[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"item-245":3,"related-245":60},{"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":22,"tags":24,"primary_topic":30,"secondary_topics":36,"classification":37,"search_phrases":54,"slug":57,"view_count":58,"doi":8,"paper":8,"created_at":59},245,"PestRefineDet: integrating DenseNet with RefineDet for precision pest detection in agriculture","https:\u002F\u002Fwww.ebiotrade.com\u002Fnewsf\u002F2026-8\u002F20260805165305044.htm","2026年8月5日《Frontiers in Plant Science》上线的研究提出PestRefineDet——一种基于深度学习的精准害虫检测与分类框架,将DenseNet-41与RefineDet集成。DenseNet-41作为骨干网络通过密集连接增强特征提取,有效学习细粒度害虫特征,提取的特征由单阶段RefineDet框架处理实现害虫同时定位与分类。该框架在具有挑战性的IP102基准数据集上评估,实现平均精度(mAP@0.5)83.61%,在保持计算效率的同时提供对多种害虫物种的准确定位与可靠分类,为自动化农业害虫监测提供前景广阔的解决方案,有望支持智能和可持续害虫管理应用。",null,"《Frontiers in Plant Science》：PestRefineDet: integrating DenseNet with RefineDet for precision pest detection in agriculture\n\n**编辑推荐：**\n\n摘要：引言：害虫是农业生产力的主要威胁，导致大量作物损失和粮食产量下降。准确及时的害虫检测对于减少经济损失和支持可持续作物管理至关重要。然而，由于许多害虫物种之间视觉相似性高且专家知识有限，人工害虫识别劳动密集、耗时且具有挑战性。方法：为应对这些挑战，研究人员\n\n摘要：引言：害虫是农业生产力的主要威胁，导致大量作物损失和粮食产量下降。准确及时的害虫检测对于减少经济损失和支持可持续作物管理至关重要。然而，由于许多害虫物种之间视觉相似性高且专家知识有限，人工害虫识别劳动密集、耗时且具有挑战性。方法：为应对这些挑战，研究人员提出了PestRefineDet，一种基于深度学习的框架，将DenseNet-41与RefineDet集成用于自动化害虫检测和分类。DenseNet-41被用作骨干网络，通过密集连接增强特征提取，从而有效学习细粒度害虫特征。提取的特征随后由单阶段RefineDet框架处理，用于同时进行害虫定位和分类。结果：该框架在具有挑战性的IP102基准数据集上进行了评估。实验结果表明，PestRefineDet实现了平均精度（mAP@0.5）83.61%，在保持计算效率的同时，提供了对多种害虫物种的准确定位和可靠分类。讨论：结果表明，该框架有效捕获了区分性害虫特征，并在大规模真实世界害虫数据集上提高了检测性能。PestRefineDet为自动化农业害虫监测提供了一种有前景的解决方案，并有可能支持智能和可持续的害虫管理应用。\n\n害虫是农业可持续生产的主要威胁，传统人工检测依赖视觉识别，受限于专家知识匮乏和物种间高视觉相似性，存在劳动密集、耗时及主观性强等问题。虽然早期机器学习方法尝试自动化，但依赖手工特征，泛化能力不足；深度学习方法虽能自动提取特征，却面临标注数据需求大、计算资源要求高及模型可解释性差等瓶颈。为克服上述局限，研究人员提出PestRefineDet框架，将DenseNet-41密集连接骨干网络与RefineDet单阶段检测框架集成，旨在以计算高效的方式实现细粒度害虫的精准定位与分类。该研究在IP102基准数据集（包含75,222张RGB图像，224×224像素，涵盖102类作物害虫，按官方划分训练集45,095张、验证集7,508张、测试集22,619张）上验证，结果显示平均精度（mAP@0.5）达83.61%，分类准确率83.57%，优于多种现有方法。该工作发表于《Frontiers in Plant Science》，为智能农业害虫监测提供了有效方案。\n\n**主要关键的技术方法**（不超过250字）： \n\n研究人员采用DenseNet-41作为特征提取骨干网络，通过密集连接（Dense Connection）实现特征重用与梯度传播，增强细粒度特征学习。该网络包含4个密集块（Dense Block）和3个过渡层（Transition Layer），共41层，输出多尺度特征图。随后，这些特征输入RefineDet框架，该框架包括锚点细化模块（Anchor Refinement Module, ARM）和对象检测模块（Object Detection Module, ODM），通过两步级联回归（Two-Step Cascaded Regression, TSCR）和负锚点过滤（Negative Anchor Filtering, NAF）优化定位与分类。训练使用ImageNet预训练权重，在IP102数据集上以随机梯度下降（SGD）优化器（学习率0.0015、动量0.9、权重衰减0.0005）微调30个周期，批量大小32，输入尺寸320×320像素，未采用数据增强以评估模型原生特征学习能力。\n\n**研究结果**： \n\n- **4.3 害虫识别评估（Evaluation of pest recognition）**：通过使用IP102测试集所有样本进行定位实验，定量计算平均精度（mAP）和交并比（IoU），得出mAP@0.5为83.61%，平均IoU为84.42%，表明模型能在复杂背景和光照变化下准确定位并识别不同尺寸、形状和颜色的害虫。 \n\n- **4.4 102类害虫识别的作物分类分析（Crop-wise analysis of 102-class pest recognition performance）**：将原始102类害虫按所关联作物（水稻、玉米、甜菜、小麦、苜蓿、芒果、柑橘、葡萄）分组，计算召回率、精确率和F1分数，平均精确率83.9%、召回率82.3%、F1分数83.57%，其中芒果、葡萄和柑橘类精度较高，水稻和苜蓿因背景相似和类内变异较大而精度较低。 \n\n- **4.5 消融研究（Ablation study）**：在保持RefineDet框架不变的情况下，替换骨干网络为VGG-16、ResNet-50、ResNet-101、EfficientNet-B0和DenseNet-121，与提出的DenseNet-41比较。结果显示，DenseNet-41以11.4百万参数和45.6 MB模型大小达到最高mAP@0.5（83.61%）和准确率（83.57%），证明密集连接机制在特征重用和计算效率间的平衡优势。 \n\n- **4.6 与其他检测模型的比较（Comparison with other detection models）**：与Faster R-CNN、SSD、YOLOv3和原始RefineDet（均使用不同骨干网络）在IP102上对比，PestRefineDet以83.61% mAP和0.22秒\u002F张的测试时间超越所有对比模型，证实DenseNet-41骨干对定位和分类的增强效果。 \n\n- **4.7 与现有先进方法的比较（Comparison with state-of-the-art approaches）**：与Ayan et al. (2020)、Nanni et al. (2020)、Albattah et al. (2023)、Ali et al. (2023)、Anwar and Masood (2023)、Rimal et al. (2023)、Agrawal et al. (2026)和Wang et al. (2025)在IP102上比较分类准确率，PestRefineDet以83.57%准确率最高，且计算效率更优，能够处理扭曲样本和类间差异。 \n\n- **4.8 实际意义与部署可行性（Practical implications and deployment feasibility）**：通过分析模型轻量化设计（11.4M参数、3.1 GFLOPs）和边缘设备部署潜力，指出该框架可用于综合害虫管理（IPM）中的智能监测，但需进一步在田间采集图像下验证光照、背景和生长阶段变化的影响。\n\n**讨论与结论**： \n\n讨论部分指出，所提框架有效捕获区分性害虫特征，在复杂真实数据集上提升了检测性能，为自动化农业害虫监测提供有前景的解决方案，但实际部署前需在田间图像上验证，并考虑误检影响（漏检延迟干预，虚报导致不必要检查）。研究结论如下：本研究提出了一种经济高效的深度学习技术，用于利用无人机图像对田间作物害虫进行自动检测和分类。该方法基于定制化的RefineDet模型，以DenseNet-41作为骨干网络提取输入图像中的判别性特征。