[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"item-4437":3,"related-4437":94},{"id":4,"title":5,"url":6,"summary":7,"summary_zh":8,"content":9,"source_name":10,"source_url":6,"published_at":11,"category":12,"cover_url":9,"hotness":13,"is_selected":14,"score":15,"score_detail":16,"sources":23,"tags":25,"primary_topic":9,"secondary_topics":31,"classification":32,"search_phrases":69,"slug":72,"view_count":73,"doi":74,"paper":75,"created_at":93},4437,"A global-local hybrid vision-language framework for interpretable grapevine disease diagnosis","https:\u002F\u002Fdoi.org\u002F10.1007\u002Fs44163-026-02463-x","Abstract Plant diseases pose a severe threat to agricultural productivity and global food security. Although Convolutional Neural Networks have achieved impressive classification accuracy in controlled environments, they often lack semantic reasoning capabilities and operate as uninterpretable “black boxes.” Recently, Vision-Language Models have emerged as powerful multimodal reasoning systems. However, they suffer from the “attention sink” phenomenon when applied to complex real-world images, wasting computational resources on background noise rather than microscopic necrotic lesions. In this paper, we propose a Global–Local Hybrid Framework that mimics the diagnostic workflow of agricultural experts. The system leverages an unsupervised Saliency Auto-Crop algorithm within the CIE-LAB color space to automatically extract local disease patches without the need for bounding box annotations. A dual image stream, comprising global context and local patches, is fed into LLaVA−1.5 7B and fine-tuned using 4-bit QLoRA to ensure parameter efficiency. Furthermore, we extract Cross-Attention Heatmaps from the final layer of the decoder to provide Explainable Artificial Intelligence capabilities. Experimental results on a grapevine disease dataset demonstrate that our method achieves an overall accuracy of 98.7%, significantly outperforming zero-shot baselines while providing explicit semantic interpretability for precision agriculture.","摘要 植物病害对农业生产力与全球粮食安全构成严重威胁。尽管卷积神经网络（Convolutional Neural Networks）在受控环境下已取得令人瞩目的分类精度，但其往往缺乏语义推理能力，且作为不可解释的“黑箱”运行。近年来，视觉-语言模型（Vision-Language Models）作为强大的多模态推理系统崭露头角。然而，将其应用于复杂的真实世界图像时，会出现“注意力汇”现象，将计算资源浪费在背景噪声上，而非微小的坏死病斑。本文提出了一种模拟农业专家诊断工作流程的全局-局部混合框架。该系统利用CIE-LAB色彩空间中的无监督显著性自动裁剪算法，无需边界框标注即可自动提取局部病害斑块。由全局上下文和局部斑块组成的双图像流被输入LLaVA−1.5 7B，并使用4位QLoRA进行微调以确保参数效率。此外，我们从解码器最后一层提取交叉注意力热图，以提供可解释人工智能能力。在葡萄藤病害数据集上的实验结果表明，本方法总体准确率达到98.7%，显著优于零样本基线，同时为精准农业提供了明确的语义可解释性。",null,"Discover Artificial Intelligence","2026-10-07T00:00:00Z","论文",10,false,80,{"impact":17,"substance":18,"depth":17,"authority":19,"freshness":20,"relevant":21,"comment":22},18,22,13,9,1,"提出全局-局部混合视觉语言框架，以98.7%准确率实现葡萄病害可解释诊断，方法新颖且具产业应用潜力。",[24],{"name":10,"url":6},[26,27,28,29,30],"智慧农业","农业人工智能","可解释AI","多模态大模型","葡萄病害",[],{"status":33,"locked":14,"version":34,"method":35,"evidence":36},"needs_review","2026-10-22.2","rules",{"candidates":37,"agricultural_evidence":41,"reason":68},[38,47,55,61],{"code":39,"score":40,"supported":41,"title":42,"abstract":43,"tags":46},"I1",7,true,[],[44,45],"精准农业","precision agriculture",[26],{"code":48,"score":49,"supported":41,"title":50,"abstract":51,"tags":54},"A8",6,[],[52,53],"粮食安全","food security",[],{"code":56,"score":49,"supported":41,"title":57,"abstract":58,"tags":60},"I2",[],[59],"artificial intelligence",[27,28,29],{"code":62,"score":63,"supported":41,"title":64,"abstract":65,"tags":67},"B2",3,[],[66],"葡萄",[],"多个产业证据相当，需要确认主归属",[70,71],"葡萄病害 视觉语言模型 诊断","LLaVA QLoRA 葡萄","葡萄病害视觉语言模型诊断-4437",4,"10.1007\u002Fs44163-026-02463-x",{"doi":74,"openalex_id":76,"authors":77,"venue":10,"cited_by_count":83,"oa_url":84,"card":85,"direction":91,"ingested_from":92},"W7220864211",[78,80],{"name":79,"orcid":9},"Lang Hoang Son",{"name":81,"orcid":82},"Bùi Thanh Hùng","https:\u002F\u002Forcid.org\u002F0000-0002-9400-7582",0,"https:\u002F\u002Flink.springer.com\u002Fcontent\u002Fpdf\u002F10.1007\u002Fs44163-026-02463-x.pdf",{"tldr":86,"method":87,"finding":88,"direction":89,"opportunity":90},"提出全局-局部混合视觉语言框架，实现可解释的葡萄病害诊断，准确率达98.7%。","无监督显著性自动裁剪提取局部病斑，双流图像输入LLaVA-1.5 