[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"item-3473":3,"related-3473":65},{"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":24,"tags":26,"search_phrases":32,"slug":35,"view_count":36,"doi":37,"paper":38,"created_at":64},3473,"TB-YOLOv8n: an improved YOLOv8n-based model for tip-burn detection in plant factory-grown pakchoi under LED lighting","https:\u002F\u002Fdoi.org\u002F10.3389\u002Ffpls.2026.1933084","Introduction Plant factory-grown pakchoi is prone to tip-burn under high-density cultivation, limited airflow, and rapid growth conditions, which reduces leaf integrity and commercial quality. Image-based detection of small or mild visible tip-burn symptoms remains challenging because the affected regions are usually small, irregularly shaped, low in color contrast, and easily disturbed by red-blue LED lighting, leaf overlap, curled leaf margins, and heart-leaf occlusion. Methods To address these challenges, this study constructed a pakchoi tip-burn image dataset under plant factory conditions and proposed TB-YOLOv8n, a lightweight detection model based on YOLOv8n. AFGCAttention was introduced at the end of the backbone to enhance informative channel responses related to subtle tip-burn regions and suppress redundant background information. GLSA was embedded between the backbone and neck to integrate local texture details with global spatial context under leaf overlap and low-contrast symptom conditions. In addition, a P2-enhanced BiFPN neck was constructed to strengthen cross-scale feature fusion and preserve fine spatial information for small visible tip-burn regions. Results Experimental results showed that TB-YOLOv8n achieved a precision of 89.4%, a recall of 87.6%, an mAP@0.5 of 93.2% ± 0.3%, and an mAP@0.5–0.95 of 59.0% ± 0.3%, which were 4.2, 3.9, 3.4, and 10.0 percentage points higher than those of the original YOLOv8n, respectively. The model contained only 2.2 M parameters, required 7.6 G FLOPs, had a model size of 4.8 MB, and achieved a model-forward inference speed of 138 FPS on the RTX 4090 and 33.0 FPS on the Jetson AGX Orin. Ablation experiments, heatmap visualization, and comparisons with mainstream detection models further confirmed the effectiveness of the proposed improvements. Discussion The results indicate that TB-YOLOv8n achieves a favorable balance among detection accuracy, lightweight structure, and real-time inference capability. The proposed model provides a basis for visible tip-burn monitoring and subsequent production decision support in plant factories.","引言 植物工厂栽培的小白菜在高密度种植、气流受限和快速生长条件下易发生焦边，导致叶片完整性和商品品质下降。基于图像的检测方法在识别微小或轻微可见的焦边症状时仍面临挑战，因为受害区域通常面积小、形状不规则、颜色对比度低，且易受红蓝LED光照、叶片重叠、叶缘卷曲和心叶遮挡的干扰。方法 为解决上述问题，本研究构建了植物工厂条件下的小白菜焦边图像数据集，并提出了TB-YOLOv8n，一种基于YOLOv8n的轻量化检测模型。在主干网络末端引入AFGCAttention，以增强与细微焦边区域相关的信息性通道响应并抑制冗余背景信息。在主干网络与颈部网络之间嵌入GLSA，以在叶片重叠和低对比度症状条件下整合局部纹理细节与全局空间上下文。此外，构建了P2增强的BiFPN颈部网络，以加强跨尺度特征融合并保留小面积可见焦边区域的精细空间信息。结果 实验结果表明，TB-YOLOv8n的精确率为89.4%，召回率为87.6%，mAP@0.5为93.2% ± 0.3%，mAP@0.5–0.95为59.0% ± 0.3%，分别比原始YOLOv8n提高了4.2、3.9、3.4和10.0个百分点。