细化的锚点与特征使得能够准确识别多种害虫类型。在包含田间害虫图像的挑战性基准IP102数据集上进行了评估，证明了该方法的实际有效性。大量实验表明，该框架能够在复杂背景和害虫特征（如形状、颜色、尺寸、方向和光照）变化下，准确地对多种害虫类别进行定位和分类。所提出的PestRefineDet框架通过结合准确害虫识别与计算效率，为智能害虫监测系统提供了有前景的基础。尽管当前评估基于IP102基准，未来工作将关注田间尺度验证以及与实际农业决策支持平台的集成，以应用于综合害虫管理（IPM）。","Frontiers in Plant Science 2026-08-05","2026-08-05T00:00:00Z","论文",10,false,82,{"impact":17,"substance":18,"depth":17,"authority":19,"freshness":13,"relevant":20,"comment":21},18,22,14,1,"提出PestRefineDet模型，在IP102数据集上取得83.61% mAP，方法新颖且结果可靠，对智慧农业害虫检测有实质贡献。",[23],{"name":10,"url":6},[25,26,27,28,29],"Frontiers","深度学习","精准农业","害虫检测","DenseNet RefineDet IP102",{"code":31,"slug":32,"name":33,"path":34,"parent_code":35},"I2","agricultural-ai","农业AI与大模型","\u002Ftopic\u002Fdigital-agriculture\u002Fagricultural-ai","I",[],{"status":38,"locked":14,"version":39,"method":40,"evidence":41},"classified","2026-10-22.2","rules",{"candidates":42,"agricultural_evidence":45},[43,49],{"code":31,"score":44,"supported":45,"title":46,"abstract":47,"tags":48},4,true,[],[26],[26],{"code":50,"score":20,"supported":14,"title":51,"abstract":52,"tags":53},"I1",[],[],[27],[55,56],"害虫检测 深度学习 精准农业 DenseNet RefineDet IP102","害虫检测 深度学习","害虫检测深度学习精准农业DenseNetRefineDetIP102-245",9,"2026-08-05T23:53:02.415763Z",{"total":61,"page":20,"page_size":61,"items":62},6,[63,138,202,259,331,406],{"id":64,"title":65,"url":66,"summary":67,"summary_zh":68,"content":8,"source_name":69,"source_url":66,"published_at":70,"category":12,"cover_url":8,"hotness":13,"is_selected":14,"score":71,"score_detail":72,"sources":78,"tags":80,"primary_topic":84,"secondary_topics":85,"classification":86,"search_phrases":117,"slug":120,"view_count":121,"doi":122,"paper":123,"created_at":137},4912,"Deep Learning Approaches for Plant Disease Detection: A Review of Convolutional Neural Networks and Emerging Techniques","https:\u002F\u002Fdoi.org\u002F10.56201\u002Fijcsmt.vol.11.no12.2025.pg220.231","Plant diseases pose a significant threat to global food security, crop yield, and agricultural sustainability. Traditional disease diagnosis methods, relying on expert visual inspection and laboratory analyses, are often time-consuming, labor-intensive, and costly, limiting their scalability. Recent advances in artificial intelligence, particularly deep learning, offer automated and accurate solutions for plant disease detection. This review synthesizes current research on deep learning–based approaches, focusing on convolutional neural networks (CNNs), hybrid CNN-transformer models, attention mechanisms, and generative data augmentation techniques. CNNs have demonstrated strong capabilities in extracting hierarchical features from leaf images, capturing complex patterns such as lesion shapes, color variations, and textures, thereby outperforming conventional machine learning and rule-based methods. Hybrid architectures and transformer-based models enhance feature extraction, particularly for subtle or visually similar disease symptoms. Data augmentation strategies, including generative adversarial networks (GANs), help address class imbalance and improve model generalization. Despite high reported accuracies, challenges remain, including limited dataset diversity, environmental variability, overfitting, lack of explainable AI, and computational constraints for mobile deployment. Future research should prioritize explainable models, multi-crop and multi-disease detection, lightweight architectures suitable for field applications, and integration with precision agriculture systems. Overall, deep learning–based plant disease detection presents a transformative approach for early diagnosis, timely intervention, and sustainable agricultural practices, offering the potential to reduce crop losses, enhance productivity, and contribute to global food security.","植物病害对全球粮食安全、作物产量和农业可持续性构成重大威胁。传统的病害诊断方法依赖专家视觉检查和实验室分析，往往耗时、费力且成本高昂，限制了其可扩展性。人工智能，特别是深度学习的最新进展，为植物病害检测提供了自动化且准确的解决方案。本文综述了基于深度学习方法的最新研究，重点关注卷积神经网络（CNN）、CNN-Transformer混合模型、注意力机制和生成式数据增强技术。