[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"item-4299":3,"related-4299":76},{"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":18,"tags":20,"primary_topic":9,"secondary_topics":25,"classification":26,"search_phrases":49,"slug":52,"view_count":44,"doi":53,"paper":54,"created_at":75},4299,"A Survey of Computer Vision Applications in Industrial Environments","https:\u002F\u002Fdoi.org\u002F10.1080\u002F08839514.2026.2739540","In recent years, Computer Vision (CV) has evolved rapidly, enabling enhanced automation, safety, inspection, and logistics across manufacturing, construction, and robotics. This systematic review surveys current computer vision applications in industrial environments, spanning factory defect inspection, construction-site safety, warehouse inventory tracking, and industrial robotics. From more than 8,700 initially identified studies, we selected 146 key papers representing cross-domain CV applications. The survey is organized into four thematic areas: visual inspection and defect detection, worker safety and monitoring, inventory and logistics tracking, and robotics and automation. For each area, we compare state-of-the-art approaches, datasets, and deployment settings while highlighting the factors that influence their performance under different operating conditions. For example, YOLO-based detectors dominate real-time edge deployment because of their high throughput, whereas transformer-based models often struggle to scale in resource-constrained environments despite their superior feature extraction capabilities. We also identify critical research gaps. Although manufacturing and construction benefit from numerous public datasets for quality inspection and safety monitoring, warehousing and robotics continue to lack large-scale, real-world benchmark datasets. Finally, we discuss emerging research directions, including domain adaptation, edge AI, and explainable artificial intelligence (XAI), to support the development of more robust, scalable, and balanced computer vision solutions for industrial applications.","近年来，计算机视觉（Computer Vision, CV）发展迅速，推动了制造业、建筑业和机器人领域的自动化、安全、检测和物流水平提升。本系统综述考察了工业环境中的当前计算机视觉应用，涵盖工厂缺陷检测、施工现场安全、仓库库存跟踪和工业机器人。在最初识别的8700余项研究中，我们选取了146篇代表性关键论文，涉及跨领域计算机视觉应用。本综述按四个主题领域组织：视觉检测与缺陷识别、工人安全与监控、库存与物流跟踪，以及机器人与自动化。针对每个领域，我们比较了最先进的方法、数据集和部署环境，同时重点分析了影响其在不同运行条件下性能的因素。例如，基于YOLO的检测器因其高吞吐量在实时边缘部署中占据主导地位，而基于Transformer的模型尽管具有更优的特征提取能力，却往往难以在资源受限环境中扩展。我们还指出了关键的研究空白。尽管制造业和建筑业在质量检测和安全监控方面受益于大量公开数据集，但仓储和机器人领域仍缺乏大规模真实世界基准数据集。最后，我们讨论了新兴研究方向，包括域适应、边缘人工智能和可解释人工智能（XAI），以支持开发更稳健、可扩展且均衡的工业应用计算机视觉解决方案。",null,"Applied Artificial Intelligence","2026-10-04T00:00:00Z","论文",10,false,0,{"impact":15,"substance":15,"depth":15,"authority":15,"freshness":15,"relevant":15,"comment":17},"该文为工业环境计算机视觉综述，与三农、农业信息化、智慧农业无直接关联，不建议进入每日精选。",[19],{"name":10,"url":6},[21,22,23,24],"计算机视觉","边缘计算","工业人工智能","智能制造",[],{"status":27,"locked":14,"version":28,"method":29,"evidence":30},"needs_review","2026-10-22.2","rules",{"candidates":31,"agricultural_evidence":14,"reason":48},[32,42],{"code":33,"score":34,"supported":35,"title":36,"abstract":38,"tags":41},"I2",18,true,[37],"computer vision",[21,39,37,40],"artificial intelligence","YOLO",[21],{"code":43,"score":44,"supported":14,"title":45,"abstract":46,"tags":47},"I5",1,[],[],[22],"农业相关性需要复核",[50,51],"工业环境 计算机视觉","YOLO 