该模型仅含2.2 M参数，需要7.6 G FLOPs，模型大小为4.8 MB，在RTX 4090上的模型前向推理速度为138 FPS，在Jetson AGX Orin上为33.0 FPS。消融实验、热力图可视化以及与主流检测模型的比较进一步证实了所提改进的有效性。讨论 结果表明，TB-YOLOv8n在检测精度、轻量化结构和实时推理能力之间取得了良好平衡。所提模型为植物工厂中可见焦边监测及后续生产决策支持提供了依据。",null,"Frontiers in Plant Science","2026-09-23T00:00:00Z","论文",10,false,78,{"impact":17,"substance":18,"depth":19,"authority":20,"freshness":21,"relevant":22,"comment":23},15,22,18,14,9,1,"面向植物工厂小白菜干烧心的轻量化检测模型，方法新颖、指标扎实且可实时部署，对设施蔬菜智能监测有实用参考价值。",[25],{"name":10,"url":6},[27,28,29,30,31],"智慧农业","农业人工智能","植物工厂","病害检测","小白菜",[33,34],"植物工厂 小白菜 干烧心 检测","TB-YOLOv8n 干烧心 识别","植物工厂小白菜干烧心检测-3473",0,"10.3389\u002Ffpls.2026.1933084",{"doi":37,"openalex_id":39,"authors":40,"venue":10,"cited_by_count":36,"oa_url":55,"card":56,"direction":62,"ingested_from":63},"W7214074852",[41,44,46,48,50,52],{"name":42,"orcid":43},"Peng Wang","https:\u002F\u002Forcid.org\u002F0000-0002-4589-0137",{"name":45,"orcid":9},"Wenwen Hu",{"name":47,"orcid":9},"Xiang Ma",{"name":49,"orcid":9},"Shipu Xu",{"name":51,"orcid":9},"Zhiwen Zhou",{"name":53,"orcid":54},"Yunsheng Wang","https:\u002F\u002Forcid.org\u002F0000-0002-0348-4608","https:\u002F\u002Fwww.frontiersin.org\u002Fjournals\u002Fplant-science\u002Farticles\u002F10.3389\u002Ffpls.2026.1933084\u002Fpdf",{"tldr":57,"method":58,"finding":59,"direction":60,"opportunity":61},"提出轻量模型TB-YOLOv8n，在植物工厂LED光下检测小白菜干烧心。","构建植物工厂小白菜干烧心图像数据集，改进YOLOv8n加入AFGCAttenti","mAP@0.5达93.2%，较YOLOv8n提升3.4个百分点，仅2.2M参数、138FPS。","农业人工智能与决策模型","可延伸至多作物苗期胁迫早期检测，结合时序图像与决策模型实现预警。","智慧农业 \u002F 农业物联网","openalex","2026-09-25T23:30:11.141664Z",{"total":66,"page":22,"page_size":66,"items":67},6,[68,113,141,164,199,237],{"id":69,"title":70,"url":71,"summary":72,"summary_zh":73,"content":9,"source_name":74,"source_url":71,"published_at":75,"category":12,"cover_url":9,"hotness":13,"is_selected":14,"score":76,"score_detail":77,"sources":79,"tags":81,"search_phrases":84,"slug":87,"view_count":36,"doi":88,"paper":89,"created_at":112},2276,"PT-GNN: A physiological topology-aware graph neural network for multimodal early detection of cucumber downy mildew using hyperspectral and chlorophyll fluorescence sensing","https:\u002F\u002Fdoi.org\u002F10.1016\u002Fj.compag.2026.112433","PT-GNN: A physiological topology-aware graph neural network for multimodal early detection of cucumber downy mildew using hyperspectral and chlorophyll fluorescence sensing。Computers and Electronics in Agriculture","PT-GNN：一种生理拓扑感知图神经网络，利用高光谱和叶绿素荧光传感进行黄瓜霜霉病的多模态早期检测。","Computers and Electronics in Agriculture","2026-09-12T00:00:00Z",81,{"impact":19,"substance":18,"depth":19,"authority":20,"freshness":21,"relevant":22,"comment":78},"提出生理拓扑感知图神经网络融合高光谱与叶绿素荧光实现黄瓜霜霉病早期检测，方法新颖且发表于农业信息领域核心期刊，具备较强专业参考价值。",[80],{"name":74,"url":71},[27,28,82,30,83],"黄瓜","高光谱遥感",[85,86],"农业人工智能 高光谱遥感 智慧农业 病害检测","农业人工智能 高光谱遥感","农业人工智能高光谱遥感智慧农业病害检测-2276","10.1016\u002Fj.compag.2026.112433",{"doi":88,"openalex_id":90,"authors":91,"venue":74,"cited_by_count":36,"oa_url":9,"card":106,"direction":110,"ingested_from":63},"W7212310174",[92,95,97,99,101,103],{"name":93,"orcid":94},"Yibin