[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"item-2213":3},{"id":4,"title":5,"url":6,"summary":7,"summary_zh":8,"content":8,"source_name":9,"source_url":8,"published_at":10,"category":11,"cover_url":8,"hotness":12,"is_selected":13,"score":14,"score_detail":15,"sources":23,"tags":25,"view_count":32,"doi":33,"paper":34,"created_at":43},2213,"《分层跨模态时空注意力网络用于复杂喜马拉雅农业生态系统作物产量预测》","https:\u002F\u002Fwww.frontiersin.org\u002Fjournals\u002Fplant-science\u002Farticles\u002F10.3389\u002Ffpls.2026.1933033\u002Ffull","Mir等提出全面的时空注意力框架,具有四项关键创新:在单树、地块、果园、区域尺度运行的分层注意力机制;融合异构数据流(物联网土壤传感器、气象站、植物生理监测仪、无人机多光谱图像)的跨模态注意力模块;显式建模土壤记忆、滞后和根区耦合的土壤感知注意力头;通过蒙特卡洛Dropout和分位数回归实现不确定性量化。在2023-2026年喜马拉雅苹果园数据集(1247棵监测树)上验证,4周产量预测R²达0.891,RMSE较CNN-LSTM基线降低29.6%,芒果园零样本迁移R²达0.674。",null,"Frontiers in Plant Science 2026年9月10日","2026-09-10T00:00:00Z","论文",10,false,82,{"impact":16,"substance":17,"depth":18,"authority":19,"freshness":20,"relevant":21,"comment":22},18,23,19,14,8,1,"方法新颖、数据规模扎实且含跨作物迁移验证，属智慧农业细分领域高水平研究，值得精选。",[24],{"name":9,"url":6},[26,27,28,29,30,31],"智慧农业","农业人工智能","产量预测","多模态融合","遥感监测","苹果园",0,"10.3389\u002Ffpls.2026.1933033\u002Ffull",{"doi":33,"openalex_id":8,"authors":35,"venue":8,"cited_by_count":32,"oa_url":8,"card":36,"direction":40,"ingested_from":42},[],{"tldr":37,"method":38,"finding":39,"direction":40,"opportunity":41},"提出分层跨模态时空注意力网络，融合多源数据预测喜马拉雅果园作物产量。","分层注意力+跨模态融合物联网、气象、无人机多光谱数据，蒙特卡洛Dropout量化","4周产量预测R²达0.891，RMSE较CNN-LSTM降低29.6%，芒果园零样本迁移R²达0.6","农业人工智能与决策模型","可探索跨物种零样本迁移的域适应机制，以及轻量化模型在边缘设备上的实时部署。","agent","2026-09-12T00:06:40.371547Z"]