A Comprehensive study on the different types of soil desiccation cracks and their implications for soil...

Area of Science:

• Geology
• Soil Science
• Computer Science

Background:

• Soil desiccation and fracture patterns are influenced by soil type.
• Characteristic crack patterns can serve as a unique identifier for different soil textures.
• Traditional methods for analyzing crack patterns can be labor-intensive.

Purpose of the Study:

• To investigate the potential of desiccation crack patterns for soil identification.
• To evaluate the effectiveness of combining conventional analysis with deep learning techniques.
• To develop a robust method for soil type classification using crack imagery.

Main Methods:

• Analysis of crack patterns in clay, silt, and sandy-clay-loam soils from the Brahmaputra river basin.
• Application of fractal dimension analysis and Euler numbers for crack pattern characterization.
• Utilizing deep learning, specifically feed-forward neural networks, on soil crack images.

Main Results:

• Fractal dimension analysis proved to be a valuable pre-processing step for deep learning.
• Data augmentation significantly improved the robustness and accuracy of the neural network model.
• An accuracy of 92.09% was achieved in soil identification, even when noise was introduced to mimic real-world conditions.

Conclusions:

• The integration of conventional crack pattern analysis with deep learning algorithms offers an effective approach for soil type identification.
• This hybrid methodology demonstrates high accuracy and robustness, suitable for practical applications in soil science.
• Desiccation crack patterns provide a reliable basis for automated soil classification systems.