Journal of Agricultural Engineering (India) Bioengineering

<b>Machine Learning-based </b><b>Predictive Modelling of Seed Mass and Terminal Velocity through Physical and Morphological Characteristics of Selected Seeds</b><b></b>

Shubham Chaudhary, Tapan Kumar Khura, Pramod Kumar Sahoo, Roaf Ahmad Parray, V. Shobhan Naik, Ajay Kumar, Rajeev Kumar, Pravin Kumar Upadhyay, Md Yeasin
Springer Nature · 2026-08-31 · DOI 10.52151/jae2026633.2038

Abstract

Precise measurement of seed mass and terminal velocity based on basic morphological characteristics is important for the design of pneumatic metering, seed handling, and grading systems used in precision agriculture. This study evaluated the performance of seven different machine learning models for predicting seed mass and terminal velocity of maize (Zea mays), soybean (Glycine max), and okra (Abelmoschus esculentus), utilizing common measurable morphological characteristics. A total of 300 seeds (100 from each species) were analysed to determine their key dimensional attributes. The dataset was divided into training (70%), validation (15%), and testing (15%) subsets for machine language based predictive modelling, and the reliability of model was evaluated using five-fold stratified cross-validation. Model interpretability was investigated through SHapley Additive exPlanations (SHAP) analysis that provided the contribution of each feature to model's predictions. For seed mass prediction, the Random Forest model achieved the highest accuracy, with an R² of 0.9994, RMSE of 2.831 mg, and MAE of 1.84 mg, while requiring only 4.43 milli-seconds (ms) for prediction. The Decision Tree model produced the fastest predictions (0.80 ms) with comparable accuracy (R² = 0.9993) for seed mass prediction. For terminal velocity prediction, the Decision Tree outperformed the other models, achieving an R² of 0.9978, RMSE of 0.089 m s⁻¹, MAE of 0.0400 m s⁻¹, and a prediction time of 1.27 milli-seconds. SHAP analysis revealed that for seed mass prediction, geometric mean diameter (GMD) was the most influential feature. However, for prediction of terminal velocity, the most influential factor was seed length. These findings highlight the critical role of seed diameter and seed length in determining seed mass and aerodynamic behaviour. Thus, the proposed approach of machine learning based predictive modelling of critical seed properties approach creates a strong foundation for the development of a smart multi-crop pneumatic metering system for different crops.

Read at publisher →