Applying AI Computer Vision and IoT Scanning Devices in Grain Quality Inspection & Rice Grading

In the rice value chain, assessing incoming paddy quality at mill intake gates and verifying polished milled rice quality before port export have historically relied on manual labor. Quality inspectors count and sort grains with tweezers on a 50g sample plate, spending 25 to 30 minutes per sample.

This manual method has major drawbacks: subjective human bias, high error rates caused by eye fatigue, inability to inspect high sample volumes during peak harvest seasons, and constant price disputes among farmers, middle traders, and mill operators.

The Viet Rice AgTech R&D unit designed GV GrainScan with two tightly integrated components: (1) An IoT flatbed optical scanner featuring an anti-reflective sample tray, coaxial multi-spectral LED lighting, and a 4K industrial CMOS sensor; (2) An embedded edge AI pipeline executing a TensorRT-optimized Grain Instance Segmentation model.

When a sample tray holding 1,500 to 3,000 grains is placed into the scanner, the system captures multi-directional images, segments each individual grain from the background, and isolates boundaries with 0.05 mm precision.

Viet Rice Holdings's multi-label grain classifier is trained on over 2.5 million labeled grain samples curated by certified international grain inspectors. The algorithm categorizes 8 standard quality parameters:

• Head rice vs. big and small broken kernel ratios. • Chalky kernel ratio distinguishing between dorsal, abdominal, and complete chalkiness. • Heat-yellow kernels, insect-damaged, and moldy kernels. • Red and red-streaked kernels, immature green kernels, and foreign paddy seeds.

A major breakthrough of Viet Rice Holdings is verifying variety purity directly on unhusked fresh paddy. By using deep learning models analyzing embryo grooves, sterile lemma curvature, and morphological aspect ratios, the system accurately identifies premium cultivars such as ST24, ST25, Dai Thom 8, Jasmine 85, Nang Hoa 9, and OM5451.

The system computes purity percentage (e.g., ST25 batch with 98.2% purity and 1.8% weedy rice adulteration) in 2–3 minutes, alerting mills to adulterated batches before they enter drying kilns.

GV GrainScan has undergone rigorous field trials at large-scale rice milling clusters in An Giang, Can Tho, and Dong Thap. Key results include:

• Sample processing speed increased 15x compared to manual inspection, clearing queues for over 60 paddy trucks daily during peak harvest. • Correlation R² = 0.985 with ISO 17025 accredited laboratory test results for broken rice and chalkiness. • 1.8% increase in head rice yield due to timely parameter tuning for huller rollers, saving hundreds of millions of VND per 10,000 tons of processed paddy.