Comparing Pan-sharpened Landsat-9 and Sentinel-2 for Land-Use Classification Using Machine Learning Classifiers

摘要

This paper evaluates the ability of two machine learning algorithms, Random Forest (RF) and Support Vector Machine (SVM), to generate land-use maps using the recently launched Landsat-9 and Sentinel-2, two of the most used and popular satellite imagery sources. The potential to improve Landsat-9 performance was tested by pan-sharpening different bands of high-resolution data (15 m). For optimal performance of both classifiers, model tuning methods were applied by trying different combinations of key parameters of each model. This comparison was made in two different areas in Central Morocco. The results show that SVM performs slightly better than RF in classifying two images. In addition, Sentinel-2 exhibits significant multivariety classification ability compared to the pan-sharpened Landsat-9, despite the improved resolution of the latter. Lastly, the best classification performances were recorded for the combination Sentinel-2/SVM classifier. At last, machine learning algorithms prove their efficiency in classifying satellite images with high performance.