[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"item-3596":3,"related-3596":38},{"id":4,"title":5,"url":6,"summary":7,"summary_zh":8,"content":9,"source_name":10,"source_url":8,"published_at":11,"category":12,"cover_url":8,"hotness":13,"is_selected":14,"score":15,"score_detail":16,"sources":24,"tags":26,"search_phrases":32,"slug":35,"view_count":36,"doi":8,"paper":8,"created_at":37},3596,"中国农科院基因组所商连光团队开发自进化AI科学家'绿洲'助力作物耐盐碱研究——Molecular Plant在线发表","https:\u002F\u002Fagis.caas.cn\u002Fxwzx\u002Fkyjz\u002Fb0960d967c964346b50cb50c4886c209.htm","2026年9月18日，中国农科院基因组所（大鹏湾实验室）商连光团队联合崖州湾国家实验室、盐碱地综合利用国家技术创新中心，在《分子植物（Molecular Plant）》上在线发表题为'OASIS, a self-evolving AI scientist that integrates omics data and literature knowledge for plant stress research'的研究论文。该研究开发了面向植物逆境研究的自进化AI科学家'绿洲'（OASIS）。系统将大语言模型、多智能体协作、植物逆境文献知识库以及组学和遗传数据纳入同一科研流程。'绿洲'将复杂科研任务拆解为多个子任务，由6个AI智能体分工协作完成。研究团队进一步以水稻耐盐为案例，对'绿洲'提出的候选基因开展实验验证，形成了从科学问题到候选假说再到实验验证的完整研究路径。",null,"2026年9月18日，中国农科院基因组所（大鹏湾实验室）商连光团队联合崖州湾国家实验室、盐碱地综合利用国家技术创新中心，在《分子植物（Molecular Plant）》上在线发表题为“OASIS, a self-evolving AI scientist that integrates omics data and literature knowledge for plant stress research”的研究论文。\n\n![Image 1](https:\u002F\u002Fagis.caas.cn\u002Fimages\u002F2026-09\u002Fefa2de3e737741ea8e2ae4ceeed0695f.png)\n\n该研究开发了面向植物逆境研究的自进化AI科学家“绿洲”（OASIS）。系统将大语言模型、多智能体协作、植物逆境文献知识库以及组学和遗传数据纳入同一科研流程，可围绕一个科学问题自主完成任务规划、文献检索、数据查询、文献与组学数据融合推理和结果审查，并能够从历史任务中的失败与纠错中学习可复用经验。研究团队进一步以水稻耐盐为案例，对“绿洲”提出的候选基因开展实验验证，形成了从科学问题到候选假说再到实验验证的完整研究路径。\n\n植物抗逆研究往往不是“缺少信息”，而是证据分散在不同载体中：机制研究主要沉淀在论文里，基因表达、遗传定位、群体变异和种质资源则分布在多个数据库中。研究人员通常需要在大量文献和多类数据之间反复检索、比对和判断，才能把一个科学问题逐步收敛为可供实验验证的假说。通用大语言模型擅长理解和生成文本，但仅依靠模型自身知识，仍难以稳定完成跨数据库、跨数据类型的连续科研分析。\n\n**从“读文献”到“查数据”，让AI进入真实科研工作流**\n\n针对这一问题，“绿洲”将复杂科研任务拆解为多个子任务，由6个AI智能体分工协作完成。系统先识别研究意图并制定任务计划，再根据需要调用文献检索、数据分析和科研数据库等工具，最终由审查与综合模块检查证据覆盖范围和任务完成情况后生成结论。任务规划、文献片段、数据库查询结果和分析记录都会保存在共享工作空间中，便于研究人员回溯和核对。整个过程类似于一个小型科研团队围绕同一科学问题持续协作。\n\n![Image 2](https:\u002F\u002Fagis.caas.cn\u002Fimages\u002F2026-09\u002Ff4ecb718f8d942b89d22b99880d0dbe3.png)\n\n图1 | “绿洲”多智能体协作科研框架。系统围绕科学问题进行任务拆解，并协调文献检索、数据分析、结果审查和综合生成，使不同来源的科研证据能够在同一工作流中被连续调用和融合推理。\n\n支撑这一流程的是专门整理的植物逆境知识和数据资源。“绿洲”目前覆盖拟南芥、水稻、玉米、小麦、大豆和陆地棉6个物种，以及盐、干旱、高温和低温4类主要非生物胁迫。文献知识库收录超过15万篇相关论文摘要和1.5万篇开放获取全文，并持续同步新发表研究；同时接入植物逆境表达数据、GWAS\u002FQTL、群体变异、eQTL和种质资源等结构化信息，使系统不仅能够回答“文献中有哪些报道”，还能够进一步判断“数据是否支持这一推断”。\n\n**OASIS的优势，体现在数据驱动的研究任务中**\n\n为评估系统在真实科研任务中的表现，研究团队构建了面向植物逆境研究的基准测试集PlantStressQA，包含200个贴近真实科研过程的问题，覆盖机制证据、基因ID映射、表达分析、遗传位点分析、品种筛选和群体变异六类任务。\n\n在与多种通用大语言模型的比较中，“绿洲”总体得分为84.0分，参评通用模型得分为33.8至50.1分。在需要实际调用并整合结构化数据的任务中，“绿洲”优势明显，其中群体变异和遗传位点分析两类任务分别比表现最好的通用模型高55.6分和52.3分，在基因ID映射和表达证据分析中也表现出明显优势。\n\n![Image 3](https:\u002F\u002Fagis.caas.cn\u002Fimages\u002F2026-09\u002Fc2b2782477a64a55a3366b44219320d5.png)\n\n图2 | “绿洲”在PlantStressQA上的系统评价。系统在基因ID转换、表达数据查询、遗传位点分析和群体变异分析等数据密集型任务中表现出更明显的优势。\n\n这些结果说明，“绿洲”的核心并不是单纯依赖语言模型“记住更多知识”，而是通过组织科研工具和数据资源，在研究过程中主动查找数据、完成格式转换、比较不同来源的证据，并将这些步骤串联为可追溯的分析过程。\n\n**做错一次之后，下一次还会不会犯同样的错误？**\n\n真实科研分析中，数据库字段不匹配、基因ID格式错误、工具调用顺序不合理等问题并不少见。如果系统每次遇到相似任务都从头试错，不仅效率低，也会造成额外的计算资源消耗。为此，“绿洲”设计了自进化经验学习模块（SEEL），从历史任务中识别“先失败、随后被成功纠正”的执行轨迹，对比错误路径和正确路径，提炼可复用的操作经验，并通过重新执行真实任务检验其有效性。只有验证通过的经验才会进入技能库，供后续相似任务调用。\n\n![Image 4](https:\u002F\u002Fagis.caas.cn\u002Fimages\u002F2026-09\u002F72cd8d054dc14f94b7b9461f4a7f8c46.png)\n\n图3 | SEEL自进化经验学习流程及效果。系统从历史任务的失败与纠错过程中提炼可复用经验，经验证后纳入技能库。\n\n研究团队从PlantStressQA的历史轨迹中筛选出86个包含可复用经验的任务。应用这些经验后，工具调用失败、数据库错误、重复操作和模型处理文本量均有所减少。在一个表达分析任务中，大模型调用次数从75次降至42次，模型处理文本量从112万降至31万，运行时间从476秒缩短到236秒，数据库错误和失败工具调用则从多次降为0次。随着有效经验持续积累，“绿洲”能够在后续相似任务中减少重复试错并优化执行过程。\n\n**最关键的一步：AI提出的候选基因，实验能不能验证？**\n\n基准测试可以衡量AI系统是否“会做题”，但科研价值最终还要回到能否帮助提出值得验证的科学假说。研究团队以水稻耐盐为案例，在未预设具体候选基因的情况下，通过多轮提问引导“绿洲”开展分析。