Assessment of Grain Harvest Moisture Content Using Machine Learning on Smartphone Images for Optimal Harvest Timing

Area of Science:

• Agricultural Engineering
• Remote Sensing
• Data Science

Background:

• Grain moisture content (GMC) is crucial for determining optimal rice harvest times.
• Traditional GMC testing methods are time-consuming, labor-intensive, and impractical for large-scale application.
• Accurate and timely GMC data is essential for efficient harvesting and yield optimization.

Purpose of the Study:

• To develop a cost-effective, non-destructive method for assessing rice GMC using smartphone imagery.
• To establish an image-based model for predicting GMC and enabling multi-day harvest forecasting.
• To facilitate optimized harvesting schedules and agricultural machinery deployment.

Main Methods:

• Smartphone images of rice panicles were captured and spectrally-geometrically corrected.
• A dataset of 517 valid samples was used to train and test machine learning models for GMC assessment.
• Principal component analysis identified key color indices for model development, including Random Forest, Multilayer Perceptron, Support Vector Regression (SVR), and Multivariate Linear Regression.

Main Results:

• The Support Vector Regression (SVR) model demonstrated high suitability for GMC assessment (below 40%) with a Mean Absolute Error (MAE) of 1.23%.
• The developed model accurately predicts GMC from image data, correlating with on-site measurements.
• The study successfully validated the use of smartphone imaging for non-destructive GMC estimation.

Conclusions:

• Smartphone-based, non-destructive GMC measurement offers a real-time, cost-effective alternative to traditional methods.
• This technology enables on-farm prediction of optimal harvest dates, improving operational efficiency.
• The findings support enhanced harvesting scheduling and better management of agricultural machinery.