7B，4-bi","方法显著优于零样本基线，并提供跨注意力热图实现可解释诊断。","农业人工智能与决策模型","可探索无标注病斑定位与轻量化VLM在田间实时诊断中的泛化能力。","农业遥感与作物表型","openalex","2026-10-08T23:30:29.641451Z",{"total":49,"page":21,"page_size":49,"items":95},[96,172,241,304,380,451],{"id":97,"title":98,"url":99,"summary":100,"summary_zh":101,"content":9,"source_name":102,"source_url":99,"published_at":103,"category":12,"cover_url":9,"hotness":13,"is_selected":14,"score":15,"score_detail":104,"sources":108,"tags":110,"primary_topic":9,"secondary_topics":112,"classification":113,"search_phrases":144,"slug":147,"view_count":148,"doi":149,"paper":150,"created_at":171},3159,"A multi-agent multi-modal LLM with Monte Carlo tree search for interpretable plant disease classification","https:\u002F\u002Fdoi.org\u002F10.3389\u002Ffpls.2026.1937679","In plant disease classification tasks, existing task specific models for plant disease classification typically achieve high accuracy by employing tailored network architectures. However, these models are unable to provide reasoning chains, which reduces interpretability and undermines trustworthiness in practical applications. In addition, although multimodal large language models (MLLMs) have made significant progress in smart agriculture, their disease classification accuracy lags behind that of task-specific models due to insufficient domain knowledge. While fine-tuning can enhance domain-specific knowledge, relying solely on unimodal data or simple multimodal information remains insufficient for adequately improving classification accuracy. Specifically, the simple multimodal information in this context refers to a single global description of the entire image produced by an MLLM. To address these issues, this paper proposes a multi-agent collaborative classification framework that integrates Monte Carlo Tree Search (MCTS) with multimodal enhancement. With an MLLM as the core, the framework first introduces an MCTSguided multi-agent collaboration mechanism that directs the agents to progressively generate region-specific descriptive textual information, which together with the image forms complex multimodal information and thereby boosts classification performance. Meanwhile, the search trajectories generated by MCTS are leveraged to construct reasoning processes for enhanced interpretability. Finally, a self-training mechanism is adopted to iteratively optimize the model, further improving overall performance. Experimental results show that the framework improves the baseline MLLM’s accuracy while providing interpretable reasoning chains, achieving gains from 0.6701 to 0.9965 on 3-class potato, 0.5340 to 0.9651 on 5-class potato, 0.5387 to 0.9467 on tomato, and 0.4705 to 0.9948 on grape.","在植物病害分类任务中，现有的植物病害分类专用模型通常通过采用定制化的网络架构来实现高准确率。然而，这些模型无法提供推理链，这降低了可解释性，并在实际应用中削弱了可信度。此外，尽管多模态大语言模型（MLLMs）在智慧农业领域取得了显著进展，但由于领域知识不足，其病害分类准确率仍落后于专用模型。虽然微调可以增强领域特定知识，但仅依赖单模态数据或简单的多模态信息仍不足以充分提升分类准确率。具体而言，此处的简单多模态信息是指MLLM对整个图像生成的单一全局描述。为解决上述问题，本文提出了一种融合蒙特卡洛树搜索（MCTS）与多模态增强的多智能体协同分类框架。该框架以MLLM为核心，首先引入MCTS引导的多智能体协同机制，引导各智能体逐步生成区域特定的描述性文本信息，这些信息与图像共同构成复杂的多模态信息，从而提升分类性能。同时，利用MCTS生成的搜索轨迹构建推理过程，以增强可解释性。最后，采用自训练机制迭代优化模型，进一步提升整体性能。实验结果表明，该框架在提升基线MLLM准确率的同时提供了可解释的推理链，在三分类马铃薯上从0.6701提升至0.9965，在五分类马铃薯上从0.5340提升至0.9651，在番茄上从0.5387提升至0.9467，在葡萄上从0.4705提升至0.9948。","Frontiers in Plant Science","2026-09-21T00:00:00Z",{"impact":17,"substance":18,"depth":17,"authority":105,"freshness":106,"relevant":21,"comment":107},14,8,"提出多智能体+MCTS+多模态增强的可解释植物病害分类框架，准确率大幅提升，方法新颖且结论可靠，对智慧农业AI应用有参考价值。",[109],{"name":102,"url":99},[26,27,28,111,29],"植物病害识别",[],{"status":33,"locked":14,"version":34,"method":35,"evidence":114},{"candidates":115,"agricultural_evidence":41,"reason":68},[116,121,128,135,140],{"code":39,"score":40,"supported":41,"title":117,"abstract":118,"tags":120},[],[26,119],"smart agriculture",[26],{"code":122,"score":49,"supported":41,"title":123,"abstract":124,"tags":127},"A5",[],[125,126],"马铃薯","potato",[],{"code":129,"score":49,"supported":41,"title":130,"abstract":131,"tags":134},"B1",[],[132,133],"番茄","tomato",[],{"code":62,"score":49,"supported":41,"title":136,"abstract":137,"tags":139},[],[66,138],"grape",[],{"code":56,"score":63,"supported":14,"title":141,"abstract":142,"tags":143},[],[],[27,28,29,111],[145,146],"多智能体 蒙特卡洛树搜索 植物病害分类","多模态大模型 马铃薯 番茄 葡萄 