边缘部署 缺陷检测","工业环境计算机视觉-4299","10.1080\u002F08839514.2026.2739540",{"doi":53,"openalex_id":55,"authors":56,"venue":10,"cited_by_count":15,"oa_url":66,"card":67,"direction":73,"ingested_from":74},"W7219574412",[57,60,63],{"name":58,"orcid":59},"Abdullah Albanyan","https:\u002F\u002Forcid.org\u002F0000-0002-2371-9400",{"name":61,"orcid":62},"Reem Abdel‐Salam","https:\u002F\u002Forcid.org\u002F0000-0002-4594-211X",{"name":64,"orcid":65},"Abdulrahman Alabduljabbar","https:\u002F\u002Forcid.org\u002F0000-0002-1066-6725","https:\u002F\u002Fwww.tandfonline.com\u002Fdoi\u002Fpdf\u002F10.1080\u002F08839514.2026.2739540?needAccess=true",{"tldr":68,"method":69,"finding":70,"direction":71,"opportunity":72},"系统综述工业环境中计算机视觉在检测、安全、物流与机器人四领域的应用现状与挑战。","系统综述8700余篇文献，精选146篇，对比方法、数据集与部署条件。","YOLO适合边缘实时部署，仓储与机器人领域缺乏大规模真实基准数据集。","农业人工智能与决策模型","可借鉴工业CV的域适应与边缘AI方法，构建农业场景专用基准数据集与轻量模型。","智慧农业 \u002F 农业物联网","openalex","2026-10-06T23:30:09.467741Z",{"total":77,"page":44,"page_size":77,"items":78},6,[79,145,199,260,303,383],{"id":80,"title":81,"url":82,"summary":83,"summary_zh":84,"content":9,"source_name":85,"source_url":82,"published_at":86,"category":12,"cover_url":9,"hotness":13,"is_selected":14,"score":87,"score_detail":88,"sources":93,"tags":95,"primary_topic":100,"secondary_topics":105,"classification":106,"search_phrases":125,"slug":128,"view_count":15,"doi":129,"paper":130,"created_at":144},5587,"Calibration-Shift Inference for Computer Vision Agriculture Yield Models Under Retrofit Sensor Replacement: An Artificial Intelligence Post-Deployment Recalibration Guarantee","https:\u002F\u002Fdoi.org\u002F10.5281\u002Fzenodo.23544195","Computer vision yield models deployed in agricultural fields are routinely paired with retrofit sensor packages that replace or supplement the original imaging hardware, and the resulting calibration shift can silently invalidate the model's predictive guarantees. The analysis characterises the operating characteristics of the design under a declared variance structure with between-cluster standard deviation 6.0, residual standard deviation 4.5 and intraclass correlation 0.64, using 2000 replications per design cell across cluster counts of 8, 16, 24 and 32 and effect sizes of 3, 5 and 8 percentage points. The design attains 80% power at a 5 percentage point effect with 54 clusters. The paper states a formal guarantee: under stated assumptions on the stability of the calibration map, the before-after-control-impact estimator yields a consistent and asymptotically normal estimate of the calibration shift, and the associated confidence interval covers the true shift at the nominal rate. The contribution is a design-theoretic account of how retrofit-induced calibration shift can be detected and bounded without observational data, together with the sample-size requirements that follow.","部署在农田中的计算机视觉产量模型通常与改装传感器组件配合使用，这些组件用于替换或补充原始成像硬件，由此产生的校准偏移可能悄然使模型的预测保证失效。该分析在声明的方差结构下刻画了设计的工作特性，其中组间标准差为6.0、残差标准差为4.5、组内相关系数为0.64，每个设计单元采用2000次重复，聚类数分别为8、16、24和32，效应量为3、5和8个百分点。该设计在54个聚类、5个百分点效应下达到80%的检验力。论文给出了一项形式化保证：在关于校准映射稳定性的既定假设下，前后对照影响（before-after-control-impact）估计量给出校准偏移的一致且渐近正态的估计，且相关置信区间以名义覆盖率覆盖真实偏移。