系统先从已有研究中归纳离子稳态与Na⁺\u002FK⁺运输、ABA信号与渗透调节、ROS稳态、MAPK级联和盐感知\u002FCa²⁺信号等主要耐盐调控模块，并整理出284个有文献证据支持的拟南芥调控基因；随后开展跨物种同源映射和功能筛选，再整合水稻耐盐GWAS\u002FQTL等遗传证据，最终逐步确定5个优先候选基因。\n\n![Image 5](https:\u002F\u002Fagis.caas.cn\u002Fimages\u002F2026-09\u002Fa9ba4a0ff48e43bc8239834881fb034c.png)\n\n图4 | “绿洲”辅助筛选水稻耐盐候选基因的多轮对话示例。系统整合耐盐机制文献、水稻同源基因信息和群体遗传证据，逐步筛选出5个优先候选基因。\n\n研究团队根据“绿洲”给出的证据链，选择此前未被直接报道参与水稻耐盐调控的OsPP2a（LOC_Os12g04290）开展实验验证。该基因的拟南芥同源基因TAP46与ABA信号相关，OsPP2a本身位于水稻耐盐meta-QTL区间内，并与一个eQTL信号共定位。两个独立的Ospp2a突变材料在150 mM NaCl处理后均表现出更明显的盐敏感表型，恢复存活率降低，同时Na⁺积累增加、Na⁺\u002FK⁺平衡发生改变。RT-qPCR结果还显示，离子稳态相关基因OsSOS1和OsHKT1;5的表达发生变化，为OsPP2a参与水稻盐胁迫响应提供了遗传和分子层面的证据。\n\n![Image 6](https:\u002F\u002Fagis.caas.cn\u002Fimages\u002F2026-09\u002Fe698f656e5c8488fb80baadceaab1bf0.png)\n\n图5 | “绿洲”优先推荐基因OsPP2a的耐盐功能验证。两个独立突变材料在盐胁迫下表现出更低的恢复存活率和更高的Na⁺\u002FK⁺比值，支持OsPP2a参与水稻耐盐响应。\n\n在这一案例中，研究人员通过多轮提问引导“绿洲”完成机制梳理、跨物种证据迁移、遗传数据查询和候选基因优先级排序，再由实验进一步检验候选假说，形成了“科学问题→证据融合→候选假说→实验验证”相衔接的研究路径。\n\n**从“AI回答科学”到“AI参与科学”**\n\n“绿洲”的目标并不是用AI替代科研人员作出判断，而是把过去需要在多篇文献、多个数据库和不同分析工具之间往返完成的工作，组织成一条可追溯、可复核的科研流程。通过多智能体协作、文献与组学数据融合推理以及自进化经验学习，系统能够帮助研究人员更高效地汇集证据、缩小候选范围并提出更有依据的科学假说。\n\n这项工作为人工智能参与植物抗逆研究提供了一个从“回答问题”走向“参与分析”的实践框架。未来，随着文献、组学数据和科研工具持续扩展，“绿洲”有望进一步服务于作物抗逆机制解析、关键基因挖掘和后续实验研究，为耐盐碱等作物抗逆改良提供智能化研究工具。\n\n基因组所（大鹏湾实验室）商连光研究员和崖州湾国家实验室钱前院士为论文共同通讯作者，基因组所（大鹏湾实验室）博士后韩漾和副研究员魏华为论文共同第一作者。该研究得到国家重点研发计划和国家自然科学基金的资助。\n\n平台入口：[https:\u002F\u002Fwww.oasis.ac.cn](https:\u002F\u002Fwww.oasis.ac.cn\u002F)\n\n原文链接：[https:\u002F\u002Fwww.cell.com\u002Fmolecular-plant\u002Fabstract\u002FS1674-2052(26)00305-9](https:\u002F\u002Fwww.cell.com\u002Fmolecular-plant\u002Fabstract\u002FS1674-2052(26)00305-9)","中国农科院基因组研究所 2026-09-18","2026-09-18T00:00:00Z","报道",10,true,90,{"impact":17,"substance":18,"depth":19,"authority":20,"freshness":21,"relevant":22,"comment":23},26,24,19,15,6,1,"国家级科研机构在《分子植物》发表的自进化AI科学家系统，实现文献与组学数据融合推理并经水稻耐盐基因实验验证，属农业人工智能与智慧育种领域的标志性突破。",[25],{"name":10,"url":6},[27,28,29,30,31],"农业人工智能","水稻","多智能体","耐盐碱","智慧育种",[33,34],"中国农科院基因组所 绿洲 OASIS","商连光 水稻耐盐 OsPP2a","中国农科院基因组所绿洲OASIS-3596",0,"2026-09-27T00:05:16.998310Z",{"total":21,"page":22,"page_size":21,"items":39},[40,115,152,183,205,243],{"id":41,"title":42,"url":43,"summary":44,"summary_zh":45,"content":8,"source_name":46,"source_url":43,"published_at":47,"category":48,"cover_url":8,"hotness":13,"is_selected":49,"score":50,"score_detail":51,"sources":58,"tags":60,"search_phrases":64,"slug":67,"view_count":36,"doi":68,"paper":69,"created_at":114},3490,"Balancing accuracy, completeness, and efficiency for rice 3D reconstruction through CBAM-UNet-based multi-view segmentation and camera configuration optimization","https:\u002F\u002Fdoi.org\u002F10.3389\u002Ffpls.2026.1882534","Multi-view 3D reconstruction has been widely applied in plant phenotyping, but the complex canopy structure of rice plants poses significant challenges for reconstruction accuracy, completene7ss, and efficiency. In this study, we developed an optimized workflow for 3D reconstruction of rice using the self-developed Metatlas V1 multi-view imaging platform combined with the COLMAP + OpenMVS pipeline. A CBAM-UNet-based image segmentation model was used to extract plant regions from complex backgrounds, outperforming thresholding, U-Net, and U-Net++, and increasing the number of reconstructed points. Camera configuration optimization was performed by systematically combining equidistant and greedy strategies, resulting in an optimal six-camera setup (#2, #3, #4, #6, #7, and #8, corresponding to −30°, −15°, 0°, +30°, +45°, and +60° relative to the horizontal viewpoint), which provided complementary views of the canopy inner structure and stem base. This configuration reduced the average