病害识别","多智能体蒙特卡洛树搜索植物病害分类-3159",2,"10.3389\u002Ffpls.2026.1937679",{"doi":149,"openalex_id":151,"authors":152,"venue":102,"cited_by_count":83,"oa_url":99,"card":165,"direction":170,"ingested_from":92},"W7213924931",[153,156,158,160,163],{"name":154,"orcid":155},"Ridong Wu","https:\u002F\u002Forcid.org\u002F0000-0003-0729-3099",{"name":157,"orcid":9},"Hua Zou",{"name":159,"orcid":9},"Zhijie Li",{"name":161,"orcid":162},"Qian Zhou","https:\u002F\u002Forcid.org\u002F0000-0001-7964-8130",{"name":164,"orcid":9},"一子 江﨑",{"tldr":166,"method":167,"finding":168,"direction":89,"opportunity":169},"提出多智能体多模态LLM结合蒙特卡洛树搜索，实现可解释的植物病害分类。","MCTS引导多智能体生成区域描述文本，构建复杂多模态信息并自训练优化。","分类准确率大幅提升，如马铃薯三分类从0.6701升至0.9965，并提供推理链。","可探索将MCTS多智能体框架迁移至其他作物病害或田间复杂场景，提升可解释性与泛化性。","智慧农业 \u002F 农业物联网","2026-09-22T23:30:11.045394Z",{"id":173,"title":174,"url":175,"summary":176,"summary_zh":177,"content":9,"source_name":178,"source_url":175,"published_at":179,"category":12,"cover_url":9,"hotness":13,"is_selected":14,"score":180,"score_detail":181,"sources":183,"tags":185,"primary_topic":188,"secondary_topics":193,"classification":194,"search_phrases":208,"slug":211,"view_count":83,"doi":212,"paper":213,"created_at":240},5562,"pxGPT: A schema-constrained multimodal LLM framework for plant phenotyping","https:\u002F\u002Fdoi.org\u002F10.1093\u002Fbioinformatics\u002Fbtag781","Abstract Summary Deep learning can scale image-based plant phenotyping but typically targets predefined traits, whereas manual phenotypic description is more flexible but difficult to scale. We present Phenotype eXplorer Generative Plant Traits (pxGPT), a Python framework that converts multi-view plant images into structured phenotype records. The workflow combines free-text description, dataset-specific schema generation with human review, and schema-constrained information extraction. We evaluated the extraction stage on three Brassica datasets using 78,973 annotations from ten annotators who had no access to pxGPT predictions. For 38-45 qualitative traits per dataset, pxGPT-human agreement ranged from 0.572 to 0.712, compared with 0.620 to 0.739 agreement among human annotators. All 32,736 required trait fields were present; 93 values were empty, and no qualitative descriptor violations were recorded. For selected morphological structures, pooled false-positive rates were 0.5-8.0%, while sensitivity varied across structure types in one mature-plant dataset (37.5-100%). The curated schemas were applied unchanged with two commercial models and one locally hosted open-weight model, with differences in mean accuracy of 0.03 or less. In the evaluated Brassica datasets, these results support schema-guided extraction of visible plant traits without separate model training for each trait. Human-guided schema development offers a path to extend pxGPT to other crops and imaging conditions, with accuracy requiring validation in each new setting. Availability and Implementation The software is available at https:\u002F\u002Fgithub.com\u002FYaoChengLab\u002Fpxgpt and an archival version of the source code is available on Zenodo (https:\u002F\u002Fdoi.org\u002F10.5281\u002Fzenodo.22931745). Curated schemas used in this study, annotation guides, phenotype records and human annotations are available at https:\u002F\u002Fdoi.org\u002F10.5281\u002Fzenodo.22907554. Supplementary information Supplementary data are available at Bioinformatics online.","摘要 深度学习能够扩展基于图像的植物表型分析，但通常针对预定义性状，而人工表型描述更为灵活却难以规模化。我们提出Phenotype eXplorer Generative Plant Traits（pxGPT），一个将多视角植物图像转换为结构化表型记录的Python框架。该工作流程结合了自由文本描述、数据集特定模式生成与人工审核，以及模式约束的信息提取。我们在三个芸薹属（Brassica）数据集上评估了提取阶段，使用了来自十名标注者的78,973条标注，这些标注者无法获取pxGPT的预测结果。对于每个数据集38-45个定性性状，pxGPT与人工标注者之间的一致性范围为0.572至0.712，而人工标注者之间的一致性为0.620至0.739。所有32,736个所需性状字段均已存在；93个值为空，未记录到定性描述符违规。对于选定的形态结构，合并假阳性率为0.5-8.0%，而在一个成熟植株数据集中，灵敏度因结构类型而异（37.5-100%）。经整理的模式在未作修改的情况下应用于两个商业模型和一个本地部署的开放权重模型，平均准确率差异不超过0.03。在所评估的芸薹属数据集中，这些结果支持在无需为每个性状单独训练模型的情况下，对可见植物性状进行模式引导的提取。人工引导的模式开发为将pxGPT扩展至其他作物和成像条件提供了途径，其准确性需在每个新设置中加以验证。可用性与实现 该软件可在https:\u002F\u002Fgithub.com\u002FYaoChengLab\u002Fpxgpt获取，源代码的存档版本可在Zenodo（https:\u002F\u002Fdoi.org\u002F10.5281\u002Fzenodo.22931745）获取。本研究中使用的经整理模式、标注指南、表型记录和人工标注可在https:\u002F\u002Fdoi.org\u002F10.5281\u002Fzenodo.22907554获取。