Li","https:\u002F\u002Forcid.org\u002F0000-0002-5906-5074",{"name":96,"orcid":9},"Zonghuan Han",{"name":98,"orcid":9},"Yong Wang",{"name":100,"orcid":9},"Wei Gao",{"name":102,"orcid":9},"Yiding Zhang",{"name":104,"orcid":105},"Lingxian Zhang","https:\u002F\u002Forcid.org\u002F0000-0002-8665-7075",{"tldr":107,"method":108,"finding":109,"direction":110,"opportunity":111},"提出生理拓扑感知图神经网络PT-GNN，融合高光谱与叶绿素荧光实现黄瓜霜霉病早期检测。","构建生理拓扑图神经网络，融合高光谱与叶绿素荧光多模态传感数据。","PT-GNN能有效利用生理拓扑关系，提升黄瓜霜霉病早期检测精度。","农业遥感与作物表型","可探索生理拓扑图构建的通用性，迁移至其他作物病害及多模态传感器融合场景。","2026-09-13T23:30:01.409569Z",{"id":114,"title":115,"url":116,"summary":117,"summary_zh":9,"content":9,"source_name":118,"source_url":9,"published_at":119,"category":12,"cover_url":9,"hotness":13,"is_selected":14,"score":120,"score_detail":121,"sources":123,"tags":125,"search_phrases":128,"slug":131,"view_count":36,"doi":9,"paper":132,"created_at":140},2270,"华中农大实现柑橘黄龙病早期无损诊断 提LLM辅助拉曼光谱相似度融合框架","https:\u002F\u002Fwww.toutiao.com\u002Farticle\u002F7682626369448165922","华中农业大学工学院联合园艺林学学院在Agriculture Communications发表研究,提出LLM辅助的多通道光谱相似度融合框架(LAMSSF),依托少量经qPCR验证的参考样本,实现柑橘HLB早期无损诊断,降低模型对大规模标注数据的依赖。","华中农业大学 Agriculture Communications","2026-09-06T16:00:00Z",82,{"impact":18,"substance":18,"depth":19,"authority":20,"freshness":66,"relevant":22,"comment":122},"华中农大提出LLM辅助拉曼光谱融合框架，实现柑橘黄龙病早期无损诊断，方法新颖且降低标注依赖，具备行业推广价值。",[124],{"name":118,"url":116},[27,28,30,126,127],"柑橘","光谱遥感",[129,130],"农业人工智能 光谱遥感 智慧农业 病害检测","农业人工智能 光谱遥感","农业人工智能光谱遥感智慧农业病害检测-2270",{"doi":9,"openalex_id":9,"authors":133,"venue":9,"cited_by_count":36,"oa_url":9,"card":134,"direction":60,"ingested_from":139},[],{"tldr":135,"method":136,"finding":137,"direction":60,"opportunity":138},"提出LLM辅助多通道光谱相似度融合框架，实现柑橘黄龙病早期无损诊断。","拉曼光谱多通道相似度融合，LLM辅助，少量qPCR验证参考样本。","少量标注样本即可实现HLB早期无损诊断，降低对大规模标注数据依赖。","可探索LLM辅助光谱融合在更多作物病害早期诊断中的迁移与少样本泛化能力。","agent","2026-09-13T00:04:04.913308Z",{"id":142,"title":143,"url":144,"summary":145,"summary_zh":9,"content":146,"source_name":147,"source_url":9,"published_at":148,"category":149,"cover_url":9,"hotness":13,"is_selected":14,"score":150,"score_detail":151,"sources":154,"tags":156,"search_phrases":159,"slug":162,"view_count":36,"doi":9,"paper":9,"created_at":163},2087,"江苏省农业科学院信息所举办\"智说三农·三农论智\"智慧农业学术论坛 浙江浙大与省农科院专家作报告","https:\u002F\u002Fiai.jaas.ac.cn\u002Fgzdt\u002Fart\u002F2026\u002Fart_bc316a767d624b76844e0d6eff013e1a.html","9月2日下午江苏农科院信息所举办第21期\"智说三农·三农论智\"智慧农业学术论坛,邀请浙江大学智能农业装备研究所副所长林涛研究员作《人工智能赋能设施农业生产管控》报告,浙江省农业科学院数字农业研究所副所长孔德栋研究员作《AI驱动型数智农业工厂关键技术与产业应用》报告,聚焦植物工厂与智慧渔业两大方向。","信息所举办第二十一期“智说三农·三农论智”智慧农业学术论坛\n\n文章来源：信息所 发布时间：2026-09-07 09:35 阅读：次\n\n9月2日下午，信息所举办第二十一期“智说三农·三农论智”智慧农业学术论坛，特别邀请浙江大学智能农业装备研究所副所长林涛研究员和浙江省农业科学院数字农业研究所副所长孔德栋研究员作学术报告。信息所相关负责人及院内相关科技人员、研究生参加论坛。\n\n林涛以《人工智能赋能设施农业生产管控》为题，围绕农业AI系统观，从机器人自动化、高频数据采集、作物生长模型及环境调控等维度，阐述了从单点技术突破向垂直场景闭环解决方案转变的发展趋势；重点分享了团队参加第四届荷兰瓦赫宁根大学温室管理挑战赛并获得冠军的实战经验，验证了AI驱动温室生产管控可有效提升投入产出比。