Lithology classification integrating multi-source remote sensing data after vegetation suppression: a case study from Inner Mongolia Autonomous Region, China

Abstract

Reducing vegetation disturbance in remote sensing images enhances lithology classification accuracy. This study utilized Gaofen-2 (GF-2), Sentinel-2A, Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER), and Gaofen-3 (GF-3) satellite images of Duolun County, Inner Mongolia. A vegetation coverage-based image filtering method was introduced to minimize vegetation interference in multispectral images, while an improved water cloud model mitigated interference in SAR backscattering images. Subsequently, a 63-dimensional feature sequence was extracted from the vegetation-suppressed images. A comparison experiment using the Support Vector Machine (SVM) classifier, both before and after vegetation suppression, was conducted. Results indicated that vegetation suppression improved the Overall Accuracy (OA) by 2.05% and the Kappa coefficient by 0.02. Specifically, the OA and Kappa coefficient for the 63-dimensional features post-suppression reached 91.52% and 0.90, respectively.