【529期】中国农业大学发布“神农·农业世界模型”

点击蓝字

关注置顶

内容共创计划，限30个名额！

Content Co-Creation Program, Limited to 30 spots! 

你出素材，我来发！

You provide materials, we publish the articles!

先到先得，快到评论区留言吧！

First come, first served. Leave your message in the comment section now!

国内农业 AI 研发再迎重大突破！中国农业大学全新迭代推出神农大模型 4.0，同步首发国内首款双域验证农业世界模型 1.0。新版模型大幅扩充作物数据体量，病虫害识别种类突破千种、识别精度超 93%，还新增作物基因组生成能力。依托因果推演、数字孪生仿真技术，可实现无田间试错的农业场景模拟推演，为育种科研、田间植保、抗旱稳产智能化决策提供全新技术支撑。

注：第一部分为英文文章及音频，第二部分为中英双语文章

（本文仅用于学习用途，For non-commercial educational use only）

英文文章及音频：（408 Words）

China Agricultural University Builds an Agricultural AI "Super Brain"

From: People.cn

During the 9th International Symposium（专题研讨会） on Innovative Development of Smart Agriculture held recently, a research team from China Agricultural University unveiled two agricultural artificial intelligence achievements consecutively: Shennong Large Model 4.0 and Shennong Agricultural World Model 1.0.

As the latest iteration（迭代版本） of the Shennong series, Shennong Large Model 4.0 centers on the agricultural AI infrastructure supporting "perception, cognition and reasoning". It delivers comprehensive upgrades in data coverage, pest and disease identification, genome analysis and service capacity, advancing agricultural artificial intelligence toward higher accuracy in recognition and richer professional knowledge.

Wang Yaojun, Associate Professor at the College of Information and Electrical Engineering, China Agricultural University, introduced the improvements in core capabilities. Shennong 4.0 has incorporated phenotypic（表型的） and water-fertilizer data for an additional 25 tropical（热带的） crops, further enriching the high-quality dataset for tropical agriculture and southern seed breeding scenarios. 

Its capacity to identify crop pests and diseases has expanded from 615 categories in the previous version to 1,016, with recognition accuracy exceeding 93%. Covering a broader spectrum（谱系） of crop varieties and disease types, the model offers more robust support for field diagnosis and plant protection decision-making. 

The team has also built generative capabilities tailored to agricultural crop genomics, providing new analytical and generative tools for breeding, trait dissection（性状解析） and genomic design.

Launched alongside the large model, Shennong Agricultural World Model 1.0 is the first agricultural world model validated across both protected agriculture and open farmland under action-conditioned frameworks.

Wang Yaojun explained that most existing agricultural AI systems excel at（擅长） recognition tasks and knowledge query. They can accurately identify disease types from images, count fruits, or offer professional answers and short-range reasoning for agricultural questions. 

By contrast, the Shennong Agricultural World Model constructs a causally structured, deducible（可推演的） digital agricultural world within the system. It functions like a digital twin experimental farm that simulates crop growth and environmental responses. Counterfactual reasoning（反事实推演） can be carried out without costly field trials. 

For instance, users can simulate how crop conditions evolve, and how yields and risks change if irrigation starts three days earlier and greenhouse temperature rises by two degrees — all questions can be answered via this world model.

Trials have been completed on three protected crops: tomatoes, lettuce and cucumbers. Action-conditioned validation focusing on drought scenarios has been conducted across 38 counties in Jilin Province. Results prove that the model can effectively deduce yield responses corresponding to different management strategies under drought conditions.

中英双语文章：

中国农大打造农业AI“超级大脑”

China Agricultural University Builds an Agricultural AI "Super Brain"

来源：人民网

在近日举行的第九届智慧农业创新发展国际研讨会期间，中国农业大学团队连续发布两项农业人工智能成果：神农大模型4.0与神农·农业世界模型1.0。

During the 9th International Symposium on Innovative Development of Smart Agriculture held recently, a research team from China Agricultural University unveiled two agricultural artificial intelligence achievements consecutively: Shennong Large Model 4.0 and Shennong Agricultural World Model 1.0.