报告还介绍了植物工厂内部自主巡检无人机、草莓生长模型与能耗模型耦合、设施农业大模型应用探索等前沿研究成果。\n\n孔德栋以《AI驱动型数智农业工厂关键技术与产业应用》为题作学术报告。报告从数据、算力、算法切入，阐述了 “先工程落地、后科学研究”的产业化思路。内容聚焦植物工厂和智慧渔业两大方向。植物工厂方面，团队在智能光调控、农业专用空调系统、绳驱欠驱动采摘机械手等方面取得突破，通过光生物学研究与第一性原理重构硬件系统，大幅降低了成本，综合成本达到国际领先水平。智慧渔业方面，团队从零搭建标准化、模块化、去中心化的技术体系，构建多模态协同感知矩阵与跨区域云管控平台，实现了高密度养殖稳产降本，并推广至国内多省区及海外多国。\n\n与会青年科研人员围绕报告内容及科研实践中遇到的问题与困惑，与两位专家展开了热烈讨论和深入交流。\n\n“智说三农·三农论智”智慧农业学术论坛是信息所智慧农业创新团队打造的品牌学术平台，将持续邀请国内外相关领域有影响力的专家学者来院做专题报告，旨在促进合作交流、提高学术水平、提升协同创新能力、共促农业产业转型。\n\n![Image 1](https:\u002F\u002Fiai.jaas.ac.cn\u002Fcms_files\u002Ffilemanager\u002F938139336\u002Fpicture\u002F20268\u002FS970562c819074d469b06a90889d2e7fe-650.png)\n\n![Image 2](https:\u002F\u002Fiai.jaas.ac.cn\u002Fcms_files\u002Ffilemanager\u002F938139336\u002Fpicture\u002F20268\u002FSa6004fb881d74fe08594601d196be2cb-650.png)\n\n![Image 3](https:\u002F\u002Fiai.jaas.ac.cn\u002Fcms_files\u002Ffilemanager\u002F938139336\u002Fpicture\u002F20268\u002FSdc404c23ce084a419d20490108cef785-650.png)\n\n![Image 4](https:\u002F\u002Fiai.jaas.ac.cn\u002Fcms_files\u002Ffilemanager\u002F938139336\u002Fpicture\u002F20268\u002FS06d169c25c9f41b1a4a0bd7f2fd2c4c2-670.png)","江苏省农业科学院","2026-09-02T00:00:00Z","报道",66,{"impact":17,"substance":19,"depth":20,"authority":152,"freshness":66,"relevant":22,"comment":153},13,"省级科研机构学术论坛，两位专家报告含AI温室管控、植物工厂降本与智慧渔业落地等实质技术进展，具备一定行业参考价值，但属会议报道、时效略滞后。",[155],{"name":147,"url":144},[27,28,157,158,29],"设施农业","智慧渔业",[160,161],"农业人工智能 智慧农业 智慧渔业 植物工厂","农业人工智能 智慧农业","农业人工智能智慧农业智慧渔业植物工厂-2087","2026-09-11T00:04:20.152178Z",{"id":165,"title":166,"url":167,"summary":168,"summary_zh":169,"content":9,"source_name":170,"source_url":167,"published_at":171,"category":12,"cover_url":9,"hotness":13,"is_selected":14,"score":172,"score_detail":173,"sources":178,"tags":180,"search_phrases":183,"slug":185,"view_count":36,"doi":186,"paper":187,"created_at":198},1601,"Thermal Cut-off, Not Frequency Throttling: Duty-Cycle Scheduling and a Dataset-Integrity Audit for On-Farm Edge Disease Detection in Durian","https:\u002F\u002Fdoi.org\u002F10.1016\u002Fj.atech.2026.102526","Vision-based disease detection is increasingly deployed on low-cost edge hardware in agriculture, yet accuracy reported on curated datasets can diverge sharply from field behaviour, and the thermal behaviour of the host device is rarely characterised at all. This paper reports the deployment of a durian (Durio zibethinus) disease-detection system on a Raspberry Pi 5 and examines three factors that determine its field reliability. First, a dataset-integrity audit combining exact (MD5) and perceptual-hash duplicate detection across the train\u002Fvalidation boundary with per-class instance-to-image tracing identifies a train\u002Fvalidation leak in the early-stage dataset: one byte-identical pair and 117 near-duplicate pairs, resolving to 22 source components under a provenance-based grouping, and traced to an augmentation-before-split workflow. Re-evaluation on a validation partition with the implicated images