CNN在从叶片图像中提取层次化特征方面展现出强大能力，能够捕捉病变形状、颜色变化和纹理等复杂模式，从而优于传统机器学习和基于规则的方法。混合架构和基于Transformer的模型增强了特征提取能力，尤其针对细微或视觉上相似的病害症状。数据增强策略，包括生成对抗网络（GAN），有助于解决类别不平衡问题并提高模型泛化能力。尽管已有研究报告了较高的准确率，但仍存在诸多挑战，包括数据集多样性有限、环境变异性、过拟合、缺乏可解释的人工智能以及移动端部署的计算约束。未来研究应优先关注可解释模型、多作物和多病害检测、适用于田间应用的轻量级架构以及与精准农业系统的集成。总体而言，基于深度学习的植物病害检测为早期诊断、及时干预和可持续农业实践提供了一种变革性方法，具有减少作物损失、提高生产力并促进全球粮食安全的潜力。","INTERNATIONAL JOURNAL OF COMPUTER SCIENCE AND MATHEMATICAL THEORY E-ISSN","2026-10-12T00:00:00Z",75,{"impact":17,"substance":73,"depth":74,"authority":75,"freshness":76,"relevant":20,"comment":77},20,17,12,8,"系统综述深度学习植物病害检测的方法进展与挑战，学术价值较高但偏理论，对产业落地有参考意义。",[79],{"name":69,"url":66},[81,82,26,27,83],"智慧农业","农业人工智能","植物病害检测",{"code":31,"slug":32,"name":33,"path":34,"parent_code":35},[],{"status":38,"locked":14,"version":39,"method":40,"evidence":87},{"candidates":88,"agricultural_evidence":45},[89,98,103,110],{"code":31,"score":90,"supported":45,"title":91,"abstract":93,"tags":97},19,[92],"deep learning",[26,94,95,92,96],"机器学习","machine learning","artificial intelligence",[82,26],{"code":50,"score":76,"supported":45,"title":99,"abstract":100,"tags":102},[],[27,101],"precision agriculture",[81,27],{"code":104,"score":61,"supported":45,"title":105,"abstract":106,"tags":109},"A8",[],[107,108],"粮食安全","food security",[],{"code":111,"score":112,"supported":45,"title":113,"abstract":114,"tags":116},"H8",3,[],[115],"可持续农业",[],[118,119],"植物病害检测 深度学习","卷积神经网络 作物病害识别","植物病害检测深度学习-4912",2,"10.56201\u002Fijcsmt.vol.11.no12.2025.pg220.231",{"doi":122,"openalex_id":124,"authors":125,"venue":69,"cited_by_count":128,"oa_url":129,"card":130,"direction":134,"ingested_from":136},"W7221637766",[126],{"name":127,"orcid":8},"NUHU ABDULLAHI",0,"https:\u002F\u002Fiiardjournals.org\u002Fget\u002FIJCSMT\u002FVOL. 11 NO. 12 2025\u002FDeep Learning Approaches for Plant 220-231.pdf",{"tldr":131,"method":132,"finding":133,"direction":134,"opportunity":135},"综述深度学习在植物病害检测中的应用，涵盖CNN、混合Transformer、注意力机制与生成式数据增","综述CNN、CNN-Transformer混合模型、注意力机制及GAN数据增强在","深度学习检测精度高，但仍受数据集多样性、环境变异、可解释性和移动部署算力限制。","农业人工智能与决策模型","可探索轻量化可解释模型、多作物多病害检测及与精准农业系统集成，解决田间部署难题。","openalex","2026-10-14T23:30:54.687367Z",{"id":139,"title":140,"url":141,"summary":142,"summary_zh":143,"content":8,"source_name":144,"source_url":141,"published_at":145,"category":12,"cover_url":8,"hotness":146,"is_selected":14,"score":147,"score_detail":148,"sources":151,"tags":155,"primary_topic":157,"secondary_topics":163,"classification":165,"search_phrases":183,"slug":186,"view_count":20,"doi":187,"paper":188,"created_at":201},4241,"Detection of Apple Plant Diseases Using Leaf Images through Convolutional Neural Network","https:\u002F\u002Fdoi.org\u002F10.5281\u002Fzenodo.23142591","Plant diseases have major social, ecological and economic impact on agriculture and frequently results in large financial losses for farmers. Therefore early and precise disease detection is crucial for efficient crop management. The severity of leaf diseases and their direct impact on yield quality have been highlighted in recent studies. Both digital cameras and flatbed scanners can be used to take leaf images but they have certain drawbacks. Although scanners offer a more secure and regulated method of acquiring images they are comparatively slow and may cause edge shadows. This study uses a hybrid deep learning approach to identify diseases in apple plant leaves. To improve classification performance; the proposed approach combines the advantages of Residual Networks (ResNet), Genetic Algorithms (GA) and Convolutional Neural Networks (CNN). CNNs are used to extract features and classify diseases whereas ResNet enhances the learning efficiency of deeper networks. The feature set is optimized using a genetic algorithm which improves accuracy and lowers computational complexity. The model also calculates the percentage of the leaf area that is impacted. According to experimental results the proposed method detects leaf diseases with high accuracy of percentage and low computational cost making it a dependable and effective precision agriculture solution. The proposed model exhibited strong classification efficiency across various leaf conditions that pertaining an accuracy of 0.97 for healthy leaves, 0.994 for apple scab, 0.89 for rust, and 0.91 for powdery mildew.","植物病害对农业产生重大的社会、生态和经济影响，并经常给农民带来巨大的经济损失。