and mutual information to enhance hyperspectral imaging classification. The ASBS-CMI method includes preprocessing for data quality improvement, clustering to group similar pixels, mutual information calculation for identifying informative spectral bands, and subsequent feature extraction from selected bands for classification. Principal Component Analysis is meant for feature extraction to capture spatial-spectral information. Class labels are assigned to hyperspectral data using a Random Forest classifier. Experimental results on Pavia University Scenes and Indian Pines datasets, along with comparisons to Support Vector Machine and Feed Forward Neural Network, demonstrate improved accuracy within a limited time. The proposed methodology's efficiency is evidenced by evaluating performance through various metrics, showcasing its potential to advance hyperspectral image analysis in remote sensing applications.