AgTech ResearchAug 10, 2026•7 min read•01 — Viet Rice AgTech Research

Fresh Paddy Variety Identification & ST25 Purity Verification Using AI Embryo Pattern Analysis

Deep learning models analyzing microscopic embryo patterns and lemma curvature of fresh unhusked paddy (Paddy Rice) to identify ST24, ST25, Jasmine 85, and Dai Thom 8 in 2 minutes without DNA testing.

#Paddy Variety#Embryo Pattern#ST25 Purity#DNA Alternative#Weedy Rice#Seed Certification
Technologies in use:Embryo Morphology Deep Neural NetHigh-Magnification Macro OpticsNational Rice Varietal Vector DBReal-time Edge Inference
Fresh Paddy Variety Identification & ST25 Purity Verification Using AI Embryo Pattern Analysis

1. Paddy adulteration issues and the bottleneck of DNA testing

ST25 commands a 2,000 to 4,000 VND/kg premium over standard varieties. This price spread tempts unscrupulous actors to blend lower-grade paddy (weedy rice, OM5451, IR50404) into ST25 batches.

When grains are unhusked in the field, even experienced inspectors cannot detect a 5–10% mix by eye. Previously, the only verification was sending samples for PCR/DNA tests: this took 5 to 7 days and cost millions of VND per test, which was impractical for fresh paddy needing immediate drying.

Urgent need for rapid seed verification

The industry needed a solution that inspects fresh paddy on-site, delivering precise purity percentages within 2–3 minutes at the intake gate.

2. Embryo pattern principle: Reading biological fingerprints on unhusked paddy

Every rice cultivar exhibits unique microscopic morphological signatures: embryo pocket geometry, sterile lemma angles, husk trichome density, and micrometric aspect ratios.

The Viet Rice optical chamber captures macro images of embryo profiles. Multi-layer convolutional neural nets extract a 512-dimensional vector representing the morphological fingerprint of each grain.

3. Building a vector database for 30+ core Vietnamese rice cultivars

Viet Rice Holdings partnered with agricultural institutes in the Mekong Delta to label over 800,000 reference grains across 30+ varieties: ST24, ST25, Dai Thom 8, Jasmine 85, Nang Hoa 9, OM18, OM5451, and DS1.

The model matches sample vectors against the national reference database, calculating variety probability and batch purity.

“The rice embryo pattern is nature's biometric ID. AI decodes that fingerprint in seconds.”

4. Benchmark: 99.2% accuracy verified against Real-time PCR genetic tests

Blind tests across 500 mixed samples (with 2% to 20% adulteration) showed Viet Rice Purity AI achieved 99.2% accuracy, with R² = 0.978 correlation against laboratory Real-time PCR.

Turnaround was 2 minutes 15 seconds, and cost per scan was less than 1% of molecular testing.

5. Protecting geographical indications and the legacy of award-winning rice

Deployed in Soc Trang, Bac Lieu, and Ca Mau, this technology intercepts degraded and blended seeds before planting and milling, protecting the reputation of Vietnam's flagship rice cultivars.

References & Technical Standards
  • [1]National Center for Biotechnology Information: Seed Variety Identification using Morphological AI
  • [2]Cục Trồng trọt - Bộ Nông nghiệp & Phát triển Nông thôn: Danh mục giống lúa thuần Việt Nam
How useful was this article?

Technical Discussion Forum (1)

Viet Rice Technical Community
Comments are published immediately after submission.
HN
Eng. Huynh Van Muoi
Technical Director, Rice Milling Plant
Just now

The flatbed scanner and kernel segmentation algorithm reduce inspection time from 30 minutes to 2 minutes, addressing critical intake bottlenecks during peak harvest.

Related industry news • Market developments & policy context
Open Newsroom
Knowledge Hub

Articles in the same category

View All