因此，早期且精确的病害检测对于高效的作物管理至关重要。近期研究强调了叶部病害的严重程度及其对产量品质的直接影响。数码相机和平板扫描仪均可用于拍摄叶片图像，但它们各有一定缺陷。尽管扫描仪提供了更安全、更可控的图像获取方式，但其速度相对较慢，且可能产生边缘阴影。本研究采用混合深度学习方法识别苹果植株叶部病害。为提高分类性能，所提方法结合了残差网络（ResNet）、遗传算法（GA）和卷积神经网络（CNN）的优势。CNN用于提取特征并分类病害，而ResNet增强了更深网络的学习效率。利用遗传算法对特征集进行优化，从而提高准确率并降低计算复杂度。该模型还计算了受影响的叶面积百分比。实验结果表明，所提方法以高准确率和低计算成本检测叶部病害，使其成为可靠且有效的精准农业解决方案。所提模型在多种叶片条件下均表现出较强的分类效率，健康叶片准确率为0.97，苹果黑星病为0.994，锈病为0.89，白粉病为0.91。","Zenodo (CERN European Organization for Nuclear Research)","2026-10-04T00:00:00Z",25,67,{"impact":75,"substance":17,"depth":149,"authority":75,"freshness":58,"relevant":20,"comment":150},16,"基于ResNet+GA+CNN的苹果叶病识别研究，准确率数据具体，方法有增量，但属实验室论文，产业影响有限。",[152,153],{"name":144,"url":141},{"name":144,"url":154},"https:\u002F\u002Fdoi.org\u002F10.5281\u002Fzenodo.23142592",[81,82,26,27,156],"苹果病害",{"code":158,"slug":159,"name":160,"path":161,"parent_code":162},"B2","fruit","果树","\u002Ftopic\u002Fhorticulture\u002Ffruit","B",[164],{"code":31,"slug":32,"name":33,"path":34,"parent_code":35},{"status":38,"locked":14,"version":39,"method":40,"evidence":166},{"candidates":167,"agricultural_evidence":45},[168,175,179],{"code":158,"score":169,"supported":45,"title":170,"abstract":172,"tags":174},15,[171],"apple",[173,171],"苹果",[156],{"code":50,"score":76,"supported":45,"title":176,"abstract":177,"tags":178},[],[27,101],[81,27],{"code":31,"score":76,"supported":45,"title":180,"abstract":181,"tags":182},[],[26,92],[82,26],[184,185],"苹果叶片病害 卷积神经网络","苹果锈病 白粉病 识别","苹果叶片病害卷积神经网络-4241","10.5281\u002Fzenodo.23142591",{"doi":187,"openalex_id":189,"authors":190,"venue":144,"cited_by_count":128,"oa_url":141,"card":195,"direction":200,"ingested_from":136},"W7219637192",[191,193],{"name":192,"orcid":8},"Mohd Arshad Raza",{"name":194,"orcid":8},"Kiran Pandey",{"tldr":196,"method":197,"finding":198,"direction":134,"opportunity":199},"提出ResNet+GA+CNN混合模型，用苹果叶片图像检测病害并计算病斑面积占比。","ResNet提取深层特征，遗传算法优化特征集，CNN分类，苹果叶片图像数据。","健康叶、黑星病、锈病、白粉病准确率分别达0.97、0.994、0.89、0.91，计算成本低。","可探索轻量化模型在田间移动端实时检测，并融合多病害严重度分级与产量损失预测。","数字乡村与农业信息化","2026-10-05T23:30:30.830366Z",{"id":203,"title":204,"url":205,"summary":206,"summary_zh":207,"content":8,"source_name":208,"source_url":205,"published_at":209,"category":12,"cover_url":8,"hotness":13,"is_selected":14,"score":147,"score_detail":210,"sources":213,"tags":215,"primary_topic":217,"secondary_topics":218,"classification":219,"search_phrases":237,"slug":240,"view_count":112,"doi":241,"paper":242,"created_at":258},2963,"A comprehensive review of deep learning methods for weed classification in precision agriculture","https:\u002F\u002Fdoi.org\u002F10.1007\u002Fs44163-026-01916-7","Weeds today are among the factors contributing to low agricultural productivity. As the world’s population continues to grow, there is an urgent need to meet global food demand. Nigeria currently lacks sufficient crop production to feed its growing population, and weeds are among the core contributors to poor agricultural yield. This study conducts a comprehensive review of Deep Learning (DL) approaches to weed classification in precision agriculture, covering literature from 2018 to 2025, was carried out. We employed a mix of quantitative and qualitative methods in the course of this review paper. Our data source is centred on Scopus-indexed papers, published with Sensors, Electronics, and Agriculture in MDPI as well as IEEE, Thomson Reuters, and Springer. The study systematically reviewed and analysed machine learning (ML), DL, and instance segmentation techniques to identify the key technological and environmental barriers, such as data limitations, class imbalance, environmental variability, and model scalability