其贡献在于从设计理论角度说明了如何在没有观测数据的情况下检测并界定改装引起的校准偏移，以及随之而来的样本量要求。","Zenodo (CERN European Organization for Nuclear Research)","2026-10-21T00:00:00Z",65,{"impact":89,"substance":34,"depth":90,"authority":91,"freshness":13,"relevant":44,"comment":92},8,17,12,"方法学扎实、统计设计严谨的预印本论文，但偏理论设计层面，缺乏田间实证数据，产业影响有限，适合作为技术参考而非每日精选头条。",[94],{"name":85,"url":82},[96,97,21,98,99],"智慧农业","农业人工智能","产量预测","传感器校准",{"code":33,"slug":101,"name":102,"path":103,"parent_code":104},"agricultural-ai","农业AI与大模型","\u002Ftopic\u002Fdigital-agriculture\u002Fagricultural-ai","I",[],{"status":107,"locked":14,"version":28,"method":29,"evidence":108},"classified",{"candidates":109,"agricultural_evidence":35},[110,115,120],{"code":33,"score":111,"supported":35,"title":112,"abstract":113,"tags":114},24,[39,37],[21,37],[97,21],{"code":116,"score":44,"supported":14,"title":117,"abstract":118,"tags":119},"I1",[],[],[96],{"code":121,"score":44,"supported":14,"title":122,"abstract":123,"tags":124},"I3",[],[],[98],[126,127],"计算机视觉 产量预测 传感器校准","农业人工智能 传感器校准 计算机视觉 产量预测","计算机视觉产量预测传感器校准-5587","10.5281\u002Fzenodo.23544195",{"doi":129,"openalex_id":131,"authors":132,"venue":85,"cited_by_count":15,"oa_url":82,"card":139,"direction":71,"ingested_from":74},"W7223104380",[133,135,137],{"name":134,"orcid":9},"Tigist Alemu Gebremariam",{"name":136,"orcid":9},"Hawi Tesfaye Desta",{"name":138,"orcid":9},"Mulugeta Bekele Wolde",{"tldr":140,"method":141,"finding":142,"direction":71,"opportunity":143},"提出改造传感器替换后视觉产量模型校准漂移的检测与置信区间保证方法。","用前后对照影响估计量，基于方差结构做2000次重复的模拟设计分析。","5个百分点效应下54个集群可达80%检验力，估计量一致且渐近正态。","可将该设计理论扩展到多传感器异构替换与在线校准监测，填补真实田间验证空白。","2026-10-22T11:21:01.554995Z",{"id":146,"title":147,"url":148,"summary":149,"summary_zh":150,"content":9,"source_name":85,"source_url":148,"published_at":86,"category":12,"cover_url":9,"hotness":13,"is_selected":14,"score":151,"score_detail":152,"sources":156,"tags":158,"primary_topic":161,"secondary_topics":162,"classification":163,"search_phrases":179,"slug":182,"view_count":15,"doi":183,"paper":184,"created_at":198},5584,"Distributionally-Robust Evaluation of Computer Vision Agriculture Disease Classifiers Under Climate-Driven Pathogen Range Shifts: An Artificial Intelligence Worst-Case Generalization Bound","https:\u002F\u002Fdoi.org\u002F10.5281\u002Fzenodo.23544150","Computer vision classifiers for crop disease are evaluated almost exclusively on held-out samples drawn from the same distribution as their training data, yet climate-driven pathogen range shifts mean that deployment environments are systematically out of distribution. The framework treats the deployment distribution as an adversarial perturbation of the training distribution within a Wasserstein ball whose radius is calibrated to observed climate displacement. Using NASA POWER monthly reanalysis for Harare, Zimbabwe (2000-2023, 312 monthly observations), the analysis estimates a mean air temperature of 18.914 C, a standard deviation of 3.107 C, and a warming trend of 0.2001 C per decade. These estimates anchor the radius of the ambiguity set and the resulting bound. The bound decomposes worst-case risk into an empirical risk term, a Wasserstein radius term scaled by the Lipschitz constant of the classifier, and a concentration term. Under plausible Lipschitz constants for convolutional backbones, the radius term dominates the empirical risk term once the climate displacement exceeds a threshold that the Harare record already approaches. The contribution is a formal evaluation criterion that replaces the i. i. d. assumption with a climate-calibrated ambiguity set, together with the analytical machinery needed to compute it.","作物病害的计算机视觉分类器几乎完全在与训练数据同分布的留出样本上进行评估，然而气候驱动的病原体分布范围变化意味着部署环境系统性地处于分布之外。