nearest-neighbor distance to 0.037–0.066 cm across different varieties and growth stages compared with the full 11-camera setup, retained 73–76% of the points, and decreased reconstruction duration by approximately 70–80%. The strong agreement between point-cloud-derived and manually measured plant height and canopy width ( R 2 = 0.989 and 0.946, respectively) supported the accuracy of the point-cloud-derived phenotypic measurements. Overall, integrating image segmentation with camera-configuration optimization offers an accurate and efficient solution for high-throughput 3D phenotyping of rice using Metatlas V1, balancing reconstruction accuracy, point-cloud completeness, and computational efficiency, and may provide a methodological reference for camera-configuration design in other rotational multi-view phenotyping platforms.","多视角三维重建已广泛应用于植物表型分析，但水稻复杂的冠层结构对重建精度、完整性和效率构成了重大挑战。本研究利用自主研发的Metatlas V1多视角成像平台，结合COLMAP + OpenMVS流程，开发了一套优化的水稻三维重建工作流。采用基于CBAM-UNet的图像分割模型从复杂背景中提取植株区域，其性能优于阈值分割、U-Net和U-Net++，并增加了重建点数量。通过系统组合等距策略和贪心策略进行相机配置优化，得到了最优的六相机方案（#2、#3、#4、#6、#7和#8，分别对应相对于水平视角的−30°、−15°、0°、+30°、+45°和+60°），该方案提供了冠层内部结构和茎基部的互补视角。与完整的11相机方案相比，该配置将不同品种和生育期的平均最近邻距离降至0.037–0.066 cm，保留了73–76%的点云，并将重建时间缩短了约70–80%。点云提取的株高和冠幅与人工测量结果高度一致（R²分别为0.989和0.946），验证了点云表型测量的准确性。总体而言，将图像分割与相机配置优化相结合，为利用Metatlas V1进行水稻高通量三维表型分析提供了一种准确高效的解决方案，在重建精度、点云完整性和计算效率之间取得了平衡，并可为其他旋转式多视角表型平台的相机配置设计提供方法学参考。","Frontiers in Plant Science","2026-09-24T00:00:00Z","论文",false,78,{"impact":52,"substance":53,"depth":54,"authority":55,"freshness":56,"relevant":22,"comment":57},16,22,18,13,9,"该研究提出结合CBAM-UNet分割与相机配置优化的水稻三维重建流程，方法新颖、数据详实，对高通量作物表型分析有参考价值，但属细分领域进展，时效性高。",[59],{"name":46,"url":43},[61,27,28,62,63],"智慧农业","作物表型","三维重建",[65,66],"Metatlas V1 水稻 三维重建","CBAM-UNet 水稻 图像分割","MetatlasV1水稻三维重建-3490","10.3389\u002Ffpls.2026.1882534",{"doi":68,"openalex_id":70,"authors":71,"venue":46,"cited_by_count":36,"oa_url":43,"card":107,"direction":111,"ingested_from":113},"W7214231705",[72,74,77,79,82,84,86,89,92,95,97,99,102,105],{"name":73,"orcid":8},"Haoyang Zhou",{"name":75,"orcid":76},"Rongjie Chen","https:\u002F\u002Forcid.org\u002F0009-0004-8078-2301",{"name":78,"orcid":8},"Hao Wang",{"name":80,"orcid":81},"Minglu Li","https:\u002F\u002Forcid.org\u002F0000-0003-1751-9418",{"name":83,"orcid":8},"Yongkang Teng",{"name":85,"orcid":8},"Shenghao Ye",{"name":87,"orcid":88},"Menglei Wei","https:\u002F\u002Forcid.org\u002F0009-0004-3070-8023",{"name":90,"orcid":91},"Kun Yu","https:\u002F\u002Forcid.org\u002F0000-0002-0190-8702",{"name":93,"orcid":94},"Pingping Fang","https:\u002F\u002Forcid.org\u002F0000-0002-9911-2963",{"name":96,"orcid":8},"Jing Cao",{"name":98,"orcid":8},"Fanglin Zhu",{"name":100,"orcid":101},"Wenyu Zhang","https:\u002F\u002Forcid.org\u002F0000-0003-3322-9736",{"name":103,"orcid":104},"Ting Sun","https:\u002F\u002Forcid.org\u002F0000-0002-9387-4852",{"name":106,"orcid":8},"Min Jiang",{"tldr":108,"method":109,"finding":110,"direction":111,"opportunity":112},"提出CBAM-UNet分割与相机配置优化结合的水稻多视角三维重建流程，兼顾精度、完整性与效率。","Metatlas V1多视角平台、COLMAP+OpenMVS、CBAM-UNe","最优6相机配置将最近邻距离降至0.037–0.066 cm，保留73–76%点云，重建耗时减少约70","农业遥感与作物表型","可将相机配置优化策略迁移到其他旋转多视角平台，并探索自适应配置与分割模型联合优化。","openalex","2026-09-25T23:30:25.303016Z",{"id":116,"title":117,"url":118,"summary":119,"summary_zh":120,"content":8,"source_name":121,"source_url":118,"published_at":122,"category":48,"cover_url":8,"hotness":13,"is_selected":49,"score":123,"score_detail":124,"sources":126,"tags":128,"search_phrases":130,"slug":133,"view_count":36,"doi":134,"paper":135,"created_at":151},3164,"Spatiotemporal Deep Learning for Rice Plant Height Estimation from Multi-Temporal UAV RGB