Abstract
As one of the precursor characteristics of slope deformation, surface cracks can provide predictive information for early identification of geological hazards and determination of kinematic instability characteristics. Constrained by terrain conditions, manual inspection suffers from low efficiency, while identification using single remote sensing data also struggles to address crack size effects and noise filtering issues in complex backgrounds. To efficiently acquire the distribution and geometric information of surface cracks on deforming slopes, point clouds and Digital Orthophoto Maps (DOM) obtained through UAV terrain-following flight are employed as data sources. First, six algorithmic models utilizing point cloud roughness, slope, dispersion, digital image pixel gradient, grayscale value, and RGB (red green blue) value as features are employed to achieve preliminary identification of slope cracks, with Receiver Operating Characteristic (ROC) curve testing conducted for different models to determine segmentation thresholds. Second, three filtering algorithms based on morphological repair and crack direction, length, and frequency indexed by density-based clustering algorithms are applied to process background noise in the initial extraction results, capable of removing up to 82.7% of noise while causing minimal crack distortion. Then, binary classification model evaluation metrics are adopted to analyze the performance of the six filtered crack extraction results, and an optimal detection model (F1=0.8350) is obtained through data fusion to address crack size effects. Finally, six quantitative characteristic indicators—quantity, length, width, direction, dispersion, and crack density—are automatically calculated based on crack skeletons and contours. The results demonstrate that multi-dimensional data fusion can resolve the spatial scale effect in surface crack identification, and the filtering approach based on crack unit indexing is applicable for noise removal in large-scale complex surface scenarios.
Full Text
Preamble
Automatic Detection of Deformation Cracks in Slopes Fused with Point Cloud and Digital Image Data
DENG Bo, XU Qiang, DONG Xiujun, JU Yuanzhen, HU Wuting
State Key Laboratory of Geohazard Prevention and Geo-environmental Protection, Chengdu University of Technology, Chengdu 610059
Abstract
Surface cracks represent a key precursor to slope deformation, offering critical predictive information for early identification of geological hazards and characterization of instability mechanisms. However, manual inspection is inefficient due to challenging terrain conditions, while single-source remote sensing data fails to resolve crack scale effects and noise filtering issues in complex environments. To efficiently acquire the spatial distribution and geometric properties of surface cracks on deforming slopes, this study employs point clouds and digital orthophoto maps captured via UAV terrain-following flights as primary data sources.
First, six algorithmic models are developed for preliminary crack detection, utilizing point cloud roughness, slope, and dispersion, as well as digital image pixel gradient, grayscale values, and RGB (red-green-blue) values as distinct features. Receiver operating characteristic (ROC) curve analysis is performed for each model to determine optimal segmentation thresholds. Second, three noise-filtering algorithms—morphological restoration and density-based clustering indexed by crack orientation, length, and frequency—are applied to process background noise in the initial extraction results, removing up to 82.7% of noise while minimizing crack distortion.
Third, binary classification evaluation metrics are employed to assess the performance of the six filtered crack extraction results, yielding an optimal detection model through data fusion that addresses crack scale effects (F1-score = 0.8350). Finally, six quantitative characteristic indicators—quantity, length, width, orientation, dispersion, and crack density—are automatically computed based on crack skeletons and contours.
The results demonstrate that multi-dimensional data fusion effectively resolves spatial scale effects in surface crack identification, while the crack-unit-indexed filtering approach is well-suited for noise removal in large-scale complex terrain scenarios.
Keywords: UAV; geological hazards; slope deformation cracks; automatic extraction; information statistics