补充信息 补充数据可在Bioinformatics在线获取。","Bioinformatics","2026-10-20T00:00:00Z",81,{"impact":17,"substance":18,"depth":17,"authority":105,"freshness":20,"relevant":21,"comment":182},"提出基于模式约束的多模态大模型植物表型提取框架，在近8万条标注上验证，方法新颖、数据规模可观，对智慧育种与表型自动化有参考价值。",[184],{"name":178,"url":175},[26,27,186,29,187],"植物表型","芸薹属",{"code":56,"slug":189,"name":190,"path":191,"parent_code":192},"agricultural-ai","农业AI与大模型","\u002Ftopic\u002Fdigital-agriculture\u002Fagricultural-ai","I",[],{"status":195,"locked":14,"version":34,"method":35,"evidence":196},"classified",{"candidates":197,"agricultural_evidence":41},[198,204],{"code":56,"score":106,"supported":41,"title":199,"abstract":200,"tags":203},[],[201,202],"深度学习","deep learning",[27,29],{"code":39,"score":21,"supported":14,"title":205,"abstract":206,"tags":207},[],[],[26],[209,210],"pxGPT 植物表型","Brassica 表型 多模态大模型","pxGPT植物表型-5562","10.1093\u002Fbioinformatics\u002Fbtag781",{"doi":212,"openalex_id":214,"authors":215,"venue":178,"cited_by_count":83,"oa_url":234,"card":235,"direction":170,"ingested_from":92},"W7223127960",[216,219,221,224,227,230,232],{"name":217,"orcid":218},"Po‐Xing Zheng","https:\u002F\u002Forcid.org\u002F0000-0001-5827-5865",{"name":220,"orcid":9},"Jheng-Yan Ou",{"name":222,"orcid":223},"Yu-Lin Wu","https:\u002F\u002Forcid.org\u002F0009-0003-8596-9047",{"name":225,"orcid":226},"Ruei-Shiuan Wang","https:\u002F\u002Forcid.org\u002F0009-0007-5159-1129",{"name":228,"orcid":229},"WU Ting-li","https:\u002F\u002Forcid.org\u002F0000-0001-7996-5515",{"name":231,"orcid":9},"Te-Chang Hsu",{"name":233,"orcid":9},"Yao-Cheng Lin","https:\u002F\u002Facademic.oup.com\u002Fbioinformatics\u002Fadvance-article-pdf\u002Fdoi\u002F10.1093\u002Fbioinformatics\u002Fbtag781\u002F72201337\u002Fbtag781.pdf",{"tldr":236,"method":237,"finding":238,"direction":91,"opportunity":239},"提出pxGPT框架，用多模态大模型将多视角植物图像转为结构化表型记录。","结合自由文本描述、人工审核的模式生成与模式约束信息抽取，在三个芸薹属数据集上评估","pxGPT与人工标注一致性0.572-0.712，接近人工间一致性，且无需逐性状训练模型。","可探索将模式引导抽取迁移至更多作物与复杂成像条件，并研究自动化模式生成以降低人工审核成本。","2026-10-22T11:20:22.173561Z",{"id":242,"title":243,"url":244,"summary":245,"summary_zh":246,"content":9,"source_name":247,"source_url":244,"published_at":248,"category":12,"cover_url":9,"hotness":13,"is_selected":14,"score":249,"score_detail":250,"sources":254,"tags":256,"primary_topic":259,"secondary_topics":260,"classification":261,"search_phrases":280,"slug":283,"view_count":21,"doi":284,"paper":285,"created_at":303},5194,"A Comprehensive Review on Multi-Modal Large Language Models and Agentic Architectures for Precision Agricultural Decision Support","https:\u002F\u002Fdoi.org\u002F10.5281\u002Fzenodo.23411181","This article reviews recent developments in multi-modal large language models and agentic architectures for agricultural decision support. Forty-five representative studies published between 2023 and 2026 were selected across perception, language, orchestration, soil intelligence and governance, organised under a four-layer taxonomy, and classified against nine functional capabilities using a stated, reproducible procedure. The comparative analysis exposes a pronounced imbalance. Crop and disease perception is covered by 36 per cent of the selected studies, while dose-bearing prescription appears in 7 per cent, profit-aware recommendation in 4 per cent, and persistent per-field memory in a single study. Advisory tools in this literature therefore remain unable to exchange state: image-based scanners are rarely conditioned on measured soil chemistry, and crop-recommendation models optimise agronomic suitability without reference to farm economics. On this basis the article synthesises a conceptual reference architecture from the reviewed literature. It is a literature-derived synthesis rather than an implemented system, and calibrated abstention, evaluation protocols for longitudinal memory, and vernacular voice coverage emerge as the most urgent gaps.","本文综述了面向农业决策支持的多模态大语言模型与智能体架构的最新进展。按照可复现的既定流程，选取2023年至2026年间发表的45项代表性研究，涵盖感知、语言、编排、土壤智能与治理等方向，将其纳入一个四层分类体系，并依据九项功能能力进行分类。比较分析揭示出明显的不均衡：作物与病害感知在所选研究中占36%，而含剂量处方的研究仅占7%，利润感知推荐占4%，持久性田块记忆仅见于一项研究。因此，该领域的咨询工具仍无法交换状态：基于图像的扫描器很少以实测土壤化学状况为条件，作物推荐模型在优化农艺适宜性时也未考虑农场经济性。在此基础上，本文综合所综述文献提出一个概念性参考架构。