Munsell Soil Colour Prediction from the Soil and Soil Colour Book Using Patching Method and Deep Learning Techniques.

Area of Science:

• Soil Science
• Computer Vision
• Machine Learning

Background:

• Soil colour is a crucial indicator of soil health and agricultural performance.
• Traditional Munsell colour charts are subjective and prone to user perception errors.
• Mobile devices offer accessible, high-quality image capture capabilities.

Purpose of the Study:

• To develop a deep learning model for predicting Munsell soil colour (MSC) using mobile-captured images.
• To evaluate the effectiveness of a patch-based mechanism for dataset enrichment.
• To determine the optimal deep learning technique for MSC prediction at both page and chip levels.

Main Methods:

• Utilized deep learning techniques on smartphone-captured images of the Munsell Soil Colour Book (MSCB).
• Implemented a patch-based mechanism to augment the dataset for training deep learning models.
• Evaluated prediction accuracy at both page and chip levels using various deep learning methods.

Main Results:

• Patch-based mechanism significantly improved accuracy, achieving around 95% for both page and chip-level predictions.
• Without patching, chip-level accuracy was below 40% and page-level below 65%.
• Identified the most effective deep learning techniques for MSC prediction.

Conclusions:

• Deep learning with a patch-based approach offers a highly accurate and scalable method for digital soil colour prediction.
• The proposed technique demonstrates potential for real-world soil analysis with limited samples.
• This digital approach overcomes the limitations of traditional subjective Munsell colour assessments.