Efficient vegetation filtering method using K-means clustering algorithm: case study of high steep rock slope

Vegetation obscures critical rock‐mass features on steep slopes, degrading the reliability of structural surface interpretation from point clouds. We propose a fast vegetation‐filtering approach tailored to high‐steep, vegetated rock slopes. The method aims to suppress vegetation noise while preserving terrain points essential for structural analysis. We first perform dual-channel dimensionality reduction by combining Principal Component Analysis (PCA) on spatial coordinates with the Red–Green Difference Index (RGDI) from RGB values, then apply K-means clustering for segmentation. A hierarchical grid plus local plane fitting is used to select vegetation seed points; distances to the fitted plane guide seed assignment and subsequent cluster-level filtering. To prevent over-filtering, a preservation mechanism based on the 3σ rule retains 5%–32% of points near the seed-point distance threshold. The method was evaluated on a rugged, vegetation-covered slope at Tiantai Mountain (Zhejiang, China) acquired with a Topcon GLS-2000 (384,663 points). The effectiveness of the filtering results is evaluated using class error, I e , II e and total error A e . Comparing with other algorithms, it is found that the class error, I e is 7.79%, the class error II e is 4.34%, and the total result is limited to 6.53%. By jointly leveraging spatial and spectral cues with grid-wise plane fitting and a preservation guarantee, the approach effectively suppresses vegetation noise while retaining terrain detail needed for downstream tasks (e.g., structural plane interpretation). The results indicate improved filtering accuracy and robustness for high-steep terrains relative to traditional methods.