Imagery","https:\u002F\u002Fdoi.org\u002F10.3390\u002Fagriculture16182034","Accurate plant height estimation is important for monitoring crop growth and supporting precision agricultural management. Manual measurements are labor-intensive, while LiDAR-based methods are expensive and require complex processing. UAV photogrammetry provides a lower-cost alternative but remains challenging in flooded rice paddies because of canopy deformation and difficulties in terrain extraction. This study proposes Rice-STNet, a spatiotemporal deep learning framework for end-to-end rice plant height estimation using multi-temporal UAV RGB imagery. Rice-STNet integrates a convolutional neural network for spatial feature extraction, Time2Vec for temporal encoding, and a gated recurrent unit network for modeling temporal dependencies across observation dates. The framework was evaluated using field data collected from rice paddies over two growing seasons. Rice-STNet achieved an R2 of 0.97, a root mean squared error of 1.97 cm, and a mean absolute error of 1.14 cm. It outperformed random forest, support vector regression, a CNN-only baseline, and a UAV photogrammetry-based point-cloud approach. In addition, the framework generated high-resolution plant height maps for field-scale analysis of spatial growth variability. These results underscore the importance of jointly modeling spatial and temporal characteristics for continuously evolving crop traits. The proposed framework offers an accurate, scalable, and non-destructive solution for large-scale crop phenotyping and precision agriculture.","准确的株高估算对于监测作物生长和支持精准农业管理具有重要意义。人工测量劳动强度大，而基于激光雷达（LiDAR）的方法成本高昂且需要复杂的处理。无人机摄影测量提供了一种成本较低的替代方案，但在淹水稻田中仍面临挑战，原因在于冠层变形和地形提取困难。本研究提出了Rice-STNet，一种时空深度学习框架，用于利用多时相无人机RGB影像进行端到端水稻株高估算。Rice-STNet集成了用于空间特征提取的卷积神经网络、用于时间编码的Time2Vec，以及用于建模观测日期之间时间依赖关系的门控循环单元网络。该框架利用两个生长季从稻田采集的田间数据进行了评估。Rice-STNet取得了R²为0.97、均方根误差为1.97 cm、平均绝对误差为1.14 cm的结果。其性能优于随机森林、支持向量回归、仅使用CNN的基线方法以及基于无人机摄影测量的点云方法。此外，该框架生成了高分辨率株高图，用于田块尺度空间生长变异性分析。这些结果凸显了联合建模空间与时间特征对于持续变化的作物性状的重要性。所提出的框架为大规模作物表型分析和精准农业提供了一种准确、可扩展且非破坏性的解决方案。","Agriculture","2026-09-21T00:00:00Z",80,{"impact":54,"substance":53,"depth":54,"authority":55,"freshness":56,"relevant":22,"comment":125},"提出时空深度学习框架Rice-STNet，用多时相无人机RGB影像实现水稻株高高精度估算，方法新颖、数据跨两个生长季，对作物表型与精准农业有实用价值。",[127],{"name":121,"url":118},[61,27,28,129,62],"遥感",[131,132],"无人机 RGB 水稻株高","Rice-STNet 水稻表型","无人机RGB水稻株高-3164","10.3390\u002Fagriculture16182034",{"doi":134,"openalex_id":136,"authors":137,"venue":121,"cited_by_count":36,"oa_url":118,"card":146,"direction":111,"ingested_from":113},"W7213887432",[138,141,143],{"name":139,"orcid":140},"Weiguo Wang","https:\u002F\u002Forcid.org\u002F0009-0003-4028-9363",{"name":142,"orcid":8},"Noboru Noguchi",{"name":144,"orcid":145},"Liangliang Yang","https:\u002F\u002Forcid.org\u002F0000-0002-5055-3987",{"tldr":147,"method":148,"finding":149,"direction":111,"opportunity":150},"提出Rice-STNet时空深度学习框架，用多时相无人机RGB影像估算水稻株高。","CNN提取空间特征，Time2Vec编码时间，GRU建模时序依赖，两季稻田数据验","R²达0.97、RMSE 1.97cm，优于随机森林、SVR、纯CNN及点云方法。","可迁移至其他作物与多源遥感融合，探索轻量化模型及实时田间部署。","2026-09-22T23:30:18.545958Z",{"id":153,"title":154,"url":155,"summary":156,"summary_zh":8,"content":8,"source_name":157,"source_url":8,"published_at":11,"category":48,"cover_url":8,"hotness":13,"is_selected":14,"score":158,"score_detail":159,"sources":163,"tags":165,"search_phrases":169,"slug":172,"view_count":36,"doi":8,"paper":173,"created_at":182},3125,"Full-Season Agentic Farm System FAIRY: Event-Driven Multi-Agent Orchestration for Soybean Production（FAIRY 全季节智能体农场系统：大豆生产的事件驱动多智能体编排）","https:\u002F\u002Faiagentstore.ai\u002Fai-agent-news\u002Ftopic\u002Fagriculture-food\u002F2026-08-11","哈尔滨工业大学研究人员发布并部署全栈、事件驱动的智能体引擎 FAIRY 于中国运行中的大豆研究农场。