该架构是文献衍生的综合框架，而非已实现的系统；校准弃权、纵向记忆的评估协议以及方言语音覆盖被列为最紧迫的空白。","Zenodo (CERN European Organization for Nuclear Research)","2026-10-16T00:00:00Z",63,{"impact":251,"substance":17,"depth":252,"authority":106,"freshness":20,"relevant":21,"comment":253},12,16,"系统性综述揭示处方决策与农场经济维度严重缺位，对农业AI研发方向有参考价值，但属文献综述而非实证突破，影响范围有限。",[255],{"name":247,"url":244},[26,27,257,29,258],"智能体","农业决策支持",{"code":56,"slug":189,"name":190,"path":191,"parent_code":192},[],{"status":195,"locked":14,"version":34,"method":35,"evidence":262},{"candidates":263,"agricultural_evidence":41},[264,269,276],{"code":56,"score":105,"supported":41,"title":265,"abstract":267,"tags":268},[266],"decision support",[266],[27,29,257],{"code":270,"score":49,"supported":41,"title":271,"abstract":272,"tags":275},"H2",[],[273,274],"土壤","soil",[],{"code":39,"score":21,"supported":14,"title":277,"abstract":278,"tags":279},[],[],[26],[281,282],"多模态大模型 农业决策支持","农业人工智能 农业决策支持 多模态大模型 智慧农业","多模态大模型农业决策支持-5194","10.5281\u002Fzenodo.23411181",{"doi":284,"openalex_id":286,"authors":287,"venue":247,"cited_by_count":83,"oa_url":244,"card":298,"direction":89,"ingested_from":92},"W7222605322",[288,290,292,294,296],{"name":289,"orcid":9},"Tanmay Bhamre",{"name":291,"orcid":9},"Yash Barde",{"name":293,"orcid":9},"Devendra Yeole",{"name":295,"orcid":9},"Gajanan Chaudhary",{"name":297,"orcid":9},"Amarsinh Verpe",{"tldr":299,"method":300,"finding":301,"direction":89,"opportunity":302},"综述多模态大模型与智能体架构在精准农业决策支持中的应用与不足。","筛选2023-2026年45篇代表性研究，按四层分类与九项功能能力归类比较。","感知研究占36%，处方仅7%、利润推荐4%、田块记忆仅1篇，系统难以交换状态。","可研究跨模态状态交换、纵向记忆评测协议与方言语音覆盖，填补处方与经济决策空白。","2026-10-17T23:30:55.282694Z",{"id":305,"title":306,"url":307,"summary":308,"summary_zh":309,"content":9,"source_name":310,"source_url":307,"published_at":311,"category":12,"cover_url":9,"hotness":13,"is_selected":14,"score":312,"score_detail":313,"sources":317,"tags":319,"primary_topic":321,"secondary_topics":322,"classification":323,"search_phrases":346,"slug":349,"view_count":83,"doi":350,"paper":351,"created_at":379},5080,"Explainable deep learning for Mango leaf disease recognition using AlexNet and grad-CAM","https:\u002F\u002Fdoi.org\u002F10.1177\u002F1088467x261495708","Early and accurate detection of plant diseases is essential for improving crop productivity and supporting sustainable agricultural practices. Mango (Mangifera indica), one of the most economically important tropical fruit crops worldwide, is highly susceptible to several leaf diseases that can significantly reduce yield and fruit quality. This study presents an explainable deep learning framework for automated ten-class mango leaf disease classification using a publicly available dataset containing approximately 5000 images distributed across Anthracnose, Bacterial Canker, Cutting Weevil, Dag Disease, Die Back, Galls, Healthy, Powdery Mildew, Red Rust, and Sooty Mould classes. Unlike conventional AlexNet-based approaches, the proposed framework integrates Bayesian hyperparameter optimization, two-stage fine-tuning, test-time augmentation (TTA), and Grad-CAM-based explainability to improve both predictive performance and interpretability. Experimental results demonstrate that the proposed framework achieves a final test accuracy of 99.2% using eight-view test-time augmentation (TTA), with corresponding precision, recall, and F1-score of approximately 99.1%. The model also achieves micro-average and macro-average ROC–AUC scores of 1.00 on the curated test set, indicating strong discriminative capability across the considered disease categories. Comparative evaluations against MobileNetV2, DenseNet201, ConvNeXt, and LeafNet demonstrate competitive or superior classification performance while maintaining moderate computational complexity and architectural simplicity. Furthermore, Grad-CAM visualizations qualitatively highlight biologically relevant disease symptoms, including lesions, necrotic regions, rust spots, fungal growth, and tissue discoloration, thereby enhancing model interpretability. Although field-scale validation and quantitative explainability assessment remain subjects for future investigation, the proposed framework provides an accurate, computationally efficient, and interpretable approach for automated mango leaf disease diagnosis with strong potential for smart agriculture and precision farming applications.","植物病害的早期准确检测对于提高作物生产力和支持可持续农业实践至关重要。芒果（Mangifera indica）是全球最具经济重要性的热带水果作物之一，极易感染多种叶部病害，这些病害可显著降低产量和果实品质。