Decoding soil properties from surface cracks using Minkowski functionals, junction crack angle distributions, and...

Area of Science:

• Geosciences
• Soil Science
• Computational Science

Background:

• Desiccation cracks form unique patterns influenced by soil type.
• Previous studies noted pattern variations but lacked advanced analytical tools.
• Physico-chemical properties of soil subclasses from different banks were similar, suggesting pattern analysis is key.

Purpose of the Study:

• To investigate the evolution of desiccation crack patterns across various soil types and subtypes.
• To utilize morphological descriptors and deep learning for accurate soil classification.
• To explore applications in fields like agriculture, planetary science, and geotechnical engineering.

Main Methods:

• Conducted desiccation crack experiments on different soil types.
• Analyzed crack patterns using morphological descriptors (Minkowski functionals, junction crack angle distribution).
• Trained convolutional neural network (CNN) algorithms on experimental image data (60% for training).

Main Results:

• Distinct desiccation crack patterns and descriptors were observed for different soil types and subtypes.
• CNN models achieved 100% prediction accuracy for both major and sub-major soil classifications.
• Morphological descriptors proved effective as soil type fingerprints.

Conclusions:

• Desiccation crack pattern analysis, enhanced by deep learning, offers a highly accurate method for soil classification.
• This approach has broad applicability in diverse scientific and engineering fields.
• The study highlights the potential of image-based soil classification using advanced computational techniques.