[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"item-3176":3,"related-3176":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},3176,"Relaxing the clear-sky assumption: Cloud-tolerant spatiotemporal fusion via mask-guided feature modulation and temporal-memory collaboration","https:\u002F\u002Fdoi.org\u002F10.1016\u002Fj.jag.2026.105571","The precise characterization of dynamic Earth surface processes relies on time-series remote sensing imagery with both high spatial fidelity and frequent temporal coverage. While spatiotemporal fusion (STF) bridges this resolution trade-off by integrating complementary observations, its practical application remains constrained by the idealized ”clear-sky assumption”. This requirement often favors the selection of temporally distant cloud-free acquisitions over temporally closer but cloud-contaminated observations, potentially reducing the relevance of reference information. To address this limitation, we propose CloudSTF, a robust framework designed to achieve high-fidelity reconstruction by effectively exploiting partially occluded observations. Our approach employs a Mask-guided Multi-scale Swin Transformer (M 2 ST) encoder to capture cloud spatial patterns and suppress noise propagation during feature extraction. This is coupled with a Cross-temporal Memory-guided Fusion (CTMF) module that adaptively integrates temporal trends with textural details retrieved from a spatio-temporal memory bank. To validate this paradigm, we introduce the Global Cloud-shrouded Regions (GCR-STF) benchmark, a geographically distributed dataset comprising 24 representative agricultural and urban sites across six continents. Extensive experiments demonstrate that CloudSTF consistently outperforms state-of-the-art methods. Additional assessments across varying cloud densities and diverse landscapes further demonstrate the robustness and applicability of CloudSTF under realistic cloud-contaminated conditions. These results suggest that CloudSTF provides a reliable solution for continuous Earth monitoring in realistic, cloud-prone environments. Source code and the GCR-STF benchmark are available at https:\u002F\u002Fgithub.com\u002FSichen-Lu\u002FCloudSTF .","精确刻画动态地球表面过程依赖于兼具高空间保真度与高频时间覆盖的时间序列遥感影像。尽管时空融合（STF）通过整合互补观测弥合了这一分辨率之间的权衡，但其实际应用仍受限于理想化的“晴空假设”。该要求往往倾向于选择时间上相距较远但无云的影像，而非时间上更近但受云污染的观测，这可能降低参考信息的相关性。为解决这一局限，我们提出了CloudSTF，一个旨在通过有效利用部分被遮挡的观测来实现高保真重建的稳健框架。我们的方法采用掩膜引导的多尺度Swin Transformer（M²ST）编码器来捕捉云的空间模式并抑制特征提取过程中的噪声传播。该编码器与跨时间记忆引导融合（CTMF）模块相结合，后者自适应地将时间趋势与从时空记忆库中检索到的纹理细节进行整合。为验证这一范式，我们引入了全球云覆盖区域（GCR-STF）基准数据集，这是一个地理分布广泛的数据集，涵盖六大洲24个代表性农业和城市站点。大量实验表明，CloudSTF始终优于现有最先进方法。在不同云密度和多样景观下的额外评估进一步证明了CloudSTF在真实云污染条件下的稳健性和适用性。这些结果表明，CloudSTF为真实且多云环境中的连续地球监测提供了可靠解决方案。源代码和GCR-STF基准数据集可在https:\u002F\u002Fgithub.com\u002FSichen-Lu\u002FCloudSTF获取。",null,"International Journal of Applied Earth Observation and Geoinformation","2026-09-21T00:00:00Z","论文",10,false,80,{"impact":17,"substance":18,"depth":19,"authority":20,"freshness":21,"relevant":22,"comment":23},16,22,19,14,9,1,"提出抗云时空融合框架CloudSTF并发布全球云覆盖基准数据集，方法新颖、数据规模大，对多云地区农业遥感监测有实用价值。",[25],{"name":10,"url":6},[27,28,29,30,31],"智慧农业","农业人工智能","遥感","时空融合","云污染",[33,34],"CloudSTF 时空融合","GCR-STF 云覆盖基准数据集","CloudSTF时空融合-3176",0,"10.1016\u002Fj.jag.2026.105571",{"doi":37,"openalex_id":39,"authors":40,"venue":10,"cited_by_count":36,"oa_url":6,"card":57,"direction":61,"ingested_from":63},"W7213903535",[41,44,47,49,52,54],{"name":42,"orcid":43},"Sichen