issues that affect the effectiveness and efficiency of these models when deployed in real time. These findings show that while weed management models like the YOLO variants, ResNet, and Vision Transformers achieved high accuracy in training and testing, they are associated with several challenges in their real world-deployment, such as occlusion, small object detection, and environmental adaptability. Overall, this research provides recommended solutions to enhance model robustness, scalability, and efficiency. It further provides a summary of the current state and future directions for AI-driven weed management.","杂草是当前导致农业生产力低下的因素之一。随着世界人口持续增长，满足全球粮食需求已成为迫切任务。尼日利亚目前的作物产量不足以养活其不断增长的人口，而杂草是导致农业产量低下的核心因素之一。本研究对精准农业中基于深度学习（Deep Learning，DL）的杂草分类方法进行了全面综述，涵盖2018年至2025年的文献。在综述过程中，我们采用了定量与定性相结合的方法。数据来源集中于Scopus索引论文，这些论文发表于MDPI旗下的Sensors、Electronics和Agriculture，以及IEEE、Thomson Reuters和Springer。本研究系统综述并分析了机器学习（Machine Learning，ML）、深度学习及实例分割技术，以识别影响这些模型实时部署效果与效率的关键技术和环境障碍，如数据局限性、类别不平衡、环境变异性及模型可扩展性问题。研究结果表明，尽管YOLO系列、ResNet和视觉Transformer（Vision Transformer）等杂草管理模型在训练和测试中达到了较高精度，但在实际部署中仍面临诸多挑战，如遮挡、小目标检测和环境适应性等问题。总体而言，本研究提出了增强模型鲁棒性、可扩展性和效率的推荐解决方案，并进一步总结了人工智能驱动杂草管理的现状与未来方向。","Discover Artificial Intelligence","2026-09-18T00:00:00Z",{"impact":75,"substance":17,"depth":149,"authority":211,"freshness":76,"relevant":20,"comment":212},13,"系统综述2018—2025年深度学习杂草分类方法，指出遮挡、小目标与环境适应性等落地瓶颈，对农业AI研究有参考价值，但属综述类论文、非突破性成果。",[214],{"name":208,"url":205},[81,82,26,216,27],"杂草识别",{"code":31,"slug":32,"name":33,"path":34,"parent_code":35},[],{"status":38,"locked":14,"version":39,"method":40,"evidence":220},{"candidates":221,"agricultural_evidence":45},[222,228,232],{"code":31,"score":90,"supported":45,"title":223,"abstract":224,"tags":227},[92],[225,26,94,95,92,226],"目标检测","YOLO",[82,26],{"code":50,"score":149,"supported":45,"title":229,"abstract":230,"tags":231},[101],[27,101],[81,27],{"code":233,"score":20,"supported":14,"title":234,"abstract":235,"tags":236},"F4",[],[],[216],[238,239],"深度学习 杂草分类 精准农业","YOLO 杂草识别 模型部署","深度学习杂草分类精准农业-2963","10.1007\u002Fs44163-026-01916-7",{"doi":241,"openalex_id":243,"authors":244,"venue":208,"cited_by_count":128,"oa_url":205,"card":253,"direction":134,"ingested_from":136},"W7213558223",[245,247,249,251],{"name":246,"orcid":8},"Njoku Camillus Ekene",{"name":248,"orcid":8},"Francis A. Okoye",{"name":250,"orcid":8},"Ebere Uzoka Chidi",{"name":252,"orcid":8},"OGBU MARY NNENNA",{"tldr":254,"method":255,"finding":256,"direction":134,"opportunity":257},"综述2018-2025年深度学习杂草分类方法，分析技术瓶颈并给出改进建议。","混合定量定性法，基于Scopus及MDPI、IEEE等文献，分析ML、DL与实例","YOLO、ResNet、ViT等精度高，但实际部署受遮挡、小目标与环境适应性限制。","可研究轻量化、跨域自适应模型，解决小目标与遮挡下的实时杂草识别难题。","2026-09-19T23:30:56.875661Z",{"id":260,"title":261,"url":262,"summary":263,"summary_zh":264,"content":8,"source_name":265,"source_url":262,"published_at":209,"category":12,"cover_url":8,"hotness":13,"is_selected":14,"score":15,"score_detail":266,"sources":268,"tags":270,"primary_topic":273,"secondary_topics":274,"classification":280,"search_phrases":304,"slug":307,"view_count":20,"doi":308,"paper":309,"created_at":330},2942,"Deep learning-driven multisource remote sensing image fusion: Advances, challenges, and future directions","https:\u002F\u002Fdoi.org\u002F10.1016\u002Fj.engappai.2026.116319","Multisource remote sensing image fusion has become an important solution to a long-standing limitation in Earth observation: individual sensors rarely provide high spatial detail, rich spectral information, reliable structural sensitivity, and frequent temporal coverage at the same time. This review examines how deep learning and artificial intelligence are being used to integrate multispectral, hyperspectral, panchromatic, optical, and synthetic aperture radar imagery for more reliable interpretation of complex ground scenes. It provides a technical synthesis of convolutional neural networks, autoencoders, generative adversarial networks, transformer architectures, diffusion models, and hybrid model driven approaches, with