A Novel Algorithm for Efficient Hyperspectral Image Preprocessing

Objectives: The main objective of this work is to develop an efficient preprocessing algorithm for classifying image bands of Hyperspectral sensor. GUI facilitate preprocessing for major group of researchers in this domain. Methods: During image acquisition using a Hyperion sensor, certain bands may not be captured properly, these are bad bands. These bad bands pose a significant hurdle in spectral analysis. A novel algorithm for identifying and classifying bad bands has been designed using statistical analysis of image bands. The machine learning classifiers the Hybrid K-nearest neighbor and Clustering techniques are used to develop the model and improve the accuracy of the spectral classification. Findings: The proposed algorithm effectively identified and classified bad bands. The Hybrid K-nearest neighbor classifier delivered 100% accuracy whereas the decision tree classifier shows 98% accuracy. The developed graphical interface significantly improved usability and efficiency in hyperspectral image preprocessing. The proposed work identified two significant thresholds 0.5 and 10.5 for bad band identification. Novelty: This work gives new benchmark for hyperspectral image preprocessing with highest accuracy. Keywords: Bad Band Identification, Hyperion, Hyperspectral, Preprocessing, Remote Sensing