Lu","https:\u002F\u002Forcid.org\u002F0009-0009-3215-6521",{"name":45,"orcid":46},"Juanjuan Jing","https:\u002F\u002Forcid.org\u002F0009-0002-0371-7245",{"name":48,"orcid":9},"Junhua Yu",{"name":50,"orcid":51},"Lei Yang","https:\u002F\u002Forcid.org\u002F0000-0001-8297-0868",{"name":53,"orcid":9},"Boyang Nie",{"name":55,"orcid":56},"Jinsong Zhou","https:\u002F\u002Forcid.org\u002F0009-0006-4704-0685",{"tldr":58,"method":59,"finding":60,"direction":61,"opportunity":62},"提出CloudSTF框架，利用含云影像实现高保真时空融合，突破晴空假设。","掩膜引导多尺度Swin Transformer编码器与跨时记忆融合模块，构建GC","在六个大洲24个站点上优于现有方法，不同云密度下均鲁棒。","农业遥感与作物表型","可探索云污染下融合结果对作物长势监测与产量估测的精度影响。","openalex","2026-09-22T23:30:23.475701Z",{"total":66,"page":22,"page_size":66,"items":67},6,[68,98,124,160,205,240],{"id":69,"title":70,"url":71,"summary":72,"summary_zh":9,"content":9,"source_name":73,"source_url":9,"published_at":74,"category":12,"cover_url":9,"hotness":13,"is_selected":14,"score":75,"score_detail":76,"sources":80,"tags":82,"search_phrases":85,"slug":88,"view_count":36,"doi":9,"paper":89,"created_at":97},3250,"农机化研究2026(11)：基于多源感知与语义分割的果树行识别与路径跟踪控制——郑州工业应用技术学院许洋洋等 DeepLabv3+ F1达0.94","http:\u002F\u002Fwww.qikanvip.com\u002Fqkml\u002F166268.html","农机化研究2026年第11期发表郑州工业应用技术学院许洋洋、柴亚珂、王莹、牛瑞利团队论文：针对果园中裸露土壤、遮蔽物干扰和冠层阴影等因素导致导航路径难以稳定提取的问题，提出融合多源感知与语义分割的自走式植保机果树识别与路径控制方法。首先利用无人机获取多光谱影像并生成正射影像（DOM）与数字表面模型（DSM），计算归一化差异绿度指数（NDGI），增强果树冠层与背景的可分性；然后采用U-Net\u002FDeepLabv3+语义分割网络在保留空间上下文信息的同时实现果树行\u002F背景端到端分割。田间对比试验结果表明，DeepLabv3+整体分割F1可达0.94，每样区生成导航线断裂1.2次，且转向工况下横向误差RMSE为0.11 m，显著优于传统阈值分割和SVM方法。","农机化研究","2026-09-22T00:00:00Z",79,{"impact":17,"substance":18,"depth":77,"authority":78,"freshness":13,"relevant":22,"comment":79},18,13,"多源遥感与语义分割结合的果园导航研究，F1达0.94且横向误差0.11米，方法新颖、数据扎实，具备行业参考价值。",[81],{"name":73,"url":71},[27,28,83,29,84],"农机导航","果园植保",[86,87],"郑州工业应用技术学院 果树行识别","DeepLabv3 果园导航 路径跟踪","郑州工业应用技术学院果树行识别-3250",{"doi":9,"openalex_id":9,"authors":90,"venue":9,"cited_by_count":36,"oa_url":9,"card":91,"direction":61,"ingested_from":96},[],{"tldr":92,"method":93,"finding":94,"direction":61,"opportunity":95},"提出融合无人机多光谱与语义分割的果树行识别与路径跟踪方法，实现植保机自主导航。","无人机多光谱DOM\u002FDSM与NDGI，U-Net\u002FDeepLabv3+语义分割，","DeepLabv3+分割F1达0.94，导航线断裂1.2次，转向横向误差RMSE 0.11 m，优于","可探索多光谱与DSM特征融合的轻量化分割模型，并迁移至多树种、多季节果园导航。","agent","2026-09-23T00:04:33.496233Z",{"id":99,"title":100,"url":101,"summary":102,"summary_zh":9,"content":9,"source_name":103,"source_url":9,"published_at":74,"category":12,"cover_url":9,"hotness":13,"is_selected":14,"score":104,"score_detail":105,"sources":107,"tags":109,"search_phrases":112,"slug":115,"view_count":36,"doi":9,"paper":116,"created_at":123},3246,"Geospatial Artificial Intelligence in Precision Agriculture: A Systematic Review（精准农业中的地理空间人工智能：系统综述）","https:\u002F\u002Fwww.mdpi.com\u002F3043-1204\u002F1\u002F1\u002F4","MDPI AI Precis. Agric. 1(1) 4 发表系统综述：综述2019-2026年Scopus、Web of Science及补充检索文献，考察精准农业中的应用、多模态数据集成与决策支持。综述表明研究范式从孤立的制图与预测转向集成工作流，整合卫星与无人机影像、田间传感器、气象、土壤与管理记录。报告效益包括产量预测改善、压力与病害更早检测、土壤制图更精细、灌溉与投入品更精准。然而，产量、环境和经济效益仍依赖当地条件和运行验证。核心发现是：地理参考数据本身不能确保真正的地理空间AI；可靠工作流必须解决空间依赖性、尺度不匹配、可迁移性和不确定性。可解释性和融入农场运营对可执行推荐同样重要。