Quantifying Desiccation Cracks for Expansive Soil Using Machine Learning Technique in Image Processing

The formation of desiccation cracks has detrimental effects on the hydraulic conductivity that affects the overall mechanical strength of expansive soil. Qualitative analysis on the desiccation cracking behaviour of expansive soil provided understanding of the subject based on various concepts and theories, while quantitative analysis aided these studies through numerical supports. In this study, a machine learning technique in image processing is developed to evaluate the surface crack ratio of expansive soil. The desiccation cracking tests were conducted on highly plastic kaolinite slurry samples with plasticity index of 29.1%. Slurry-saturated specimens with thickness of 10 mm were prepared. The specimens were subjected to cyclic drying-wetting conditions. The images are acquired through a digital camera (12 MP) at constant distance to monitor the desiccation cracks. The images are then pre-processed using OpenCV before crack feature extraction. In this study, a total of 54 desiccation crack images were processed, along with 8 images from trial test to train the model. The processed images are used to quantify the desiccation cracks by evaluating surface crack ratio and average crack width. It was identified that the accuracy of the model for the quantification of surface crack ratio and average crack width were 97.24% and 93.85% respectively with average processing time of 1.51s per image. The results show that the model was able to achieve high accuracy with sufficient efficiency in determining important parameters used for crack characterization.