China builds the measurement layer for livestock AI — an industry standard and a national behaviour-annotation tool

East-Asia (China); national, with Chongqing as the standard's lead province

Content

Two items, one function: China is building the layer that makes livestock AI comparable, auditable and shareable — the part that is usually missing everywhere.

1. A standard for pig-farm liquid-feeding digital management. 《猪场液态饲喂数字化管理系统技术要求》(NY/T 5655—2026) is, per Chinese reporting, the first agricultural industry standard in the field of pig-farm liquid-feeding digital management, filling a domestic standards gap. It was drafted under the lead of the Chongqing Animal Husbandry Technology Extension Station with the municipal pig-industry innovation team and twelve units including Henan Hesun Automation Equipment (河南河顺自动化设备股份有限公司), and it passed expert review on 11 October 2025 under MARA’s agricultural informatisation standardisation technical committee, issuing as an NY/T standard in 2026. The problem it names is explicitly a data problem: breaking “data islands” (数据孤岛) — the situation where each farm’s liquid-feeding system speaks its own format and no cross-farm comparison is possible. The lead drafter was 朱燕, a senior livestock engineer at the Chongqing station.

2. A national behaviour-annotation tool for livestock. In December 2025 CAAS’s Institute of Animal Science smart-livestock innovation team published HABLer — Humanoid Animal Behavior Labeler, described as the first system to fuse a large vision-language model with traditional computer vision, behaviour quantification and expert knowledge for animal behaviour annotation. The pipeline: animal detection, instance segmentation and keypoint recognition identify individuals in video; behaviour metrics supply quantitative, explainable evidence; the vision-language model reasons about behaviour semantics; expert interaction corrects and updates policy, giving continuous learning.

Reported results (pigs and dairy cattle; zero-shot or few-shot):

• Short-behaviour tasks: mean frame accuracy 0.87, temporal consistency 0.94.
• A 125-minute continuous pig-posture annotation task: expert annotation time cut by roughly 60%, with initial prediction accuracy holding near 94%.

The paper appeared in Computers and Electronics in Agriculture (DOI 10.1016/j.compag.2025.111307); first author 周梦婷, corresponding authors 李建功 and 唐湘方. Funding came from the national pig industry technology system, the CAAS science and technology innovation programme, and central public-interest research institute funds. The tool is live at ai4as.cn and is to be integrated into the HERD platform — the national agricultural science data centre’s smart-livestock data-sharing service platform (国家农业科学数据中心智慧畜牧业数据共享服务平台).

The Chongqing backdrop. The standard’s home province runs the “pig industry brain 2.0 + future pig farm” (生猪产业大脑2.0 + 未来猪场) programme, presented by the municipal government in 2025 as a move toward unmanned pig farms with automatic precision feeding and robotic cleaning; and the Rongchang district launched China’s first livestock-industry technology trading market in November 2025 — a digital matchmaking platform with expert, output and demand databases. Livestock AI in China is being institutionalised provincially as much as nationally.

What this unit is doing in the taxonomy

The measurement-and-standards unit for Chinese livestock AI — the counterpart to what the corpus tracks in data-governance units elsewhere. It exists because the corpus’s Chinese livestock material is all production claims; this is the layer that decides whether those claims can be compared.

Distinct from:

• muyuan-smart-pig-farming.md and china-pig-farming-ecosystem-2018-2026.md — production and vendor layers; this is standards and research infrastructure.
• china-mara-agricultural-data-resources-2026.md — the ministry’s own data estate and statistics; this unit is the technical-standard and research-tool layer beneath it, and HERD is the domain instance of the national science data centre system.
• canadian-dairy-ai / milk-quality-standards-type units elsewhere in the corpus — comparable standardisation moves in other jurisdictions; China’s is happening at the same time and in the same direction.

Why it matters for talks

• Standards are the unglamorous half of AI adoption. A country that writes the liquid-feeding data standard first is doing the thing that makes vendor claims checkable later. Worth naming when the corpus’s other stories document claim inflation.
• The stated enemy is “data islands” — the same fragmentation problem the corpus records in the global data-infrastructure units, named here by a provincial extension station rather than a researcher.
• HABLer is the corpus’s best-verified Chinese livestock AI item. Peer-reviewed, with reported effect sizes and a released tool: V1 on verification, which almost nothing else in the corpus’s Chinese livestock material achieves.
• Extension stations write standards. The lead body is a technology extension station, which is notable for the corpus’s Extension thread: in China the extension system appears in standard-setting and platforms, not only in farmer advice.
• HERD is the state’s answer to shared livestock data — a national platform that researchers and industry users may join, with data-sharing framed as public infrastructure rather than as a market.

Critical context

• The standard’s scope is liquid feeding, not livestock AI. NY/T 5655—2026 concerns digital management systems for liquid-feeding — an automation-heavy segment of pig production, not models. Its inclusion here is honest only if labelled that way; it is the closest thing to a digital-management standard the sector has, not an AI standard.
• HABLer’s numbers are the authors’ own, though peer-reviewed and with code published. Effect sizes (0.87 accuracy, 60% time reduction) come from the paper’s own test scenarios — three named production settings — not from independent replication. V1, not V2.
• The standard’s adoption is unverified. Issuing a standard is not adopting it; no take-up data was located among Chinese farms or equipment vendors.
• “First” claims in Chinese standards reporting are narrow. “First agricultural industry standard in pig-farm liquid-feeding digital management” is a specific and defensible claim; it should not be paraphrased as “China’s first livestock AI standard”.
• The Chongqing “unmanned pig farm” language is promotional (from a government press conference), not a deployment report; no named farm or operational figure accompanies it.