Morphological Approach to Quantifying Soil Cracks: Application to Dynamic Crack Patterns during Wetting‐Drying Cycles

Abstract

Core Ideas We propose a methodology for soil crack recognition and quantification. Crack skeleton, outline and nodes are produced using morphological algorithms. A novel method for separating cracks is proposed based on SKIZ, distance transform, and watershed segmentation. The approach is assessed on dynamic crack patterns derived from three wetting‐drying cycles. Crack degree, velocity, similarity and network topology within wet‐dry cycles are analyzed. Water flow and solute transport through shrink‐swell soils with desiccation cracks are dominated by the crack distribution and variable hydraulic properties. The quantification of cracks is therefore inevitable for predicting moisture movement in cracked soils. This study presents a morphological approach to crack recognition and quantification based on digital image processing and morphological algorithms. The methodology comprises three sections: image segmentation, morphological processing and parameter calculation. A crack image is first segmented and then subjected to morphological operations including skeletonization, node identification and crack outline extraction. Based on the resulting binary crack image, skeleton and nodes of crack network, we propose a novel approach to crack separation by employing the concept of skeleton by zones of influence (SKIZ) and algorithms including distance transform and watershed segmentation. Calculations of crack parameters are presented and we apply this methodology to quantify the characteristics of dynamic crack patterns derived from wetting‐drying (W‐D) cycles. Crack degree, propagation velocity of cracks, similarity between crack patterns and network topology are exemplified to delineate the dynamic crack patterns. We find that the cracking degree present hysteresis between W‐D processes, similarly to soil shrink‐swell characteristics and water retention curves. The crack skeleton simultaneously, non‐hierarchically and rapidly propagated in successive drying cycles compared to the initial drying process. The similarity between two crack patterns with similar moisture contents derived from W‐D cycles is positively correlated with the drying intensity, and the evolution of the similarity gradually decelerates with decreasing water content. The crack patterns at full desiccation across W‐D cycles exhibit a marked similarity, with a maximum similarity measure of approximately 0.8. Overall, this approach could achieve efficient recognition and quantification of crack patterns and has great potential in the 3‐D quantification of cracks or related structures.