attention to their fusion mechanisms, reconstruction behavior, computational demand, and suitability for operational use. Applications include land cover mapping, precision agriculture, environmental monitoring, urban analysis, disaster assessment, and defense related interpretation. Rather than treating each fusion task separately, this review connects sensor heterogeneity, spatial and spectral resolution trade offs, radiometric correction, geometric correction, registration, noise reduction, and fusion level design within a single framework. The analysis indicates that convolutional models remain effective for stable local detail recovery, adversarial learning can improve visual sharpness but may introduce spectral distortion, transformer models better capture long range spatial and spectral relationships, and diffusion models offer refined reconstruction at greater computational cost. The review further identifies open challenges involving misregistration, spectral bias, limited labeled data, weak generalization across sensors, high memory requirements, and limited interpretability. Future progress should prioritize sensor aware learning, self supervised training, uncertainty aware evaluation, lightweight deployment, and application oriented benchmarks to improve reliability in operational Earth observation.","多源遥感图像融合已成为解决地球观测领域一个长期局限的重要方案：单一传感器很少能够同时提供高空间细节、丰富光谱信息、可靠的结构敏感性以及频繁的时间覆盖。本文综述了如何利用深度学习和人工智能整合多光谱、高光谱、全色、光学和合成孔径雷达（synthetic aperture radar, SAR）影像，以更可靠地解译复杂地表场景。文章对卷积神经网络、自编码器、生成对抗网络、Transformer架构、扩散模型以及混合模型驱动方法进行了技术综合，重点关注其融合机制、重建行为、计算需求以及业务化适用性。应用领域包括土地覆盖制图、精准农业、环境监测、城市分析、灾害评估和国防相关解译。本文并非将每种融合任务分开处理，而是在一个统一框架内将传感器异质性、空间与光谱分辨率权衡、辐射校正、几何校正、配准、降噪和融合层级设计联系起来。分析表明，卷积模型在稳定的局部细节恢复方面仍然有效，对抗学习可以提升视觉锐度但可能引入光谱失真，Transformer模型能更好地捕捉长程空间与光谱关系，而扩散模型以更高的计算成本提供精细重建。本文进一步指出了涉及配准误差、光谱偏差、标注数据有限、跨传感器泛化能力弱、高内存需求以及可解释性有限等开放挑战。未来的进展应优先关注传感器感知学习、自监督训练、不确定性感知评估、轻量化部署以及面向应用的基准测试，以提高业务化地球观测的可靠性。","Engineering Applications of Artificial Intelligence",{"impact":17,"substance":18,"depth":90,"authority":19,"freshness":58,"relevant":20,"comment":267},"发表于核心期刊的综述，系统梳理深度学习多源遥感融合的方法、应用与挑战，对农业遥感与精准农业有直接参考价值，时效性强，值得进入每日精选。",[269],{"name":265,"url":262},[82,26,27,271,272],"遥感","多源数据融合",{"code":31,"slug":32,"name":33,"path":34,"parent_code":35},[275],{"code":276,"slug":277,"name":278,"path":279,"parent_code":35},"I3","remote-sensing","遥感与农情监测","\u002Ftopic\u002Fdigital-agriculture\u002Fremote-sensing",{"status":38,"locked":14,"version":39,"method":40,"evidence":281},{"candidates":282,"agricultural_evidence":45},[283,287,294,299],{"code":31,"score":90,"supported":45,"title":284,"abstract":285,"tags":286},[92],[26,92,96],[82,26],{"code":276,"score":17,"supported":45,"title":288,"abstract":290,"tags":293},[289],"remote sensing",[271,289,291,292],"multispectral","hyperspectral",[271],{"code":50,"score":295,"supported":45,"title":296,"abstract":297,"tags":298},7,[],[27,101],[27],{"code":300,"score":20,"supported":14,"title":301,"abstract":302,"tags":303},"I5",[],[],[272],[305,306],"多源遥感 图像融合 深度学习","农业人工智能 多源数据融合 深度学习 精准农业","多源遥感图像融合深度学习-2942","10.1016\u002Fj.engappai.2026.116319",{"doi":308,"openalex_id":310,"authors":311,"venue":265,"cited_by_count":128,"oa_url":262,"card":324,"direction":328,"ingested_from":136},"W7213544281",[312,315,318,321],{"name":313,"orcid":314},"Shahid Karim","https:\u002F\u002Forcid.org\u002F0000-0001-9986-5052",{"name":316,"orcid":317},"Akeel Qadir","https:\u002F\u002Forcid.org\u002F0000-0003-0358-6505",{"name":319,"orcid":320},"Asif Ali Laghari","https:\u002F\u002Forcid.org\u002F0000-0001-5831-5943",{"name":322,"orcid":323},"Irfana Bibi","https:\u002F\u002Forcid.org\u002F0000-0003-2794-504X",{"tldr":325,"method":326,"finding":327,"direction":328,"opportunity":329},"综述深度学习多源遥感图像融合方法、挑战与未来方向。","综述CNN、GAN、Transformer、扩散模型等融合机制与重建行为。","CNN擅局部细节，GAN易谱失真，Transformer长程关系强，扩散模型精度高但算力大。","农业遥感与作物表型","面向农业的轻量、自监督、不确定性感知融合与基准数据集构建。","2026-09-19T23:30:32.767384Z",{"id":332,"title":333,"url":334,"summary":335,"summary_zh":336,"content":8,"source_name":337,"source_url":334,"published_at":338,"category":12,"cover_url":8,"hotness":13,"is_selected":14,"score":339,"score_detail":340,"sources":342,"tags":344,"primary_topic":346,"secondary_topics":352,"classification":354,"search_phrases":388,"slug":391,"view_count":392,"doi":393,"paper":394,"created_at":405},2940,"A