本研究提出了一种可解释的深度学习框架，用于自动化十类芒果叶部病害分类，所使用的公开数据集包含约5000张图像，分布于炭疽病（Anthracnose）、细菌性溃疡病（Bacterial Canker）、切叶象甲（Cutting Weevil）、枯枝病（Dag Disease）、回枯病（Die Back）、虫瘿（Galls）、健康（Healthy）、白粉病（Powdery Mildew）、红锈病（Red Rust）和煤烟病（Sooty Mould）十个类别。与传统的基于AlexNet的方法不同，所提出的框架集成了贝叶斯超参数优化、两阶段微调、测试时增强（TTA）以及基于Grad-CAM的可解释性，以同时提升预测性能和可解释性。实验结果表明，所提出的框架在使用八视图测试时增强（TTA）时达到了99.2%的最终测试准确率，相应的精确率、召回率和F1分数约为99.1%。该模型在精选测试集上还取得了1.00的微平均和宏平均ROC–AUC分数，表明在所考虑的病害类别间具有较强的判别能力。与MobileNetV2、DenseNet201、ConvNeXt和LeafNet的对比评估表明，该框架在保持适中计算复杂度和架构简洁性的同时，具有竞争性或更优的分类性能。此外，Grad-CAM可视化定性地突出了具有生物学相关性的病害症状，包括病斑、坏死区域、锈斑、真菌生长和组织变色，从而增强了模型的可解释性。尽管田间规模验证和定量可解释性评估仍有待未来研究，但所提出的框架为自动化芒果叶部病害诊断提供了一种准确、计算高效且可解释的方法，在智慧农业和精准农业应用中具有强大潜力。","Intelligent Data Analysis","2026-10-15T00:00:00Z",76,{"impact":252,"substance":314,"depth":315,"authority":19,"freshness":20,"relevant":21,"comment":316},21,17,"将贝叶斯调参、两阶段微调与Grad-CAM结合用于十类芒果叶病识别，准确率99.2%，方法新颖且可解释性强，对智慧农业植保应用有参考价值。",[318],{"name":310,"url":307},[26,27,28,44,320],"芒果病害",{"code":56,"slug":189,"name":190,"path":191,"parent_code":192},[],{"status":195,"locked":14,"version":34,"method":35,"evidence":324},{"candidates":325,"agricultural_evidence":41},[326,330,335,340],{"code":56,"score":252,"supported":41,"title":327,"abstract":328,"tags":329},[202],[201,202],[27,28],{"code":39,"score":331,"supported":41,"title":332,"abstract":333,"tags":334},11,[],[26,44,119],[26,44],{"code":62,"score":63,"supported":41,"title":336,"abstract":337,"tags":339},[],[338],"芒果",[],{"code":341,"score":63,"supported":41,"title":342,"abstract":343,"tags":345},"H8",[],[344],"可持续农业",[],[347,348],"芒果叶片病害 AlexNet","Grad-CAM 芒果病害识别","芒果叶片病害AlexNet-5080","10.1177\u002F1088467x261495708",{"doi":350,"openalex_id":352,"authors":353,"venue":310,"cited_by_count":83,"oa_url":9,"card":374,"direction":170,"ingested_from":92},"W7222328160",[354,357,360,362,365,368,371],{"name":355,"orcid":356},"Anay Ghosh","https:\u002F\u002Forcid.org\u002F0009-0006-6237-7905",{"name":358,"orcid":359},"Tawfikur Rahman","https:\u002F\u002Forcid.org\u002F0000-0002-0456-5143",{"name":361,"orcid":9},"Md Moniruzzaman",{"name":363,"orcid":364},"Nibedita Deb","https:\u002F\u002Forcid.org\u002F0000-0002-1579-1138",{"name":366,"orcid":367},"Hanan Abdullah Mengash","https:\u002F\u002Forcid.org\u002F0000-0002-4103-2434",{"name":369,"orcid":370},"Noorlindawaty Md Jizat","https:\u002F\u002Forcid.org\u002F0000-0001-7618-2283",{"name":372,"orcid":373},"Samir Salem Al‐Bawri","https:\u002F\u002Forcid.org\u002F0000-0003-2852-575X",{"tldr":375,"method":376,"finding":377,"direction":89,"opportunity":378},"提出可解释深度学习框架，用AlexNet结合贝叶斯调参、微调和Grad-CAM实现芒果叶病十分类。","AlexNet+贝叶斯超参优化、两阶段微调、八视图TTA、Grad-CAM，约5","测试准确率达99.2%，ROC-AUC为1.00，Grad-CAM能定位病斑等生物相关症状。","可延伸至田间复杂背景验证、可解释性定量评估及轻量化边缘部署研究。","2026-10-16T23:30:08.303796Z",{"id":381,"title":382,"url":383,"summary":384,"summary_zh":385,"content":9,"source_name":386,"source_url":383,"published_at":311,"category":12,"cover_url":9,"hotness":13,"is_selected":14,"score":387,"score_detail":388,"sources":391,"tags":393,"primary_topic":396,"secondary_topics":397,"classification":398,"search_phrases":430,"slug":433,"view_count":21,"doi":434,"paper":435,"created_at":450},5076,"Trustworthy Explainable AI for Lettuce Growth Stage Prediction Using Sensor Data in Smart Agriculture","https:\u002F\u002Fdoi.org\u002F10.20944\u002Fpreprints202610.1452.v1","Background In precision agriculture, especially in hydroponic systems where environmental management affects both production and resource efficiency, interpretable modeling of crop phenological stages is crucial. Typical visualization-based approaches are generally expensive and ecologically dependent. To classify and visualize lettuce growth stages (Seedling, Vegetative Growth, Heading, Bolting and Senescence) using temperature, humidity, total dissolved solids (TDS) and pH, this study proposed XGB-StageVision, a sensor-powered machine learning setup combining supervised and unsupervised learning techniques. Using XG Boost, LightGBM, CatBoost, ExtraTrees, Random Forest and LSTM under an 80\u002F20 train-test split, a dataset of 3,169 unique observations from 70 hydroponically grown lettuce plants was evaluated. Methods and Results XGB-StageVision achieved 99.86% test accuracy, a ROC AUC of 1.00, an MCC of 0.9982 and perfect precision, recall, F1 scores and specificity