Automatic Classification Method of Quaternary Lithology in Vegetation Cover Area Combining Spectral, Textural, Topographic, Geothermal, and Vegetation Features

Abstract

To explore the automatic classification method of Quaternary lithology in vegetation covered areas is significantly helpful to improve the efficiency of Quaternary lithology mapping. Due to the vegetation cover and human modification effects, the traditional lithology identification methods based on image spectra and textures are often challenging to be effective. This paper uses multi-source remote sensing data such as OLI, TIRS, and ASTER GDEM to extract multiple types of spectral (SPEC), textural (TEX), topographic (TOPO), geothermal (TEM), and vegetation (VEG) based on principal component transform, gray co-occurrence matrix, topographic factor calculation, thermal radiation transport model and vegetation index in the Quaternary distribution area of Viet Chi, Vietnam. Remote sensing features were selected and combined to form 16 kinds of classification datasets. The lithological units was automatically classified using the random forest method, The method’s accuracy was evaluated to study the effectiveness of multi-type remote sensing features on the automatic classification of Quaternary lithology in vegetation cover area. The results show that the geothermal, textural, and topographic features can effectively improve the lithological classification accuracy, and the overall classification accuracy is improved by 0.32%, 0.87%, and 2.25%, respectively, compared with the use of spectral data alone. Among the 16 classification datasets constructed, the dataset combining spectral, textural, topographic, and geothermal features (SPEC+ TEX+ TOPO+ TEM) obtained the highest automatic lithology classification accuracy of 80.99%. This study can provide a technical idea for rapid differentiation of regional Quaternary surface sediment lithologies.

Lithological classification and analysis based on random forest and multiple features: a case study in the Qulong copper deposit, China

“…Polarization features are acquired through SAR bands that penetrate vegetation and shallow soil, providing insights into rock morphology and structure 27 . Texture features play a crucial role in capturing the spatial distribution and organizational structure of surface objects, along with their relationships with the surrounding environment 28 . Terrain features provide valuable insights into variations in erosion and weathering across different lithologic areas, which can be employed in lithological mapping 29 …”

“…27 Texture features play a crucial role in capturing the spatial distribution and organizational structure of surface objects, along with their relationships with the surrounding environment. 28 Terrain features provide valuable insights into variations in erosion and weathering across different lithologic areas, which can be employed in lithological mapping. 29 Given the large number of features, the PDF and the Bhattacharyya distance 30 were used to evaluate the distinguishability of each feature for different lithologies.…”

Machine learning algorithms for lithological mapping using Sentinel-2 and SRTM DEM in highly vegetated areas

“…These methods establish a solid basis for the rapid classification of rocks using "geobotany" principles in remote sensing. Furthermore, HAN utilized the RF method to map Quaternary rock (including Pleistocene gravel, Holocene sand, Holocene clay, and Holocene gravel) in vegetation-covered areas of Vietnam based on multiple remote sensing data sources, achieving OA of 80.99% (Han et al, 2021). This highlights the potential of the RF algorithm in geological mapping.…”

“…High-resolution optical and radar remote sensing data, along with terrain information, are valuable for extracting rock-type information from densely vegetated areas. In the past decade, medium-resolution remote sensing imagery such as Landsat series and ASTER, has been extensively employed for rock type mapping in vegetated areas (Knepper, 1989;Langford, 2015;Han et al, 2021;Zeng et al, 2023). It establishes a strong foundation for rock-type identification by offering cost-effective, wide coverage, high spatial resolution (Chen et al, 2022;Zou et al, 2022), valuable indications of vegetation and rock-soil information, rich surface information and a small mixed pixel effect (Meroni et al, 2021).…”

“…An appropriate algorithm is one of the key factors contributing to achieving satisfactory classification results. Machine learning algorithms such as maximum likelihood (ML) (Grebby et al, 2011), partial least squares discriminant analysis (PLSDA) (Lu et al, 2021), support vector machine (SVM) (Othman and Gloaguen, 2014;Bachri et al, 2019), and random forest (RF) (Han et al, 2021) have been extensively used for rock classification in vegetation-covered areas because of the rapid advancement of machine learning. In Grebby's study, airborne multispectral imagery and laser scanning data were used to map rock types in the Troodos ophiolite.…”