Hybrid CNN–Transformer–LSTM Deep Learning Framework for Automated Cotton Disease Detection","https:\u002F\u002Fdoi.org\u002F10.53365\u002Fnrfhh.1816","Cotton production is frequently affected by leaf diseases that can reduce plant productivity, deteriorate crop quality, and cause considerable financial losses for farmers. Consequently, rapid and reliable disease identification is an important requirement for precision agriculture and effective crop protection. Conventional image-based approaches predominantly employ CNN architectures for visual classification; however, such methods may have limited capability in learning long-range spatial dependencies and modeling changes associated with disease development over time. To overcome these limitations, this research introduces an integrated hybrid deep learning framework that combines Convolutional Neural Networks (CNNs), Transformer-based self-attention, and Long Short-Term Memory (LSTM) networks for intelligent cotton leaf disease recognition. The proposed framework utilizes a pre-trained EfficientNet network to extract discriminative spatial characteristics from cotton leaf images. The extracted representations are subsequently processed through a multi-head self-attention Transformer module, which enables the network to identify relationships between distant and relevant regions of the leaf. An LSTM component is then incorporated to learn sequential dependencies and provide a foundation for analyzing disease evolution and progression. To improve model reliability and generalization, the network is trained using an appropriately organized cotton leaf image dataset together with image augmentation, transfer learning, and fine-tuning techniques. The experimental evaluation indicates that the proposed CNN–Transformer–LSTM architecture provides improved disease classification performance and more effective feature learning compared with conventional CNN-based models. Performance assessment using classification metrics, confusion matrices, and ROC curves demonstrates strong discrimination among the considered cotton disease categories. The combined learning of local visual characteristics, global contextual information, and sequential dependencies provides a comprehensive framework for automated cotton disease assessment. The proposed approach can serve as a scalable artificial intelligence solution for precision agriculture applications. Its architecture also provides opportunities for future integration with IoT-based crop monitoring systems, environmental data acquisition, and disease progression prediction, supporting early warning systems and intelligent crop health management in smart farming environments.","棉花生产常受叶片病害影响，这些病害会降低植株生产力、恶化作物品质，并给农民造成相当大的经济损失。因此，快速可靠的病害识别是精准农业和有效作物保护的重要需求。传统的基于图像的方法主要采用CNN架构进行视觉分类；然而，此类方法在学习长程空间依赖关系以及建模与病害随时间发展相关的变化方面能力有限。为克服这些局限，本研究提出了一种集成混合深度学习框架，将卷积神经网络（CNN）、基于Transformer的自注意力机制和长短期记忆（LSTM）网络相结合，用于智能棉花叶片病害识别。所提出的框架利用预训练的EfficientNet网络从棉花叶片图像中提取具有判别力的空间特征。提取到的表示随后通过多头自注意力Transformer模块进行处理，使网络能够识别叶片中相距较远且相关区域之间的关系。随后引入LSTM组件以学习序列依赖关系，并为分析病害演变和进展提供基础。为提高模型可靠性和泛化能力，网络使用组织良好的棉花叶片图像数据集进行训练，并结合图像增强、迁移学习和微调技术。实验评估表明，与传统的基于CNN的模型相比，所提出的CNN–Transformer–LSTM架构提供了更好的病害分类性能和更有效的特征学习。使用分类指标、混淆矩阵和ROC曲线进行的性能评估表明，该方法在所考虑的棉花病害类别之间具有较强的判别能力。局部视觉特征、全局上下文信息和序列依赖关系的联合学习为自动化棉花病害评估提供了一个全面的框架。所提出的方法可作为精准农业应用中可扩展的人工智能解决方案。