across all classes, performing competitively with the best baseline models, including LightGBM, which achieved statistically comparable cross validation performance. The unsupervised t-SNE visualization showed clearly distinct structural clusters across growth stages, while SHAP revealed temperature as the strongest predictor (≈75% contribution), followed by TDS. Novel Contributions SHAP, LIME and t-SNE provided global, local and structural interpretability and transparent perspectives on environmental influences regarding plant development. Statistical analysis and 10-fold cross-validation were used to further test the model’s robustness, reliability and generalizability. Conclusion XGB-StageVision provides an IoT-ready, interpretable and imaging-free system for monitoring lettuce growth stages in real time. The framework combines supervised predictive modeling with unsupervised structural visualization to ensure transparency, sustainability and high performance, supporting resource-efficient, data-driven strategic planning in contemporary hydroponic agriculture to lower monitoring costs, optimize inputs and improve yield consistency.","背景 在精准农业中，尤其是在环境管理同时影响产量和资源效率的水培系统中，作物物候阶段的可解释建模至关重要。典型的基于可视化的方法通常成本高昂且依赖生态条件。为了利用温度、湿度、总溶解固体（TDS）和pH对生菜生长阶段（幼苗期、营养生长期、结球期、抽薹期和衰老期）进行分类和可视化，本研究提出了XGB-StageVision，一种由传感器驱动的机器学习方案，结合了监督学习和无监督学习技术。在80\u002F20的训练-测试划分下，使用XGBoost、LightGBM、CatBoost、ExtraTrees、随机森林和LSTM，对来自70株水培生菜的3，169个独特观测值数据集进行了评估。方法与结果 XGB-StageVision达到了99.86%的测试准确率、1.00的ROC AUC、0.9982的MCC，并在所有类别上实现了完美的精确率、召回率、F1分数和特异性，与最佳基线模型表现相当，包括LightGBM，其交叉验证性能在统计上具有可比性。无监督t-SNE可视化显示了各生长阶段之间明显不同的结构聚类，而SHAP揭示温度是最强的预测因子（约75%贡献），其次是TDS。新颖贡献 SHAP、LIME和t-SNE提供了全局、局部和结构可解释性，以及关于环境对植物发育影响的可透明视角。使用统计分析和10折交叉验证进一步检验了模型的稳健性、可靠性和泛化性。结论 XGB-StageVision提供了一种物联网就绪、可解释且无需成像的系统，用于实时监测生菜生长阶段。该框架将监督预测建模与无监督结构可视化相结合，以确保透明度、可持续性和高性能，支持当代水培农业中资源高效、数据驱动的战略规划，以降低监测成本、优化投入并提高产量一致性。","Preprints.org",64,{"impact":251,"substance":389,"depth":315,"authority":49,"freshness":20,"relevant":21,"comment":390},20,"预印本论文，方法新颖、数据规模可观且强调可解释性，但尚未经同行评审，属细分领域技术进展。",[392],{"name":386,"url":383},[26,27,28,394,395],"水培生菜","传感器监测",{"code":56,"slug":189,"name":190,"path":191,"parent_code":192},[],{"status":195,"locked":14,"version":34,"method":35,"evidence":399},{"candidates":400,"agricultural_evidence":41},[401,406,412,419,424],{"code":39,"score":402,"supported":41,"title":403,"abstract":404,"tags":405},15,[119],[44,45],[26],{"code":56,"score":106,"supported":41,"title":407,"abstract":408,"tags":411},[],[409,410],"机器学习","machine learning",[27,28],{"code":413,"score":49,"supported":41,"title":414,"abstract":415,"tags":418},"I5",[],[416,417],"物联网","IoT",[],{"code":129,"score":63,"supported":41,"title":420,"abstract":421,"tags":423},[],[422],"生菜",[],{"code":425,"score":63,"supported":41,"title":426,"abstract":427,"tags":429},"B7",[],[428],"hydroponic",[],[431,432],"XGB-StageVision 生菜 生长阶段","水培生菜 传感器 生长预测","XGB-StageVision生菜生长阶段-5076","10.20944\u002Fpreprints202610.1452.v1",{"doi":434,"openalex_id":436,"authors":437,"venue":386,"cited_by_count":83,"oa_url":383,"card":445,"direction":170,"ingested_from":92},"W7222080602",[438,440,443],{"name":439,"orcid":9},"Muhammad Majid",{"name":441,"orcid":442},"Attia Ghafoor","https:\u002F\u002Forcid.org\u002F0009-0000-3828-6479",{"name":444,"orcid":9},"Amna Munir",{"tldr":446,"method":447,"finding":448,"direction":170,"opportunity":449},"提出XGB-StageVision，用传感器数据与可解释AI预测水培生菜生长阶段。","XGBoost等模型结合SHAP、LIME、t-SNE，基于3169条温湿度TD","测试准确率99.86%，温度贡献约75%，t-SNE显示各生长阶段聚类清晰。","可探索多作物、多环境迁移与低成本传感器部署下的可解释模型泛化能力。","2026-10-16T23:30:08.020788Z",{"id":452,"title":453,"url":454,"summary":455,"summary_zh":456,"content":9,"source_name":457,"source_url":454,"published_at":458,"category":12,"cover_url":9,"hotness":13,"is_selected":14,"score":459,"score_detail":460,"sources":462,"tags":464,"primary_topic":467,"secondary_topics":473,"classification":475,"search_phrases":496,"slug":499,"view_count":21,"doi":500,"paper":501,"created_at":526},4993,"Effects of within-season temporal aggregation of weather variables on performance and interpretability in rice yield machine learning models","https:\u002F\u002Fdoi.org\u002F10.1038\u002Fs41598-026-75646-2","While crop responses are inherently stage-dependent, machine learning yield models often rely on