FAIRY 集成传感器、无人机、卫星植被产品、机械 API、作物过程模型与多智能体编排层，执行起垄→播种→灌溉→病虫害防治→收获→干燥→存储工作流，并在 64 垄研究场上跨 100 个全季节场景评估 9 个智能体控制器。这是智能体系统能够在大农业时间尺度和延迟结果下进行推理的最清晰演示之一，将农业中的智能体工作从实验室演示推进到全过程评估。同期 arXiv 推出 HarvestBench 基准将 LLM 驱动智能体置于农场网格世界（拖拉机面临动物选择绕行或碾压），结果显示模型差异巨大、对道德简报高度敏感、避免意愿具有可衡量的价格弹性。","Harbin Institute of Technology \u002F arXiv",89,{"impact":18,"substance":160,"depth":19,"authority":161,"freshness":56,"relevant":22,"comment":162},23,14,"哈工大在真实大豆农场部署全季节事件驱动多智能体系统并配套 HarvestBench 基准，是农业智能体从演示走向全过程评估的标志性进展，专业深度与信息增量俱佳。",[164],{"name":157,"url":155},[61,166,27,167,29,168],"无人农场","农业遥感","大豆生产",[170,171],"哈工大 FAIRY 大豆","HarvestBench 智能体 农场","哈工大FAIRY大豆-3125",{"doi":8,"openalex_id":8,"authors":174,"venue":8,"cited_by_count":36,"oa_url":8,"card":175,"direction":179,"ingested_from":181},[],{"tldr":176,"method":177,"finding":178,"direction":179,"opportunity":180},"部署全季节事件驱动多智能体系统FAIRY，在大豆农场完成从起垄到存储的全流程评估。","集成传感器、无人机、卫星、作物模型与机械API，用9个智能体控制器在64垄100","智能体系统能在大农业时间尺度下推理，模型差异大且对道德简报敏感。","农业人工智能与决策模型","可研究多智能体在长周期、延迟反馈农业任务中的鲁棒性与伦理约束机制。","agent","2026-09-22T00:05:38.611717Z",{"id":184,"title":185,"url":186,"summary":187,"summary_zh":8,"content":188,"source_name":189,"source_url":8,"published_at":190,"category":12,"cover_url":8,"hotness":13,"is_selected":49,"score":191,"score_detail":192,"sources":195,"tags":197,"search_phrases":200,"slug":203,"view_count":36,"doi":8,"paper":8,"created_at":204},3101,"AI 接管稻田：四川眉山永丰村 300 亩 AI 试点水稻亩产 826.8-863.6 公斤","https:\u002F\u002Fai-damn.com\u002Fai-takes-over-the-rice-fields-863-6-kg-per-mu-in-sichuan-pilot-1789513373662","9-14 四川省眉山市东坡区太和镇永丰村千亩高标准农田 300 亩 AI 试点田通过专家组测产验收：\"华浙优 210\"（高产优质杂交稻）亩产 826.8 公斤、\"胜两优 222\"（超高产籼粳杂交稻）亩产 863.6 公斤、\"全优 169\"（超高产杂交籼稻）亩产 858.8 公斤。AI 系统通过无人机巡检采集数据，对种植、水肥调控和病虫害早期预警提供精准建议。四川农业大学水稻栽培专家马均教授表示，结合良种、良法与 AI 精准管理可有效释放水稻增产潜力，为大规模单产提升提供可复制技术路径；今年永丰村共有 240 余个新品种在产量\u002F株型\u002F米质上表现良好，智能精准播种技术与 AI 应用已初步见效。","## AI Takes Over the Rice Fields: 863.6 kg per Mu in Sichuan Pilot\n\nIn the rolling fields of Yongfeng Village, Tahe Town, Dongpo District, Meishan City, Sichuan Province, something unusual happened this harvest season. On September 14, as combines rolled through the thousand-mu high-standard farmland, 300 mu of it had been managed not by traditional farming wisdom alone, but by an **AI model** specifically designed for rice cultivation.\n\nGone are the days of \"judging fields by experience.\" Now, it's all about **making decisions based on data**.\n\n### From Experience to Data\n\nThe embankments were crowded with agricultural experts and curious farmers, all gathered to witness a field test. An expert group organized by the Sichuan Provincial Science and Technology Department was evaluating a project led by Sichuan Agricultural University: the \"Integrated Demonstration and Application of High-quality, High-yield, and Efficient Production Technologies for Rice-Vegetable (Medicinal) Crops in the Chengdu Plain.\"\n\nSo how does it work? The AI system collects data through **drone inspections**, then provides precise recommendations on planting, water and fertilizer regulation, and pest and disease early warning.\n\nLocal large-scale grain farmer Zhao Youyong put it simply: \"Before, farming relied on experience for field inspections. Now, using drones and the AI system, we get timely information about pests and diseases, so we can handle them directly. Farming has become more convenient.