removed shows the leak inflated the affected class by 35.0% in AP@0.5 and aggregate mAP@0.5 by 9.5%. Second, eleven training runs across three architecture families — including a transformer-based, anchor-free detector with three times the parameter count — leave two classes at or near zero average precision, while bootstrap resampling of the validation set, applied to the two original-submission checkpoints, shows 93% of the nominal difference between them to originate in a single validation instance. Retraining on a source-level leak-free partition recovers one of those classes from 0.000 to 0.399, which is consistent with its zero having been a property of the validation sample rather than of the class, and motivates a candidate three-way distinction between feature-space, label-space and evaluation-space failure. All detection figures are computed on a 126-image validation partition holding one and two instances in two classes; they are not estimates of general detection performance, and no independent source-level test partition was available. Third, a duty-cycle thermal characterisation across five configurations, replicated four weeks apart, shows that the operative failure mode on this platform is not throughput loss but protective shutdown: continuous inference triggered 99 and 108 software thermal cut-offs across the two rounds and a 15 s sleep interval 41 and 3, whereas a 30 s interval recorded none in either round under the tested conditions, with 45 s retaining the larger thermal margin. Throttling itself cost only 2.6% of median inference latency (419.0 to 408.0 ms), so what duty-cycle scheduling protects is thermal margin and uninterrupted operation rather than frame rate. Because each cut-off suspends inference for a fixed interval, continuous operation delivers 95.5% rather than the 100% coverage its schedule nominally promises: for an operator walking an inspection round, 8.2 min of the 183 min run examine no tree, with no indication of when. Direct power measurement at the system-on-chip domain shows the exchange is not free in the other direction either: duty cycling lowers mean SoC power by 28.2% but raises SoC energy per inference by 39.4%, because the idle floor is paid throughout the sleep interval. The audit protocol, the failure taxonomy and the cut-off-based thermal criterion are stated as procedures that could be applied to other resource-constrained agricultural deployments; their generality is not tested here, since the evidence is one durian dataset, one Raspberry Pi 5 and one software stack.","基于视觉的病害检测正越来越多地部署于农业中的低成本边缘硬件上，然而在精选数据集上报告的准确率与田间实际表现可能存在显著差异，且宿主设备的热行为几乎从未被表征。本文报告了在树莓派5（Raspberry Pi 5）上部署榴莲（Durio zibethinus）病害检测系统的情况，并考察了决定其田间可靠性的三个因素。首先，一项数据集完整性审计结合了训练\u002F验证边界上的精确（MD5）与感知哈希重复检测，以及逐类实例到图像的追踪，识别出早期阶段数据集中的训练\u002F验证泄漏：一对字节完全相同的图像和117对近似重复图像，在基于来源的分组下归结为22个源组件，并追溯至先增强后分割的工作流程。在移除相关图像后的验证分区上重新评估显示，该泄漏使受影响类别在AP@0.5上虚增了35.0%，并使总体mAP@0.5虚增了9.5%。