该框架将部署分布视为训练分布在Wasserstein球内的对抗性扰动，球的半径根据观测到的气候位移进行校准。利用NASA POWER对津巴布韦哈拉雷的月度再分析数据（2000—2023年，312个月度观测值），分析估计平均气温为18.914°C，标准差为3.107°C，每十年升温趋势为0.2001°C。这些估计值锚定了模糊集的半径及由此得到的界。该界将最坏情况风险分解为经验风险项、由分类器Lipschitz常数缩放后的Wasserstein半径项以及集中项。在卷积骨干网络合理的Lipschitz常数下，一旦气候位移超过哈拉雷记录已接近的阈值，半径项便主导经验风险项。本文的贡献在于提出了一种形式化评估准则，以气候校准的模糊集替代独立同分布（i.i.d.）假设，并提供了计算该准则所需的分析工具。",75,{"impact":91,"substance":153,"depth":34,"authority":154,"freshness":13,"relevant":44,"comment":155},22,13,"方法新颖、数据具体，但属方法论预印论文，产业影响有限，适合作为专业精选而非头条。",[157],{"name":85,"url":148},[96,97,21,159,160],"作物病害","气候变化",{"code":33,"slug":101,"name":102,"path":103,"parent_code":104},[],{"status":107,"locked":14,"version":28,"method":29,"evidence":164},{"candidates":165,"agricultural_evidence":35},[166,170,175],{"code":33,"score":111,"supported":35,"title":167,"abstract":168,"tags":169},[39,37],[21,37],[97,21],{"code":171,"score":44,"supported":14,"title":172,"abstract":173,"tags":174},"H7",[],[],[160],{"code":116,"score":44,"supported":14,"title":176,"abstract":177,"tags":178},[],[],[96],[180,181],"NASA POWER 哈拉雷 气温","Wasserstein 分布鲁棒 病害分类","NASAPOWER哈拉雷气温-5584","10.5281\u002Fzenodo.23544150",{"doi":183,"openalex_id":185,"authors":186,"venue":85,"cited_by_count":15,"oa_url":148,"card":193,"direction":71,"ingested_from":74},"W7223087892",[187,189,191],{"name":188,"orcid":9},"Kossi Afeli",{"name":190,"orcid":9},"Akouvi Amegashie",{"name":192,"orcid":9},"Komlan Sélom Dossou",{"tldr":194,"method":195,"finding":196,"direction":71,"opportunity":197},"提出气候校准的分布鲁棒评估准则，替代i.i.d.假设评价作物病害视觉分类器。","用Wasserstein球建模部署分布偏移，以NASA POWER哈拉雷气温数据","气候位移超阈值后，半径项主导经验风险，现有评估会高估模型泛化能力。","可将气候校准的鲁棒评估扩展到多作物多区域，并研究适应分布偏移的鲁棒训练方法。","2026-10-22T11:21:00.036937Z",{"id":200,"title":201,"url":202,"summary":203,"summary_zh":9,"content":9,"source_name":204,"source_url":9,"published_at":205,"category":12,"cover_url":9,"hotness":13,"is_selected":14,"score":206,"score_detail":207,"sources":213,"tags":215,"primary_topic":219,"secondary_topics":225,"classification":226,"search_phrases":247,"slug":250,"view_count":15,"doi":9,"paper":251,"created_at":259},5533,"基于低成本遥操作机械臂模仿学习的草莓果柄定点采摘","https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.22765","该研究构建视觉运动模仿学习流程，统一感知、运动规划与接触操作，从人工遥操作演示训练基于ResNet-50骨干的动作分块Transformer策略，并在Jetson AGX Orin边缘平台以15赫兹自主部署。策略在130条演示、十万梯度步内损失由2.84降至0.091，数据集与模型权重已开源。预印本，已被IROS农业机器人研讨会接收。","arXiv 预印本 2026-10-16","2026-10-16T00:00:00Z",68,{"impact":208,"substance":209,"depth":34,"authority":210,"freshness":211,"relevant":44,"comment":212},16,21,9,4,"方法新颖、数据与开源材料扎实，但属预印本且已过一周，适合作为细分技术进展收录而非头条精选。",[214],{"name":204,"url":202},[97,216,22,217,218],"农业机器人","模仿学习","草莓采摘",{"code":220,"slug":221,"name":222,"path":223,"parent_code":224},"B1","vegetables","蔬菜","\u002Ftopic\u002Fhorticulture\u002Fvegetables","B",[],{"status":107,"locked":14,"version":28,"method":29,"evidence":227},{"candidates":228,"agricultural_evidence":35},[229,234,239,243],{"code":220,"score":89,"supported":35,"title":230,"abstract":232,"tags":233},[231],"草莓",[],[],{"code":235,"score":211,"supported":35,"title":236,"abstract":237,"tags":238},"F6",[],[216],[216],{"code":33,"score":44,"supported":14,"title":240,"abstract":241,"tags":242},[],[],[97],{"code":43,"score":44,"supported":14,"title":244,"abstract":245,"tags":246},[],[],[22],[248,249],"草莓果柄 