\"\n\n### The Numbers That Matter\n\nThe expert group's standardized yield test delivered solid results. All three core varieties in the 300-mu AI pilot fields performed impressively:\n\n*   **\"Huazheyous 210\"** (high-yield, high-quality hybrid rice): 826.8 kg per mu\n*   **\"Shengliangyou 222\"** (super-high-yield indica-japonica hybrid rice): **863.6 kg per mu**\n*   **\"Quanyou 169\"** (super-high-yield hybrid indica rice): 858.8 kg per mu\n\nMa Jun, a rice cultivation expert at Sichuan Agricultural University, explained that these yields prove that combining quality seeds with appropriate methods and AI precision management can **effectively release the potential for rice yield increase**. It offers a replicable technical path for large-scale yield improvement.\n\nHe also noted that more than 240 new varieties demonstrated good performance in yield, plant shape, and rice quality in Yongfeng Village this year. The application of intelligent precision sowing technology and AI has already shown initial results.\n\n### What This Means for the Future\n\nThis pilot isn't just about one good harvest. It's a glimpse into how **AI can transform traditional agriculture**. By moving from experience-based to data-driven farming, growers can make more informed decisions, reduce risks, and potentially achieve higher yields sustainably.\n\nAs Ma Jun pointed out, the combination of quality seeds, appropriate methods, and AI precision management provides a technical path that can be replicated on a larger scale. For a country that feeds 20% of the world's population with less than 10% of its arable land, such innovations are more than welcome—they're essential.\n\n### Key Points\n\n*   **AI-managed pilot field** in Sichuan achieved rice yields up to **863.6 kg per mu**.\n*   **Drones and data** replaced traditional experience-based farming for planting, fertilization, and pest control.\n*   **Three rice varieties** all exceeded 826 kg per mu, proving the effectiveness of AI precision management.\n*   **Experts say** this approach offers a replicable path for large-scale yield improvement.\n*   **The future of farming** is shifting from \"judging fields by experience\" to \"making decisions based on data.\"","AI DAMN","2026-09-14T10:00:00Z",76,{"impact":53,"substance":53,"depth":193,"authority":56,"freshness":21,"relevant":22,"comment":194},17,"AI精准管理水稻试点实测亩产数据具体、多方信源，具备可复制的智慧农业示范价值，值得入选每日精选。",[196],{"name":189,"url":186},[61,27,28,198,199],"精准农业","无人機巡田",[201,202],"四川眉山 永丰村 AI水稻","四川农业大学 水稻 AI试点","四川眉山永丰村AI水稻-3101","2026-09-22T00:05:33.996455Z",{"id":206,"title":207,"url":208,"summary":209,"summary_zh":210,"content":8,"source_name":211,"source_url":208,"published_at":212,"category":48,"cover_url":8,"hotness":13,"is_selected":49,"score":213,"score_detail":214,"sources":218,"tags":220,"search_phrases":222,"slug":225,"view_count":36,"doi":226,"paper":227,"created_at":242},3009,"AgriMAC: An Attention Based Multimodal Deep Clustering Framework for Rice Health Assessment","https:\u002F\u002Fdoi.org\u002F10.22266\u002Fijies2026.1031.06","Rice is Indonesia's staple crop, yet its productivity has declined in recent years because pest and disease outbreaks remain difficult to detect at an early stage.Existing precision agriculture approaches commonly process Internet of Things (IoT) sensor data and remote sensing imagery independently and often rely on supervised learning, requiring large amounts of labeled data.Meanwhile, multispectral drone imagery producing the Normalized Difference Vegetation Index (NDVI) provides richer information on crop physiological conditions than RGB-based vegetation indices.This study proposes Agricultural Multimodal Attention Clustering (AgriMAC), an unsupervised framework that integrates UAV derived NDVI imagery, 7-in-1 IoT soil sensor measurements, and historical weather data from the Open-Meteo API for rice field condition monitoring.Each modality is encoded