其次，跨三个架构家族的十一次训练运行——包括一个基于Transformer、无锚框、参数量为其三倍的检测器——使两个类别的平均精度保持为零或接近零，而对验证集进行自助重采样（bootstrap resampling），应用于两个原始提交检查点，显示它们之间名义差异的93%源于单个验证实例。在无泄漏的源级分区上重新训练，使其中一个类别从0.000恢复至0.399，这与该零值源于验证样本特性而非类别本身属性相一致，并促使提出特征空间、标签空间与评估空间失败之间的候选三重区分。所有检测数据均在包含两个类别中分别含一个和两个实例的126幅图像验证分区上计算；它们并非一般检测性能的估计，且不存在独立的源级测试分区。第三，跨五种配置、相隔四周重复两次的占空比热表征显示，该平台上的操作性失效模式并非吞吐量损失而是保护性关机：连续推理在两轮中分别触发了99次和108次软件热切断，15秒睡眠间隔分别为41次和3次，而在测试条件下30秒间隔在两轮中均未记录到热切断，45秒则保留了更大的热裕度。节流本身仅造成中位推理延迟2.6%的损失。","Smart Agricultural Technology","2026-09-01T00:00:00Z",70,{"impact":174,"substance":18,"depth":175,"authority":174,"freshness":176,"relevant":22,"comment":177},12,20,4,"论文深入剖析了边缘设备在田间病害检测中的热失控与数据集泄漏问题，方法严谨，对实际部署有重要参考价值。",[179],{"name":170,"url":167},[27,28,181,30,182],"边缘计算","榴莲",[184,161],"农业人工智能 智慧农业 病害检测 边缘计算","农业人工智能智慧农业病害检测边缘计算-1601","10.1016\u002Fj.atech.2026.102526",{"doi":186,"openalex_id":188,"authors":189,"venue":170,"cited_by_count":36,"oa_url":167,"card":193,"direction":60,"ingested_from":63},"W7207768578",[190],{"name":191,"orcid":192},"Lin Ding Shan","https:\u002F\u002Forcid.org\u002F0009-0009-6031-8479",{"tldr":194,"method":195,"finding":196,"direction":60,"opportunity":197},"研究榴莲病害边缘检测系统的数据集泄漏与热关断问题，提出占空比调度方案。","数据集完整性审计（MD5\u002F感知哈希）、重训练、五配置占空比热特性测试。","数据泄漏虚增AP 35%，热关断而非降频是主要失效模式，30秒间隔可避免。","可探索边缘AI部署中热管理与数据泄漏的通用评估框架，及占空比调度在其他作物检测的适用性。","2026-09-04T23:30:04.697246Z",{"id":200,"title":201,"url":202,"summary":203,"summary_zh":204,"content":9,"source_name":170,"source_url":202,"published_at":171,"category":12,"cover_url":9,"hotness":13,"is_selected":14,"score":205,"score_detail":206,"sources":208,"tags":210,"search_phrases":213,"slug":215,"view_count":22,"doi":216,"paper":217,"created_at":236},1599,"Continual Learning Model Evaluation for Hyperspectral-Based Tomato Bacterial Spot Detection in Greenhouse","https:\u002F\u002Fdoi.org\u002F10.1016\u002Fj.atech.2026.102535","Detection of tomato bacterial spot (TBS) at canopy level is critical in greenhouse seedling production, yet most previous hyperspectral studies have been conducted at leaf scale under tightly controlled conditions. This study evaluated a domain-incremental continual-learning framework for automated canopy-scale TBS detection using hyperspectral sensing across five greenhouse experiments. Three spectral feature sets were evaluated, and Full-memory and Bounded-memory Gradient Episodic Memory multilayer perceptrons (GEM-MLP) were compared with Sequential and Joint-training MLP models under two conditions: a future unseen-experiment generalization test (Scenario 1) and a same-distribution benchmark (Scenario 2). Genetic-algorithm feature selection reduced input dimensionality but did not identify a single feature subset that was stable across all folds and scenarios. Under Scenario 1, Full-memory GEM-MLP using the fixed 18-VI feature set achieved 58.1% asymptomatic recall and 81.2% symptomatic recall. Its clearest advantage was knowledge retention: backward transfer (BWT) was -1.4%, compared with -7.2% for Sequential MLP, while forward transfer was similar (7.4% versus 7.0%). Under Scenario 2, Full-memory GEM-MLP achieved 46.5% asymptomatic recall and 86.2% symptomatic recall, while Joint-training MLP was highly competitive and reached the highest mean asymptomatic recall (53.4%). Full-memory GEM-MLP again showed better retention, with BWT of +1.8% compared with -3.2% for Sequential MLP. These results indicate that GEM did not provide universally superior classification performance, but it consistently reduced forgetting during sequential learning. The persistent difficulty of asymptomatic detection, even when all experimental domains were represented during model development, further suggests that early canopy-scale disease detection is limited by factors beyond cross-experiment distribution shift.","在温室育苗生产中，冠层尺度上的番茄细菌性斑点病（TBS）检测至关重要，然而以往大多数高光谱研究都是在严格控制条件下于叶片尺度进行的。本研究评估了一种基于域增量持续学习框架的自动化冠层尺度TBS检测方法，该方法利用跨五个温室实验的高光谱传感数据。研究评估了三组光谱特征集，并在两种条件下将全记忆与有界记忆梯度情景记忆多层感知器（GEM-MLP）与顺序训练和联合训练MLP模型进行了比较：一种是对未来未见实验的泛化测试（情景1），另一种是同分布基准测试（情景2）。遗传算法特征选择降低了输入维度，但未能识别出在所有折和情景下均稳定的单一特征子集。在情景1下，使用固定18-VI特征集的全记忆GEM-MLP实现了58.1%的无症状召回率和81.2%的有症状召回率。其最显著的优势在于知识保持能力：后向迁移（BWT）为-1.4%，而顺序MLP为-7.2%，同时前向迁移相近（7.4%对7.0%）。在情景2下，全记忆GEM-MLP实现了46.5%的无症状召回率和86.2%的有症状召回率，而联合训练MLP表现极具竞争力，达到了最高的平均无症状召回率（53.4%）。全记忆GEM-MLP再次展现出更好的保持能力，其BWT为+1.8%，而顺序MLP为-3.2%。这些结果表明，GEM并未提供普遍优越的分类性能，但在顺序学习过程中始终减少了遗忘。即使在模型开发阶段涵盖了所有实验域，无症状检测仍持续存在困难，这进一步表明早期冠层尺度病害检测受到跨实验分布偏移之外因素的限制。",64,{"impact":174,"substance":19,"depth":19,"authority":174,"freshness":176,"relevant":22,"comment":207},"研究评估持续学习模型在温室番茄细菌性斑点病高光谱检测中的应用，虽为细分领域进展，但方法新颖且数据详实，对农业AI落地有参考价值。",[209],{"name":170,"url":202},[27,28,211,212,30],"番茄","遥感",[214,161],"农业人工智能 智慧农业 病害检测 番茄","农业人工智能智慧农业病害检测番茄-1599","10.1016\u002Fj.atech.2026.102535",{"doi":216,"openalex_id":218,"authors":219,"venue":170,"cited_by_count":36,"oa_url":202,"card":231,"direction":62,"ingested_from":63},"W7207729517",[220,222,224,227,229],{"name":221,"orcid":9},"Gaoshoutong Si",{"name":223,"orcid":9},"Peter Ling",{"name":225,"orcid":226},"Anna L. Testen","https:\u002F\u002Forcid.org\u002F0000-0002-6504-9599",{"name":228,"orcid":9},"Heping Zhu",{"name":230,"orcid":9},"Hongyoung Jeon",{"tldr":232,"method":233,"finding":234,"direction":62,"opportunity":235},"评估持续学习框架用于温室高光谱番茄细菌性斑点病冠层检测。","域增量持续学习，GEM-MLP与顺序\u002F联合训练MLP对比，遗传算法特征选择。","GEM-MLP减少遗忘但分类性能非最优，无症状检测仍困难。","可探索结合多模态数据或更先进持续学习方法，提升早期无症状病害检测精度。","2026-09-04T23:30:04.578607Z",{"id":238,"title":239,"url":240,"summary":241,"summary_zh":242,"content":9,"source_name":243,"source_url":240,"published_at":148,"category":12,"cover_url":9,"hotness":13,"is_selected":14,"score":244,"score_detail":245,"sources":248,"tags":250,"search_phrases":252,"slug":254,"view_count":36,"doi":255,"paper":256,"created_at":268},1430,"Hyperspectral Imaging for Disease Detection in Wheat Crops: A Comprehensive Survey","https:\u002F\u002Fdoi.org\u002F10.18805\u002Fijare.a-6533","Hyperspectral imaging (HSI) has emerged as a game-changing technology for precision agriculture because it captures detailed spectral