Evaluation of an AI-Integrated Laboratory Tool for Estimation of Rice Milling Yield

Highlights This study introduces an AI-based tool for rapid and objective quantification of head rice yield (HRY) and compares its performance to the conventional laboratory method. HRY was estimated using both the conventional method and the MachVision analyzer across five milling durations (10, 15, 20, 30, and 40 s) for twenty-two rice cultivars dried to 12.5% moisture content. The MachVision analyzer demonstrated a strong correlation with the conventional method (r 0.9), with R2 ranging from 85% to 94%, indicating high agreement and predictive accuracy. The MachVision analyzer is a promising tool for efficient and accurate prediction of HRY across multiple cultivars, supporting the integration of AI-based tools into lab and industrial rice milling evaluation. ABSTRACT. Milling is a critical postharvest process in rice production that significantly influences the head rice yield (HRY). This study introduces and evaluates the MachVision rice analyzer, a new machine vision system developed for easy quantification of rice milling assessments. The analyzer quantifies HRY, identifies defects and foreign materials, provides detailed shape characteristics, and classifies kernels as paddy, whole, or broken while generating detailed analytical reports. Twenty-two rough rice cultivars (long-grain purelines and hybrids and medium-grain) at 12.5% moisture content were husked and milled at five different durations (10, 15, 20, 30, and 40 s) using the McGill #2 mill. The milled rice samples were then passed through the MachVision rice analyzer, which employed a CNN model to classify kernels and estimate HRY. The Bland–Altman analysis showed a mean bias of –3 and 95% limits of agreement between –8.30 and +2.41, indicating that the MachVision analyzer slightly underestimated HRY compared to the conventional method but maintained overall good agreement. Further results indicated that the rice analyzer exhibited strong agreement with the conventional method, with correlation coefficient (r) exceeding 0.9 and R 2 value ranging from 0.85 to 0.94 across the milling durations. HRY estimates were consistent across most cultivars, with notable discrepancies observed in cultivars such as DG263L, Ozark, and ProGold1, and lesser variations seen in Diamond and PVL03. Overall, these findings highlight the potential of the MachVision analyzer as a rapid, consistent, and accurate tool for HRY estimation. Its application could benefit breeders, millers, and researchers by improving the efficiency and objectivity of rice milling assessment in both laboratory and industrial settings, while also supporting the broader integration of machine vision and AI-based tools within the rice industry. Keywords: Head rice yield, Machine vision, MachVision, Milling, Rice.