using a dedicated autoencoder and fused through an entropy-regularized attention mechanism before Deep Embedded Clustering is performed.To reduce the influence of crop growth stage, the IoT representation is residualized using growth-phase statistics estimated exclusively from the training fold, enabling the discovered clusters to represent within-phase agronomic conditions rather than crop age.Experiments conducted under a grouped leave-one-field-out protocol produced a Silhouette Score of 0.465 ± 0.048, a Davies Bouldin Index of 0.807 ± 0.036, and a Calinski Harabasz Index of 195 ± 27.The learned groups also showed low normalized mutual information with growth phase (0.079 ± 0.043) and near chance phase decodability (balanced accuracy = 0.554 ± 0.042), indicating that they are only weakly associated with crop growth stage.The learned attention weights identified IoT soil measurements (0.570 ± 0.024) as the dominant modality, while NDVI imagery (0.210 ± 0.014) and weather information (0.220 ± 0.014) provided complementary spatial and temporal context.Overall, AgriMAC provides an interpretable and leakage-aware framework for multimodal clustering of rice field conditions.Although its clustering performance is comparable to that of a capacity-matched IoT-only model, it additionally quantifies the contribution of each sensing modality through attention weights and explicitly mitigates the growth-phase confound, making it suitable for field level agronomic condition monitoring and spatial decision support in precision agriculture.","水稻是印度尼西亚的主要作物，但近年来其生产力有所下降，因为病虫害暴发在早期阶段仍难以检测。现有的精准农业方法通常独立处理物联网（IoT）传感器数据和遥感影像，且往往依赖监督学习，需要大量标注数据。与此同时，生成归一化植被指数（NDVI）的多光谱无人机影像比基于RGB的植被指数能提供更丰富的作物生理状况信息。本研究提出农业多模态注意力聚类（AgriMAC），这是一个无监督框架，整合了无人机获取的NDVI影像、七合一IoT土壤传感器测量数据以及来自Open-Meteo API的历史天气数据，用于稻田状况监测。每种模态均使用专用自编码器进行编码，并通过熵正则化注意力机制进行融合，随后执行深度嵌入聚类。为减少作物生长阶段的影响，IoT表征利用仅从训练折估计的生长阶段统计量进行残差化处理，使发现的聚类能够表征阶段内的农艺状况而非作物年龄。在分组留一田块协议下进行的实验产生了0.465 ± 0.048的轮廓系数、0.807 ± 0.036的Davies-Bouldin指数和195 ± 27的Calinski-Harabasz指数。学习到的分组还显示出与生长阶段的低归一化互信息（0.079 ± 0.043）以及接近随机的阶段可解码性（平衡准确率 = 0.554 ± 0.042），表明它们与作物生长阶段仅存在弱关联。学习到的注意力权重将IoT土壤测量（0.570 ± 0.024）识别为主导模态，而NDVI影像（0.210 ± 0.014）和天气信息（0.220 ± 0.014）则提供了互补的空间和时间背景。总体而言，AgriMAC为稻田状况的多模态聚类提供了一个可解释且感知数据泄漏的框架。尽管其聚类性能与容量匹配的仅IoT模型相当，但它还通过注意力权重量化了每种传感模态的贡献，并明确减轻了生长阶段混杂因素，使其适用于田块级农艺状况监测和精准农业中的空间决策支持。","International journal of intelligent engineering and systems","2026-09-19T00:00:00Z",72,{"impact":215,"substance":216,"depth":193,"authority":55,"freshness":56,"relevant":22,"comment":217},12,21,"提出无监督多模态注意力聚类框架，融合无人机NDVI、IoT土壤与气象数据评估水稻健康，方法新颖且实验严谨，对精准农业田间监测有参考价值。",[219],{"name":211,"url":208},[61,27,28,221,129],"多模态融合",[223,224],"AgriMAC 水稻 多模态聚类","无人机 NDVI 水稻 病害监测","AgriMAC水稻多模态聚类-3009","10.22266\u002Fijies2026.1031.06",{"doi":226,"openalex_id":228,"authors":229,"venue":211,"cited_by_count":36,"oa_url":208,"card":236,"direction":240,"ingested_from":113},"W7213619014",[230,232,234],{"name":231,"orcid":8},"Nurfadhilah Mardianti Andini",{"name":233,"orcid":8},"Mike Yuliana",{"name":235,"orcid":8},"Moch. Zen Samsono Hadi",{"tldr":237,"method":238,"finding":239,"direction":240,"opportunity":241},"提出无监督多模态聚类框架AgriMAC，融合无人机NDVI、IoT土壤与气象数据评估水稻健康。","各模态自编码器编码，熵正则注意力融合，深度嵌入聚类，按生长阶段残差化。","聚类性能与仅IoT模型相当，但注意力权重可解释模态贡献并弱化生长阶段混淆。","智慧农业 \u002F 农业物联网","可探索注意力融合机制在更多作物与传感器组合下的泛化性，并引入时序动态聚类。","2026-09-20T23:30:08.419613Z",{"id":244,"title":245,"url":246,"summary":247,"summary_zh":8,"content":248,"source_name":249,"source_url":8,"published_at":11,"category":12,"cover_url":8,"hotness":13,"is_selected":14,"score":250,"score_detail":251,"sources":254,"tags":256,"search_phrases":260,"slug":263,"view_count":22,"doi":8,"paper":8,"created_at":264},2984,"中国农科院基因组所李奎团队发布\"伯乐\"多智能体AI平台 开启基因组智能育种新范式","https:\u002F\u002Fwap.sciencenet.cn\u002Fblog-3618084-1552418.html","中国农科院深圳基因组所李奎教授团队在Journal of Integrative Agriculture发表的BOLE多智能体AI平台，将原本高度依赖专业知识的复杂工程任务简化为直观的自然语言对话，显著降低技术使用门槛同时提升分析结果可重复性。