The Effect of Satellite Image Fusion on the Classification Process by Using Multiple Sensors
Abstract
Accurate land use and land cover (LU/LC) classification is essential for various geospatial applications, which demand high spatial and spectral resolution imagery. This study investigates the efficacy of pan-sharpening and hybrid data fusion techniques to enhance Landsat 9 and Sentinel 2A imagery for LU/LC classification. The study aims to compare the classification accuracy of different datasets using multiple classifiers: Support Vector Machines (SVM) with various kernels, Maximum Likelihood (ML), Artificial Neural Networks (ANN), Mahalanobis distance, and Minimum Distance. Datasets encompassed the Landsat (30 m) imagery, Landsat pan-sharpened to (15 m) and (10 m), Sentinel 2 pan-sharpened to (10 m), and the hybrid datasets that combined the un-overlapping bands from both satellites at (10 m). The results show a significant improvement in classification accuracy: from 82.28% for Landsat 30m to 89.91% for Landsat 10m and 91.55% for the hybrid dataset using the linear SVM. The hybrid dataset outperformed the individual Landsat and Sentinel 2 pan-sharpened imagery, especially for spectrally overlapped features, demonstrating the value of fusing the information from multiple sensors. SVM classifier with the linear kernel are emerged as the most accurate and robust classifier. This study concludes that pan-sharpening and hybrid data fusion effectively enhance LU/LC classification accuracy. The SVM classifier with the linear kernel is identified as a suitable choice for this task, offering insights valuable for applications in diverse fields.Datasets encompassed the Landsat (30 m) imagery, Landsat pan-sharpened to (15 m) and (10 m), Sentinel 2 pan-sharpened to (10 m), and the hybrid datasets that combined the un-overlapping bands from both satellites at (10 m). Results showed a significant improvement in classification accuracy: from 82.28% for Landsat 30m to 89.91% for Landsat 10m and 91.55% for the hybrid dataset using the linear SVM. The hybrid dataset outperformed the individual Landsat and Sentinel 2 pan-sharpened imagery, especially for spectrally overlapped features, demonstrating the value of fusing the information from multiple sensors. SVM classifier with the linear kernel emerged as the most accurate and robust classifier.This study concluded that pan-sharpening and hybrid data fusion effectively enhance LU/LC classification accuracy. The SVM classifier with the linear kernel was identified as a suitable choice for this task, offering insights valuable for applications in diverse fields.