其架构也为未来与基于物联网的作物监测系统、环境数据采集和病害进展预测的集成提供了机会，从而支持预警系统和智能","Natural Resources for Human Health","2026-09-16T00:00:00Z",69,{"impact":75,"substance":73,"depth":74,"authority":75,"freshness":76,"relevant":20,"comment":341},"提出CNN-Transformer-LSTM混合框架用于棉花叶病识别，方法有创新但属实验室验证阶段，产业影响有限。",[343],{"name":337,"url":334},[81,82,26,27,345],"棉花病害识别",{"code":347,"slug":348,"name":349,"path":350,"parent_code":351},"A6","cash-crops","棉麻糖与特色经作","\u002Ftopic\u002Ffield-crops\u002Fcash-crops","A",[353],{"code":31,"slug":32,"name":33,"path":34,"parent_code":35},{"status":38,"locked":14,"version":39,"method":40,"evidence":355},{"candidates":356,"agricultural_evidence":45},[357,361,367,371,377,383],{"code":31,"score":90,"supported":45,"title":358,"abstract":359,"tags":360},[92],[26,92,96],[82,26],{"code":347,"score":19,"supported":45,"title":362,"abstract":364,"tags":366},[363],"cotton",[365,363],"棉花",[],{"code":50,"score":76,"supported":45,"title":368,"abstract":369,"tags":370},[],[27,101],[81,27],{"code":300,"score":61,"supported":45,"title":372,"abstract":373,"tags":376},[],[374,375],"物联网","IoT",[],{"code":378,"score":112,"supported":45,"title":379,"abstract":380,"tags":382},"A4",[],[381],"花生",[],{"code":233,"score":112,"supported":45,"title":384,"abstract":385,"tags":387},[],[386],"crop protection",[],[389,390],"棉花叶病 CNN Transformer LSTM","EfficientNet 棉花病害检测","棉花叶病CNNTransformerLSTM-2940",5,"10.53365\u002Fnrfhh.1816",{"doi":393,"openalex_id":395,"authors":396,"venue":337,"cited_by_count":128,"oa_url":8,"card":399,"direction":404,"ingested_from":136},"W7213544712",[397],{"name":398,"orcid":8},"Prajakta Sunil Gupta",{"tldr":400,"method":401,"finding":402,"direction":134,"opportunity":403},"提出CNN-Transformer-LSTM混合框架，实现棉花叶片病害自动识别。","EfficientNet提取特征，Transformer自注意力与LSTM建模，","混合模型分类性能优于传统CNN，能同时学习局部特征、全局上下文和序列依赖。","可融合IoT环境数据与时间序列，开展病害进展预测和早期预警系统研究。","智慧农业 \u002F 农业物联网","2026-09-19T23:30:19.558449Z",{"id":407,"title":408,"url":409,"summary":410,"summary_zh":411,"content":8,"source_name":412,"source_url":409,"published_at":413,"category":12,"cover_url":8,"hotness":13,"is_selected":14,"score":71,"score_detail":414,"sources":416,"tags":418,"primary_topic":419,"secondary_topics":420,"classification":421,"search_phrases":443,"slug":446,"view_count":112,"doi":447,"paper":448,"created_at":462},2335,"Deep Learning-Based Weed Classification, Detection, and Segmentation in Precision Agriculture: a Bibliometric and Comprehensive Review","https:\u002F\u002Fdoi.org\u002F10.1007\u002Fs10343-026-01424-9","Deep Learning-Based Weed Classification, Detection, and Segmentation in Precision Agriculture: a Bibliometric and Comprehensive Review。Journal of Plant Diseases and Protection","基于深度学习的精准农业杂草分类、检测与分割：文献计量与综合综述。《植物病害与保护杂志》","Journal of Plant Diseases and Protection","2026-09-12T00:00:00Z",{"impact":169,"substance":73,"depth":17,"authority":19,"freshness":76,"relevant":20,"comment":415},"核心期刊发表的深度学习杂草识别综述，方法梳理与文献计量兼具，对精准农业植保智能化有参考价值，但属学术综述而非产业级突破。",[417],{"name":412,"url":409},[81,82,26,216,27],{"code":31,"slug":32,"name":33,"path":34,"parent_code":35},[],{"status":38,"locked":14,"version":39,"method":40,"evidence":422},{"candidates":423,"agricultural_evidence":45},[424,428,432,439],{"code":50,"score":149,"supported":45,"title":425,"abstract":426,"tags":427},[101],[27,101],[81,27],{"code":31,"score":149,"supported":45,"title":429,"abstract":430,"tags":431},[92],[26,92],[82,26],{"code":433,"score":19,"supported":45,"title":434,"abstract":436,"tags":438},"I7",[435],"bibliometric",[437,435],"文献计量",[],{"code":233,"score":20,"supported":14,"title":440,"abstract":441,"tags":442},[],[],[216],[444,445],"农业人工智能 智慧农业 杂草识别 深度学习","农业人工智能 智慧农业","农业人工智能智慧农业杂草识别深度学习-2335","10.1007\u002Fs10343-026-01424-9",{"doi":447,"openalex_id":449,"authors":450,"venue":412,"cited_by_count":128,"oa_url":8,"card":457,"direction":134,"ingested_from":136},"W7212352052",[451,454],{"name":452,"orcid":453},"Furkan Ulaş","https:\u002F\u002Forcid.org\u002F0009-0002-3052-4457",{"name":455,"orcid":456},"Muhammad Aasım","https:\u002F\u002Forcid.org\u002F0000-0002-8524-9029",{"tldr":458,"method":459,"finding":460,"direction":134,"opportunity":461},"对精准农业中深度学习杂草分类、检测与分割研究进行文献计量与综述分析。","文献计量分析与综合综述，基于相关文献数据库。","梳理了深度学习在杂草识别三大任务中的方法、趋势与挑战。","可针对田间复杂光照与多物种场景，探索轻量化实时分割模型与跨数据集泛化。","2026-09-13T23:30:38.907624Z"]