定点采摘 机械臂","Jetson AGX Orin 采摘 模仿学习","草莓果柄定点采摘机械臂-5533",{"doi":9,"openalex_id":9,"authors":252,"venue":9,"cited_by_count":15,"oa_url":9,"card":253,"direction":71,"ingested_from":258},[],{"tldr":254,"method":255,"finding":256,"direction":71,"opportunity":257},"用低成本遥操作机械臂采集演示，模仿学习实现草莓果柄定点采摘。","ResNet-50骨干的动作分块Transformer策略，Jetson边缘部署","130条演示训练后损失降至0.091，策略可15赫兹自主采摘。","可探索小样本跨品种泛化与失败恢复，降低演示采集成本。","agent","2026-10-22T00:07:09.891422Z",{"id":261,"title":262,"url":263,"summary":264,"summary_zh":265,"content":9,"source_name":266,"source_url":263,"published_at":267,"category":12,"cover_url":9,"hotness":13,"is_selected":14,"score":15,"score_detail":268,"sources":270,"tags":272,"primary_topic":9,"secondary_topics":274,"classification":275,"search_phrases":284,"slug":287,"view_count":15,"doi":288,"paper":289,"created_at":302},5494,"YOLO-Based Real-Time Object Detection Using Computer Vision for Robotics","https:\u002F\u002Fdoi.org\u002F10.64388\u002Firev10i4-1724117","Robots operating in dynamic environments require rapid and reliable perception to identify people, tools, obstacles, and other objects before making movement or manipulation decisions. Conventional vision pipelines may require separate stages for region proposal, classification, and localization, which can increase computational cost and latency. You Only Look Once (YOLO) formulates object detection as a single-network prediction task and has become a practical family of detectors for real-time computer-vision applications. This paper presents a research-oriented framework for integrating a lightweight YOLO detector with a robot-mounted camera. The framework covers image acquisition, preprocessing, object detection, confidence filtering, coordinate interpretation, and communication with a robot controller. It also discusses model selection, deployment constraints, evaluation metrics, safety considerations, and a reproducible testing plan. Rather than reporting unperformed experiments, the paper defines how a prototype should be evaluated using precision, recall, mean average precision, inference latency, frame rate, and end-to-end response time. The analysis indicates that compact YOLO variants are suitable starting points for resource-constrained robotic platforms, while reliable navigation requires additional depth or distance sensing and a safety controller. The proposed approach can support obstacle awareness, object following, pick-and-place assistance, and indoor mobile-robot perception.","在动态环境中运行的机器人需要快速且可靠的感知能力，以便在做出移动或操作决策之前识别人、工具、障碍物和其他物体。传统视觉流水线可能需要区域提议、分类和定位等独立阶段，这会增加计算成本和延迟。You Only Look Once（YOLO）将目标检测表述为单网络预测任务，已成为实时计算机视觉应用中实用的一类检测器。本文提出了一个面向研究的框架，用于将轻量级YOLO检测器与机器人搭载的相机集成。该框架涵盖图像采集、预处理、目标检测、置信度过滤、坐标解释以及与机器人控制器的通信。文中还讨论了模型选择、部署约束、评估指标、安全考虑以及可复现的测试方案。本文不报告未进行的实验，而是定义了应如何使用精确率、召回率、平均精度均值、推理延迟、帧率和端到端响应时间来评估原型。分析表明，紧凑型YOLO变体适合作为资源受限机器人平台的起点，而可靠导航还需要额外的深度或距离传感以及安全控制器。