BOLE已上线并免费开放使用同时支持本地化部署。论文第一作者为广东省农科院王子帅与深圳基因组所梁崇霄，研究得到国家重点研发计划、猪禽种业全国重点实验室基金等项目支持。","随着高通量基因型与表型数据的快速积累，基因组育种已进入数据密集型时代，但如何将异构的生物信息学工具有效整合为连贯的分析工作流仍是一个主要瓶颈。现有平台多依赖预定义的静态流程，要求用户具备深厚的计算遗传学与编程能力，限制了方法的实际应用与规模化推广。\n\n近期，中国农业科学院（深圳）农业基因组研究所 联合 广东省农业科学院农业生物基因研究中心、香港科技大学（广州）等单位，成功开发了基于知识驱动的多智能体AI平台——BOLE（伯乐）。该平台以“聊天对话框”式的极简交互，实现了基因组育种分析的端到端自动化。相关研究成果以“BOLE: A script-knowledge driven multi-agent framework for reproducible llm-assisted genomic breeding”为题 在**_Journal of Integrative A_****_gricultur_****_e（《农业科学学报_（英文）》，JIA）**优先在线 发表。\n\n**从“手动工程”到“自然语言对话”**\n\nBOLE的核心创新在于将基因组育种分析从传统的手动工程模式，转变为自然语言驱动的对话式操作。研究团队构建了结构化的脚本知识库，并将核心功能模块集成于统一的“聊天对话框”交互界面。在大语言模型驱动下，系统通过交互智能体、规划智能体、流程组装智能体和代码生成智能体四个专业化智能体的协同配合，自主完成从用户意图解析到最终数据分析的全流程。\n\n![Image 1: 2.png](http:\u002F\u002Fimage.sciencenet.cn\u002Fhome\u002F202609\u002F14\u002F115137p77o5o9n00bo5pon.png)\n\n图1 伯乐平台的多智能体架构\n\n**从GWAS到全基因组选择的一站式解决方案**\n\nBOLE整合了多个符合产业标准的生物信息学工具，覆盖了基因组育种分析的四大核心任务：全基因组关联分析（GWAS）、遗传力估计、种质资源评价和全基因组选择。该平台将原本高度依赖专业知识的复杂工程任务，简化为直观的自然语言对话，显著降低了技术使用门槛，同时提升了分析结果的可重复性，为人工智能技术在农业育种领域的应用探索了全新路径。\n\n目前，BOLE已上线并免费开放使用（[https:\u002F\u002Fbole.zishuailab.com\u002F](https:\u002F\u002Fbole.zishuailab.com\u002F)），同时支持本地化部署，充分保障育种数据的安全性。\n\n![Image 2: 1.png](http:\u002F\u002Fimage.sciencenet.cn\u002Fhome\u002F202609\u002F14\u002F115124le0bibyslbru320w.png)\n\n图2 伯乐核心功能示意图\n\n论文第一作者：王子帅（广东省农业科学院农业生物基因研究中心）、梁崇霄（中国农业科学院深圳基因组所）\n\n通讯作者：李奎教授（中国农业科学院深圳基因组所）\n\n本研究得到国家重点研发计划、猪禽种业全国重点实验室基金、深圳市优秀人才培养基金和广东省基金等项目支持。\n\nCite the article:\n\nZishuai Wang, Chongxiao Liang, Yanlin Zhang, Rong Zhou, Xiaoai Zhang, Wenkang Wei, Kui Li. 2026. BOLE: A script-knowledge driven multi-agent framework for reproducible llm-assisted genomic breeding.Journal of Integrative Agriculture, Doi:10.1016\u002Fj.jia.2026.07.011\n\n[https:\u002F\u002Fdoi.org\u002F10.1016\u002Fj.jia.2026.07.011](https:\u002F\u002Fdoi.org\u002F10.1016\u002Fj.jia.2026.07.011)\n\n_Journal of Integrative Agriculture_（《农 业科学学报（英文）》, JIA）由中华人民共和国农业农村部主管，中国农业科学院与中国农学会主办，中国农业科学院农业信息研究所承办。综合性英文学术期刊，月刊。创刊于2002年，现任主编为中国科学院院士陈化兰。JIA主要栏目有作物科学、园艺、植物保护、动物科学、动物医学、农业生态环境、食品科学、农业经济与管理等。刊稿类型有综述、研究论文、简报以及评述等。全部论文在Elsevier-ScienceDirect (SD) 平台OA出版。最新SCI影响因子5.7，位于SCI-JCR农业综合学科Q1区。中国科学院分区农林科学1区。2016年以来先后获得中国科协等部委 “提升计划”“登峰计划”“卓越计划”项目支持。\n\n转载本文请联系原作者获取授权，同时请注明本文来自王宁科学网博客。  \n链接地址： [https:\u002F\u002Fwap.sciencenet.cn\u002Fblog-3618084-1552418.html](https:\u002F\u002Fwap.sciencenet.cn\u002Fblog-3618084-1552418.html)\n\n上一篇：[JIA | 四川农业大学小麦研究所马建教授课题组鉴定并遗传解析一个新的小麦粒长位点](https:\u002F\u002Fblog.sciencenet.cn\u002Fblog-3618084-1552417.html)  \n下一篇：[JIA优先上线｜多单位联合综述从“传统棉作”到“智能设计”：中国棉花产业十年的跃迁](https:\u002F\u002Fblog.sciencenet.cn\u002Fblog-3618084-1552981.html)\n\n[欢迎参加科学网十佳博文评选活动！](https:\u002F\u002Fblog.sciencenet.cn\u002Fblog-45-1520553.html)\n\n主办单位： [![Image 3](https:\u002F\u002Fimage.sciencenet.cn\u002Fhome\u002F202606\u002F03\u002F170346g6h6jzlddzxnvgk1.jpg)](https:\u002F\u002Fwww.sciencenet.cn\u002F)支持单位：[![Image 4](https:\u002F\u002Fimage.sciencenet.cn\u002Fhome\u002F202606\u002F04\u002F150033i1jb7dpbv12pbsbs.jpg)](https:\u002F\u002Fais.cn\u002Fu\u002FQnia6z)","科学网博客",86,{"impact":18,"substance":53,"depth":54,"authority":161,"freshness":252,"relevant":22,"comment":253},8,"国家级科研机构发布的多智能体AI育种平台，方法新颖、开放可用，对智慧育种有实质推动，值得进入每日精选。",[255],{"name":249,"url":246},[27,257,29,258,259],"种业振兴","智能育种","基因组选择",[261,262],"中国农科院基因组所 伯乐 育种平台","BOLE 多智能体 基因组育种","中国农科院基因组所伯乐育种平台-2984","2026-09-20T00:03:03.016540Z"]