神农大模型4.0是神农系列的最新迭代版本，聚焦农业AI“感知—认知—推理”底座，在数据覆盖、病虫害识别、基因组分析及服务能力等方面实现全面升级，推动农业人工智能向“看得更准、懂得更多”迈进。

As the latest iteration of the Shennong series, Shennong Large Model 4.0 centers on the agricultural AI infrastructure supporting "perception, cognition and reasoning". It delivers comprehensive upgrades in data coverage, pest and disease identification, genome analysis and service capacity, advancing agricultural artificial intelligence toward higher accuracy in recognition and richer professional knowledge.

中国农业大学信息与电气工程学院副教授王耀君介绍，在核心能力升级方面，神农4.0新增25种热带作物的表型与水肥数据，进一步丰富了热带与南繁场景的高质量数据底座。

Wang Yaojun, Associate Professor at the College of Information and Electrical Engineering, China Agricultural University, introduced the improvements in core capabilities. Shennong 4.0 has incorporated phenotypic and water-fertilizer data for an additional 25 tropical crops, further enriching the high-quality dataset for tropical agriculture and southern seed breeding scenarios. 

作物病虫害识别能力由上一代的615种扩展至1016种，识别准确率达93%以上，覆盖作物谱系和病害类型更加广泛，为田间诊断与植保决策提供更可靠支撑。

Its capacity to identify crop pests and diseases has expanded from 615 categories in the previous version to 1,016, with recognition accuracy exceeding 93%. Covering a broader spectrum of crop varieties and disease types, the model offers more robust support for field diagnosis and plant protection decision-making. 

团队还构建了面向农业作物基因组学的生成能力，为育种、性状解析与基因组设计提供新的分析与生成工具。

The team has also built generative capabilities tailored to agricultural crop genomics, providing new analytical and generative tools for breeding, trait dissection and genomic design.

同步发布的神农·农业世界模型1.0，则是首个动作条件化并在设施与大田双域验证的农业世界模型。

Launched alongside the large model, Shennong Agricultural World Model 1.0 is the first agricultural world model validated across both protected agriculture and open farmland under action-conditioned frameworks.

王耀君解释说，以往的大多数农业AI，主要擅长识别和专业知识库，可从图像中准确判别病害种类、统计果实数量，或是对农业问题给出专业解答和短程推理。

Wang Yaojun explained that most existing agricultural AI systems excel at recognition tasks and knowledge query. They can accurately identify disease types from images, count fruits, or offer professional answers and short-range reasoning for agricultural questions. 

神农·农业世界模型是在机器内部构建一个具备因果结构、可推演的农业世界，相当于为农田建立一块会生长、会响应的数字孪生试验田，无需实地试错，即可开展反事实推演。

By contrast, the Shennong Agricultural World Model constructs a causally structured, deducible digital agricultural world within the system. It functions like a digital twin experimental farm that simulates crop growth and environmental responses. Counterfactual reasoning can be carried out without costly field trials. 

比如提前三天灌溉并将棚温调高两度，作物状态将如何逐步演化，产量与风险将随之如何变化，都能在神农·农业世界模型中找到答案。

For instance, users can simulate how crop conditions evolve, and how yields and risks change if irrigation starts three days earlier and greenhouse temperature rises by two degrees — all questions can be answered via this world model.

据介绍，该模型已在番茄、生菜、黄瓜三种设施作物上跑通，在吉林省38个县开展了以干旱为切入的动作条件化验证，结果表明模型能够有效推演旱情情景下不同管理方案的产量响应。

Trials have been completed on three protected crops: tomatoes, lettuce and cucumbers. Action-conditioned validation focusing on drought scenarios has been conducted across 38 counties in Jilin Province. Results prove that the model can effectively deduce yield responses corresponding to different management strategies under drought conditions.

- 词汇盘点 -

symposium、iteration、phenotypic、tropical、spectrum、trait dissection、excel at、deducible、counterfactual reasoning

- END -

- 推荐阅读 -

-Recommended reading-