所提出的方法可支持障碍物感知、目标跟随、拾取与放置辅助以及室内移动机器人感知。","Iconic Research and Engineering Journals","2026-10-19T00:00:00Z",{"impact":15,"substance":15,"depth":15,"authority":15,"freshness":15,"relevant":15,"comment":269},"该文为通用机器人视觉论文，未涉及农业场景，与三农及农业信息化无直接关联，不建议进入每日精选。",[271],{"name":266,"url":263},[97,21,273],"机器人感知",[],{"status":27,"locked":14,"version":28,"method":29,"evidence":276},{"candidates":277,"agricultural_evidence":14,"reason":48},[278],{"code":33,"score":279,"supported":35,"title":280,"abstract":281,"tags":283},27,[37,40],[282,21,40],"目标检测",[97,21],[285,286],"YOLO 机器人 目标检测","农业人工智能 机器人感知 计算机视觉","YOLO机器人目标检测-5494","10.64388\u002Firev10i4-1724117",{"doi":288,"openalex_id":290,"authors":291,"venue":266,"cited_by_count":15,"oa_url":296,"card":297,"direction":71,"ingested_from":74},"W7222933754",[292,294],{"name":293,"orcid":9},"Aafareen Sikandar Shaikh",{"name":295,"orcid":9},"Archana Nilesh Pawar","https:\u002F\u002Fwww.irejournals.com\u002Fformatedpaper\u002F1724117.pdf",{"tldr":298,"method":299,"finding":300,"direction":71,"opportunity":301},"提出将轻量YOLO检测器与机器人相机集成的框架，并定义原型评估方案。","轻量YOLO单阶段检测、机器人相机图像采集与预处理、置信度过滤及控制器通信。","紧凑YOLO适合资源受限机器人平台，但可靠导航仍需深度感知与安全控制器。","可将该框架迁移至农业机器人，研究田间遮挡与光照变化下的轻量检测与深度感知融合。","2026-10-21T23:30:40.721716Z",{"id":304,"title":305,"url":306,"summary":307,"summary_zh":308,"content":9,"source_name":309,"source_url":306,"published_at":267,"category":12,"cover_url":9,"hotness":13,"is_selected":14,"score":310,"score_detail":311,"sources":314,"tags":316,"primary_topic":9,"secondary_topics":319,"classification":320,"search_phrases":354,"slug":357,"view_count":15,"doi":358,"paper":359,"created_at":382},5402,"AI‐Enabled RGB Light‐to‐Frequency Sensor for Nondestructive Salinity Stress Monitoring in Medicinal Plants","https:\u002F\u002Fdoi.org\u002F10.1002\u002Faisy.70561","Salinity stress constrains medicinal plant productivity, particularly in coastal and salt affected agroecosystems where rapid, deployable diagnostics are essential. Here, we develop an AI‐enabled RGB light‐to‐frequency proximal sensing platform integrated with Tiny Machine Learning (TinyML) for nondestructive, real‐time salinity discrimination. Plants are exposed to three salinity levels (0, 100, and 200 mM NaCl). RGB‐derived and CIELAB chromatic descriptors are engineered, normalized, and partitioned (75:25) into training and test datasets. ReliefF‐based feature ranking and systematic subset optimization are combined with supervised learning using fivefold cross‐validation, and interpretability is achieved through SHAP analysis. Biomass reductions under saline treatments confirm physiological perturbations underlying spectral variability. Subspace KNN achieved the highest predictive performance (96.28% validation; 95.55% independent test accuracy). Framing model selection as a multiobjective optimization problem identified an optimal 10‐feature subset. Under this constraint, a Wide Neural Network attained 0.904 accuracy with an ultracompact 0.017 MB model size and >246 000 observations s −1 inference speed, demonstrating full TinyML compatibility. SHAP attribution consistently highlighted green intensity, blue channel metrics, and chromatic ratios as dominant predictors. This work establishes a lightweight, interpretable Edge‐AI framework for scalable, real‐time salinity diagnostics in precision agriculture.","盐胁迫制约药用植物的生产力，尤其是在沿海及受盐影响的农业生态系统中，亟需可快速部署的诊断手段。本研究开发了一种基于人工智能的RGB光-频率近端传感平台，并集成微型机器学习（TinyML），用于无损、实时的盐胁迫判别。植物暴露于三个盐浓度水平（0、100和200 mM NaCl）。研究构建了RGB衍生描述符与CIELAB色度描述符，经归一化后按75:25划分为训练集与测试集。将基于ReliefF的特征排序与系统性特征子集优化相结合，采用五折交叉验证进行监督学习，并通过SHAP分析实现模型可解释性。盐处理下的生物量下降证实了光谱变异性背后的生理扰动。