signatures over hundreds of finely-resolved, continuous wavelength bands. This feature makes possible a detailed description of species-specific plant physiological states to enable early, precise and non-destructive identification of diseases in crops like wheat. With the rising demands worldwide for food security and healthy cultivation, the application of HSI in agro-diagnostics has expanded considerably. This paper provides a comprehensive and state-of-the-art review of hyperspectral imaging strategies devoted to wheat disease discrimination. It essentially introduces the theoretical backgrounds of HSI, such as data acquisition, preprocessing and spectral-spatial feature extraction. The survey further details a plethora of analytical approaches from classical statistical learning methods to deep learning and their utility in modelling HSI data. In particular, a comparative study of these methods is carried out and discussed for distinguishing healthy and diseased wheat, as well as different types of diseases under diverse environmental conditions. The paper also considers the benchmark datasets, sensors and platforms (field\u002Flaboratory-based) for which there are practical limitations. Key challenges are addressed, such as data dimensionality, variability as a consequence of atmospheric conditions and the requirement for online implementation in open-field scenarios. A set of suggestions for applying HSI in other domains has been proposed, such as the fusion of HSI with additional remote sensing information, introduction of compact models to edge devices and construction of a scalable and economical solution of HSI in commercial agriculture.","高光谱成像（HSI）已成为精准农业领域的一项变革性技术，因为它能够在数百个精细分辨、连续分布的波段上捕获详细的光谱特征。这一特性使得对物种特异性植物生理状态进行细致描述成为可能，从而能够对小麦等作物病害实现早期、精准且非破坏性的识别。随着全球对粮食安全和健康栽培需求的日益增长，HSI在农业诊断中的应用已显著扩展。本文对用于小麦病害区分的高光谱成像策略进行了全面且前沿的综述。文章首先介绍了HSI的理论基础，包括数据采集、预处理及光谱-空间特征提取。随后，综述详细探讨了从经典统计学习方法到深度学习的众多分析方法及其在HSI数据建模中的适用性。特别地，文中对这些方法进行了比较研究，并讨论了其在区分健康与患病小麦，以及在不同环境条件下识别不同类型病害方面的表现。本文还考虑了基准数据集、传感器及平台（田间\u002F实验室）在实际应用中存在的局限性。针对数据维度、大气条件引起的变异性以及开放田间场景下在线实施的需求等关键挑战进行了阐述。最后，提出了将HSI应用于其他领域的一系列建议，例如将HSI与额外遥感信息融合、将紧凑模型引入边缘设备，以及构建在商业农业中可扩展且经济可行的HSI解决方案。","Indian Journal of Agricultural Research",77,{"impact":19,"substance":18,"depth":19,"authority":174,"freshness":246,"relevant":22,"comment":247},7,"高光谱成像用于小麦病害检测的综述，系统性强，方法比较全面，对智慧农业病害诊断有参考价值。",[249],{"name":243,"url":240},[27,28,251,212,30],"小麦",[253,161],"农业人工智能 智慧农业 病害检测 小麦","农业人工智能智慧农业病害检测小麦-1430","10.18805\u002Fijare.a-6533",{"doi":255,"openalex_id":257,"authors":258,"venue":243,"cited_by_count":36,"oa_url":240,"card":263,"direction":110,"ingested_from":63},"W7205005591",[259,261],{"name":260,"orcid":9},"M. Deepak",{"name":262,"orcid":9},"Gopal K. Shyam",{"tldr":264,"method":265,"finding":266,"direction":110,"opportunity":267},"综述高光谱成像在小麦病害检测中的应用，涵盖方法、挑战与未来方向。","综述数据采集、预处理、特征提取及统计学习与深度学习方法。","高光谱成像可早期、无损识别小麦病害，但面临维度高、环境变化和实时性挑战。","可探索高光谱与多源遥感融合、轻量化模型部署于边缘设备，以及低成本商业化方案。","2026-09-02T23:30:51.057495Z"]