--- title: "One AI model names 8,800 crop pests from a single field photo" date: 2026-10-05 category: AI in Action site: NeuroAI canonical: https://neuroai.site/a/na-app-topcloud-agri-pest-ai language: en --- # One AI model names 8,800 crop pests from a single field photo > Top Cloud-Agri's (托普云农) See-Insect (见虫) model IDs 8,800+ pests and its Wen Ji (问稷) agent advises on spraying, deployed across 30+ Chinese provinces. A county agronomist kneels in a rice paddy, holds a phone over a speckled leaf, and taps once. A second later the screen names the pest, grades how bad the infestation is, and suggests whether today is the day to spray. No microscope, no reference book, no waiting for the regional lab. That small moment is the front end of a much larger machine. The model behind it is See-Insect (见虫), built by Zhejiang-based Top Cloud-Agri (托普云农), and it is part of China's push to pour artificial intelligence into the least glamorous link in the food chain: spotting what is eating the crops before it is too late. ## A bug photo that answers back Top Cloud-Agri is not a drone company; it is an agriculture-information firm that has spent more than a decade wiring fields with sensors and cameras. Its See-Insect model is a computer-vision system trained on hundreds of terabytes of pest and disease data collected across crop life cycles. According to People's Daily, the model can identify more than 8,800 species of agricultural and forestry pests and over 70 types of crop-disease symptoms from a field image. The company reports its pest-and-disease recognition accuracy sits above 90%. In practice the tool is deployed several ways: fixed light- and colour-trap stations that photograph insects automatically, portable collectors for mobile checks, and lightweight mini-apps — "See-Bug" and "Count-Bug" — that let a farmer identify a catch on the spot. What makes it deployment, not a demo, is the closed loop. Traps feed images to a cloud platform that fuses the reads with weather and crop-growth data, then predicts where the next outbreak will land. A purpose-built agricultural agent called Wen Ji (问稷), built on retrieval-augmented generation, turns those signals into a plain-language pest report and a spray recommendation. ## From one photo to a whole farm's defense plan The "sky-ground" monitoring network ties single photos to a regional picture. Top Cloud-Agri runs intelligent insect-light traps and a "plant-protection online" platform that statistically analyses live pest pressure across an area and, using its own outbreak-prediction models, issues risk warnings to station staff. The same logic reaches the field itself through a digital smart field (数智大田) system that manages the full cycle — raise, till, plant, manage, harvest, dry. A documented pilot shows why farms care. At the Fengtian family farm in Jiangshan, close to 10,000 mu (about 670 hectares) of rice ran on this model, and People's Daily reports the results: water use fell about 10%, fertiliser use dropped 5%–10%, pest-control efficiency rose roughly 30%, management efficiency rose over 50%, and yield went up 6%–10%. The farm went on to set four Zhejiang provincial yield records in 2024. For a country whose leaders talk about "winning grain from the mouths of insects" (虫口夺粮), a 30% lift in pest-control efficiency is not a productivity footnote. It is the difference between a spray-everything habit and a spray-only-when-needed one. ## The company behind the model Top Cloud-Agri (托普云农) listed on the Shenzhen Stock Exchange in October 2024 (code 301556.SZ) and is, by its own description, a种植业农业-detection and agricultural-IoT firm. Its 2025 results put revenue at about 511 million yuan (≈ US$72M / HK$562M) with net profit near 106 million yuan. Its plant-protection products are listed as recommended gear in several provinces, and the company says its services cover more than 30 provincial-level regions in China. The firm also works with state bodies — the China National Agro-Tech Extension and Service Centre and provincial plant-protection stations among them — to field-test and standardise the equipment. That public-sector pull is what turns a model into infrastructure rather than a gadget a few rich farms buy. ## Why this belongs on a global reading list Food systems everywhere face the same squeeze: fewer young people want farm work, pests are migrating with the climate, and chemical overuse is exhausting the soil. China's answer at scale is a sensor-and-model layer that tells a human exactly when to act. The pattern — photograph, identify, predict, advise, keep the farmer's decision — is portable to rice bowls well beyond Zhejiang. The twist worth noting is that the AI here is mostly about restraint. By naming the pest precisely, it argues against blanket spraying, which is the opposite of the "more technology equals more inputs" story many readers expect. ## Honest limitations The 8,800-species count and the Fengtian farm yield figures come from People's Daily and the company's own materials; this article did not find an independent, replicated agronomy trial isolating the model's contribution from weather and husbandry. The 90% accuracy claim is vendor-reported and depends on image quality and pest mix. Listed-company financials are disclosed but unaudited at the summary level used here. The piece does not compare Top Cloud-Agri against overseas precision-agriculture vendors. ## What readers can do now - If you farm or advise farmers, start with one trap and one crop; the value shows up in spray-timing discipline, not in the photo alone. - Procurement teams should ask for the model's confusion matrix by pest class — a high overall accuracy can hide blind spots on the species that actually hurt your yield. - Watch input bills, not just yields: the real win from pest AI is fewer, better-timed sprays, and that is the metric to track quarter over quarter. --- Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-app-topcloud-agri-pest-ai Free to quote with attribution and a link to the original.