子空间KNN取得了最高的预测性能（验证集准确率96.28%；独立测试集准确率95.55%）。将模型选择构建为多目标优化问题，确定了最优的10特征子集。在该约束下，宽神经网络（Wide Neural Network）达到0.904的准确率，模型体积仅0.017 MB，推理速度超过246 000次观测\u002F秒，展现出完全的TinyML兼容性。SHAP归因一致表明，绿色强度、蓝色通道指标及色度比是主要预测因子。本研究为精准农业中可扩展的实时盐胁迫诊断建立了一个轻量、可解释的边缘人工智能框架。","Advanced Intelligent Systems",81,{"impact":34,"substance":153,"depth":34,"authority":312,"freshness":210,"relevant":44,"comment":313},14,"该研究提出一种轻量、可解释的边缘AI盐胁迫诊断框架，方法新颖、数据扎实，对精准农业具有参考价值，但属于实验室阶段成果，产业影响有限。",[315],{"name":309,"url":306},[96,97,22,317,318],"盐胁迫监测","药用植物",[],{"status":27,"locked":14,"version":28,"method":29,"evidence":321},{"candidates":322,"agricultural_evidence":35,"reason":353},[323,330,336,342,349],{"code":324,"score":325,"supported":35,"title":326,"abstract":327,"tags":329},"B6",7,[],[318,328],"medicinal plant",[318],{"code":116,"score":325,"supported":35,"title":331,"abstract":332,"tags":335},[],[333,334],"精准农业","precision agriculture",[96],{"code":33,"score":325,"supported":35,"title":337,"abstract":338,"tags":341},[],[339,340],"机器学习","machine learning",[97],{"code":343,"score":344,"supported":35,"title":345,"abstract":346,"tags":348},"H6",3,[],[347],"农业生态",[],{"code":43,"score":44,"supported":14,"title":350,"abstract":351,"tags":352},[],[],[22],"多个产业证据相当，需要确认主归属",[355,356],"RGB 光频传感器 盐胁迫","TinyML 药用植物 无损监测","RGB光频传感器盐胁迫-5402","10.1002\u002Faisy.70561",{"doi":358,"openalex_id":360,"authors":361,"venue":309,"cited_by_count":15,"oa_url":306,"card":377,"direction":71,"ingested_from":74},"W7222914402",[362,365,368,371,374],{"name":363,"orcid":364},"Ahmad Ali","https:\u002F\u002Forcid.org\u002F0000-0001-5530-7374",{"name":366,"orcid":367},"Vinie Lee Silva-Alvarado","https:\u002F\u002Forcid.org\u002F0009-0000-5857-3248",{"name":369,"orcid":370},"Arman Heydari","https:\u002F\u002Forcid.org\u002F0009-0002-4705-0729",{"name":372,"orcid":373},"Sandra Sendra","https:\u002F\u002Forcid.org\u002F0000-0001-9556-9088",{"name":375,"orcid":376},"Jaime Lloret","https:\u002F\u002Forcid.org\u002F0000-0002-0862-0533",{"tldr":378,"method":379,"finding":380,"direction":73,"opportunity":381},"开发AI驱动的RGB光频传感器，实现药用植物盐胁迫无损实时监测。","RGB与CIELAB色度特征、ReliefF特征选择、SHAP解释及TinyML","Subspace KNN验证准确率96.28%，10特征Wide NN模型仅0.017MB且推理超2","可探索多作物多胁迫通用色度特征库，及田间光照变化下的模型鲁棒性与长期部署验证。","2026-10-20T23:30:35.809539Z",{"id":384,"title":385,"url":386,"summary":387,"summary_zh":388,"content":9,"source_name":389,"source_url":386,"published_at":390,"category":12,"cover_url":9,"hotness":13,"is_selected":14,"score":391,"score_detail":392,"sources":394,"tags":396,"primary_topic":399,"secondary_topics":404,"classification":405,"search_phrases":428,"slug":431,"view_count":15,"doi":432,"paper":433,"created_at":453},5401,"A HIERARCHICAL DISTRIBUTED EDGE AI FRAMEWORK FOR REAL-TIME APPLE LEAF DISEASE CLASSIFICATION","https:\u002F\u002Fdoi.org\u002F10.22190\u002Ffuacr260506002g","The application of artificial intelligence (AI) in agriculture has transformed crop management, particularly in the detection of plant diseases. Apple trees, a key agricultural commodity, are vulnerable to various diseases, which can severely impact crop yields. Traditional disease detection methods, often reliant on manual inspections, are time-consuming and prone to error. In response, this study proposes a novel two-stage edge AI framework for apple leaf disease classification that integrates RGB imagery with distributed