Non-destructive image analysis for assessing rice kernel integrity during milling

Introduction: Conventional evaluation of rice milling quality largely relies on end-point measures such as head-rice yield and broken-rice percentage to quantify the final milling outcome. However, they provide limited information about the lost kernel integrity along the processing sequence due to a lack of stage-specific non-destructive diagnostics to track kernel integrity throughout and the milling process.

Materials and methods: A rapid, non-destructive flatbed scanner-based method was used to describe rice kernel morphology and stage-wise changes in milling quality. Aromatic long-grain Pishori rice was sampled at brown-rice, white-rice-before-grading, and white-rice-after-grading points over three weekly production batches. Approximately 10 g from each stage sample was scanned at 200 dpi and analysed with seed analyser software to obtain kernel length, breadth, and aspect ratio. The analysis distinguishes standard rice-integrity definitions from a study-specific long-grain morphology screen and reports head rice as an image-derived kernel-count proportion rather than gravimetric recovery.

Results: Brown rice had the highest long-grain morphology-screen proportion (63.98%) and image-derived head-rice kernel proportion (72.17%). White rice before grading had the lowest mean length (5.10 mm) and the highest broken-kernel proportion (43.12%), while the post-grading stream contained a higher proportion of longer kernels (head-rice kernel proportion 68.12%; broken-kernel proportion 31.88%). The largest descriptive deterioration occurred between the brown-rice and pre-grading sampling points, a process interval that includes whitening and associated transfer operations.

Conclusions: The findings show that a flatbed scanner-derived morphology can reveal stage-wise shifts in kernel integrity. Therefore, this experimental method could be suitable for routine monitoring of head-rice recovery, broken-rice generation, and process adjustments in rice mills. Replicate-aware validation using fixed-reference thresholds is recommended before predictive or causal claims are made.