weather variables aggregated over the entire growing season. This study evaluates how within-season temporal aggregation of weather variables affects both predictive performance and interpretability in machine learning-based rice yield modeling. Using CYGMA (Crop Yield Growth Model with Assumptions on Climate and Socioeconomics)-simulated potential yield as a synthetic response variable, we minimized confounding effects from observational noise and management variability, enabling a clearer evaluation of the role of within-season weather aggregation. Daily weather variables were averaged over the rice growing period after dividing each season into multiple equal-length segments, and several machine learning algorithms were trained and evaluated using temporally independent validation. Increasing within-season temporal resolution improved model performance up to a moderate level, after which gains diminished. Interpretation using SHAP (SHapley Additive exPlanations) and response diagnostics consistently indicated that yield sensitivity concentrates in specific windows of the season, with temperature-related effects emerging most strongly early in the season and humidity-related effects later. Finer temporal aggregation sharpened the identification of window-specific nonlinear and threshold-like responses that were obscured under coarse seasonal averages. Overall, moderate within-season aggregation provides a practical balance between predictive skill and mechanistic interpretability, supporting more reliable identification of phenology-relevant meteorological risks for climate-impact assessment and decision support.","尽管作物响应本质上具有阶段依赖性，但机器学习产量模型往往依赖于整个生长季聚合的天气变量。本研究评估了天气变量的季节内时间聚合如何影响基于机器学习的水稻产量建模的预测性能与可解释性。以CYGMA（Crop Yield Growth Model with Assumptions on Climate and Socioeconomics，气候与社会经济假设下的作物产量生长模型）模拟的潜在产量作为合成响应变量，我们最大限度地减少了观测噪声和管理变异性带来的混杂效应，从而能够更清晰地评估季节内天气聚合的作用。将每个生长季划分为多个等长时段后，对水稻生长期内的逐日天气变量进行平均，并采用时间独立验证对多种机器学习算法进行训练和评估。提高季节内时间分辨率可提升模型性能，直至达到中等水平，此后增益递减。利用SHAP（SHapley Additive exPlanations，沙普利加性解释）和响应诊断进行的解释一致表明，产量敏感性集中在季节的特定窗口，温度相关效应在季节早期最为显著，湿度相关效应则在后期出现。更精细的时间聚合增强了对窗口特异性非线性和阈值型响应的识别，而这些响应在粗粒度季节平均下被掩盖。总体而言，中等程度的季节内聚合在预测能力与机制可解释性之间提供了实用平衡，有助于更可靠地识别与物候相关的气象风险，以支持气候影响评估和决策支持。","Scientific Reports","2026-10-13T00:00:00Z",72,{"impact":251,"substance":389,"depth":17,"authority":105,"freshness":106,"relevant":21,"comment":461},"该研究通过CYGMA模拟数据系统评估了生长季内气象变量时间聚合对水稻产量机器学习模型性能与可解释性的影响，方法新颖、结论可靠，对农业气象风险识别与决策支持有参考价值，但属学术论文，公共影响范围有限。",[463],{"name":457,"url":454},[26,27,28,465,466],"农业气象","水稻产量预测",{"code":468,"slug":469,"name":470,"path":471,"parent_code":472},"A1","rice","水稻","\u002Ftopic\u002Ffield-crops\u002Frice","A",[474],{"code":56,"slug":189,"name":190,"path":191,"parent_code":192},{"status":195,"locked":14,"version":34,"method":35,"evidence":476},{"candidates":477,"agricultural_evidence":41},[478,483,487,492],{"code":56,"score":479,"supported":41,"title":480,"abstract":481,"tags":482},19,[410],[409,410,266],[27,28],{"code":468,"score":105,"supported":41,"title":484,"abstract":485,"tags":486},[469],[470,469],[],{"code":488,"score":21,"supported":14,"title":489,"abstract":490,"tags":491},"H7",[],[],[465],{"code":39,"score":21,"supported":14,"title":493,"abstract":494,"tags":495},[],[],[26],[497,498],"水稻产量 机器学习 气象","SHAP 水稻 产量预测","水稻产量机器学习气象-4993","10.1038\u002Fs41598-026-75646-2",{"doi":500,"openalex_id":502,"authors":503,"venue":457,"cited_by_count":83,"oa_url":520,"card":521,"direction":170,"ingested_from":92},"W7221657053",[504,506,508,510,512,514,517],{"name":505,"orcid":9},"Noh-Hyun Lee",{"name":507,"orcid":9},"Wonjae Jeong",{"name":509,"orcid":9},"Hyeon-Ji Yang",{"name":511,"orcid":9},"Hana Cho",{"name":513,"orcid":9},"Hyeon-Yeong Ju",{"name":515,"orcid":516},"Toshichika Iizumi","https:\u002F\u002Forcid.org\u002F0000-0002-0611-4637",{"name":518,"orcid":519},"Ki‐Tae Kim","https:\u002F\u002Forcid.org\u002F0000-0002-8249-5283","https:\u002F\u002Fwww.nature.com\u002Farticles\u002Fs41598-026-75646-2_reference.pdf",{"tldr":522,"method":523,"finding":524,"direction":89,"opportunity":525},"评估生长季内天气变量时间聚合粒度对水稻产量机器学习模型预测与可解释性的影响。","用CYGMA模拟潜在产量作响应变量，按等长时段聚合逐日天气，训练多种机器学习模型","适度提高季内时间分辨率可提升性能，产量敏感性集中在特定窗口，温度效应偏早、湿度效应偏晚。","可探索不同作物与气候区下最优聚合粒度，并将窗口特异性气象风险嵌入气候影响评估与决策支持系统。","2026-10-15T23:30:10.483881Z"]