edge computing. The first stage performs binary classification to distinguish between healthy and diseased leaves, achieving an F1 score of 95.50%. This optimization is achieved by utilizing lightweight models on edge devices, thereby enhancing computational efficiency. The second stage employs a consensus-based multiclass classification approach, achieving a 96.44% F1 score in identifying specific diseases. This dual-stage system, supported by an edge AI architecture, ensures real-time disease detection while minimizing resource consumption and enhancing scalability in large-scale agricultural environments. The full system achieved a 95.50% accuracy and F1 score on the special dataset. The results demonstrate the system’s efficacy in controlled laboratory conditions, offering a promising solution for precise, real-time agricultural monitoring. This study underscores the potential of distributed edge AI systems for scalable and efficient disease detection in precision agriculture.","人工智能（AI）在农业中的应用已深刻改变了作物管理方式，尤其是在植物病害检测方面。苹果树作为重要的农业经济作物，易受多种病害侵袭，严重影响作物产量。传统的病害检测方法通常依赖人工检查，耗时且易出错。为此，本研究提出了一种新颖的两阶段边缘AI框架，用于苹果叶片病害分类，该框架将RGB图像与分布式边缘计算相结合。第一阶段执行二分类以区分健康叶片与患病叶片，F1分数达到95.50%。该优化通过在边缘设备上采用轻量级模型实现，从而提升了计算效率。第二阶段采用基于共识的多类分类方法，在识别具体病害方面F1分数达到96.44%。该双阶段系统在边缘AI架构的支持下，实现了实时病害检测，同时最大限度地降低了资源消耗，并增强了在大规模农业环境中的可扩展性。完整系统在专用数据集上达到了95.50%的准确率和F1分数。结果表明，该系统在受控实验室条件下具有有效性，为精准、实时的农业监测提供了一种有前景的解决方案。本研究凸显了分布式边缘AI系统在精准农业中实现可扩展且高效病害检测的潜力。","Facta Universitatis Series Automatic Control and Robotics","2026-10-18T00:00:00Z",76,{"impact":34,"substance":209,"depth":90,"authority":91,"freshness":89,"relevant":44,"comment":393},"该研究提出两级边缘AI框架，在苹果叶片病害识别上取得95%以上F1值，方法新颖、数据可靠，对精准农业实时监测有参考价值，但尚处实验室验证阶段，产业影响有限。",[395],{"name":389,"url":386},[96,97,22,397,398],"作物病害识别","苹果病害",{"code":400,"slug":401,"name":402,"path":403,"parent_code":224},"B2","fruit","果树","\u002Ftopic\u002Fhorticulture\u002Ffruit",[],{"status":107,"locked":14,"version":28,"method":29,"evidence":406},{"candidates":407,"agricultural_evidence":35},[408,414,418,423],{"code":400,"score":34,"supported":35,"title":409,"abstract":411,"tags":413},[410],"apple",[402,412,410],"苹果",[398],{"code":116,"score":325,"supported":35,"title":415,"abstract":416,"tags":417},[],[333,334],[96],{"code":43,"score":325,"supported":35,"title":419,"abstract":420,"tags":422},[],[22,421],"edge computing",[22],{"code":33,"score":424,"supported":35,"title":425,"abstract":426,"tags":427},5,[],[39],[397,97],[429,430],"苹果叶片病害 边缘AI","分布式边缘计算 病害分类","苹果叶片病害边缘AI-5401","10.22190\u002Ffuacr260506002g",{"doi":432,"openalex_id":434,"authors":435,"venue":389,"cited_by_count":15,"oa_url":386,"card":448,"direction":71,"ingested_from":74},"W7222964752",[436,439,442,445],{"name":437,"orcid":438},"Milan Grujev","https:\u002F\u002Forcid.org\u002F0009-0002-3211-6844",{"name":440,"orcid":441},"Dragan H. Stojanovic","https:\u002F\u002Forcid.org\u002F0000-0003-4893-3814",{"name":443,"orcid":444},"Miloš Ilić","https:\u002F\u002Forcid.org\u002F0000-0002-9696-2874",{"name":446,"orcid":447},"Aleksandar Milosavljević","https:\u002F\u002Forcid.org\u002F0000-0003-3057-7727",{"tldr":449,"method":450,"finding":451,"direction":73,"opportunity":452},"提出两级边缘AI框架，用RGB图像实时分类苹果叶片病害。","分布式边缘计算+轻量模型，先二分类再共识多分类。","两级系统准确率与F1均达95.50%，多分类F1为96.44%。","仅在受控实验室验证，可延伸至田间复杂光照与多作物泛化及边缘能耗优化。","2026-10-20T23:30:33.830748Z"]