采用仍受数据互操作性、连接性、可负担性、隐私和技术能力限制。","MDPI AI in Precision Agriculture",82,{"impact":77,"substance":18,"depth":77,"authority":20,"freshness":13,"relevant":22,"comment":106},"系统综述梳理2019-2026年GeoAI在精准农业的集成工作流与落地瓶颈，结论扎实、时效性强，对智慧农业技术路线有参考价值。",[108],{"name":103,"url":101},[27,28,29,110,111],"决策支持","多模态数据",[113,114],"GeoAI 精准农业 系统综述","地理空间人工智能 精准农业","GeoAI精准农业系统综述-3246",{"doi":9,"openalex_id":9,"authors":117,"venue":9,"cited_by_count":36,"oa_url":9,"card":118,"direction":61,"ingested_from":96},[],{"tldr":119,"method":120,"finding":121,"direction":61,"opportunity":122},"系统综述2019-2026年地理空间AI在精准农业的应用、多模态数据集成与决策支持。","系统综述Scopus、Web of Science等文献，分析多模态数据集成与决","地理参考数据不等于地理空间AI，可靠工作流须解决空间依赖、尺度、可迁移性与不确定性。","可研究跨尺度空间依赖建模与可迁移性评估，提升模型在不同农场条件下的泛化能力。","2026-09-23T00:04:33.172821Z",{"id":125,"title":126,"url":127,"summary":128,"summary_zh":129,"content":9,"source_name":130,"source_url":127,"published_at":11,"category":12,"cover_url":9,"hotness":13,"is_selected":14,"score":15,"score_detail":131,"sources":133,"tags":135,"search_phrases":138,"slug":141,"view_count":36,"doi":142,"paper":143,"created_at":159},3164,"Spatiotemporal Deep Learning for Rice Plant Height Estimation from Multi-Temporal UAV RGB Imagery","https:\u002F\u002Fdoi.org\u002F10.3390\u002Fagriculture16182034","Accurate plant height estimation is important for monitoring crop growth and supporting precision agricultural management. Manual measurements are labor-intensive, while LiDAR-based methods are expensive and require complex processing. UAV photogrammetry provides a lower-cost alternative but remains challenging in flooded rice paddies because of canopy deformation and difficulties in terrain extraction. This study proposes Rice-STNet, a spatiotemporal deep learning framework for end-to-end rice plant height estimation using multi-temporal UAV RGB imagery. Rice-STNet integrates a convolutional neural network for spatial feature extraction, Time2Vec for temporal encoding, and a gated recurrent unit network for modeling temporal dependencies across observation dates. The framework was evaluated using field data collected from rice paddies over two growing seasons. Rice-STNet achieved an R2 of 0.97, a root mean squared error of 1.97 cm, and a mean absolute error of 1.14 cm. It outperformed random forest, support vector regression, a CNN-only baseline, and a UAV photogrammetry-based point-cloud approach. In addition, the framework generated high-resolution plant height maps for field-scale analysis of spatial growth variability. These results underscore the importance of jointly modeling spatial and temporal characteristics for continuously evolving crop traits. The proposed framework offers an accurate, scalable, and non-destructive solution for large-scale crop phenotyping and precision agriculture.","准确的株高估算对于监测作物生长和支持精准农业管理具有重要意义。人工测量劳动强度大，而基于激光雷达（LiDAR）的方法成本高昂且需要复杂的处理。无人机摄影测量提供了一种成本较低的替代方案，但在淹水稻田中仍面临挑战，原因在于冠层变形和地形提取困难。本研究提出了Rice-STNet，一种时空深度学习框架，用于利用多时相无人机RGB影像进行端到端水稻株高估算。Rice-STNet集成了用于空间特征提取的卷积神经网络、用于时间编码的Time2Vec，以及用于建模观测日期之间时间依赖关系的门控循环单元网络。该框架利用两个生长季从稻田采集的田间数据进行了评估。Rice-STNet取得了R²为0.97、均方根误差为1.97 cm、平均绝对误差为1.14 cm的结果。其性能优于随机森林、支持向量回归、仅使用CNN的基线方法以及基于无人机摄影测量的点云方法。此外，该框架生成了高分辨率株高图，用于田块尺度空间生长变异性分析。这些结果凸显了联合建模空间与时间特征对于持续变化的作物性状的重要性。所提出的框架为大规模作物表型分析和精准农业提供了一种准确、可扩展且非破坏性的解决方案。","Agriculture",{"impact":77,"substance":18,"depth":77,"authority":78,"freshness":21,"relevant":22,"comment":132},"提出时空深度学习框架Rice-STNet，用多时相无人机RGB影像实现水稻株高高精度估算，方法新颖、数据跨两个生长季，对作物表型与精准农业有实用价值。",[134],{"name":130,"url":127},[27,28,136,29,137],"水稻","作物表型",[139,140],"无人机 RGB 水稻株高","Rice-STNet 水稻表型","无人机RGB水稻株高-3164","10.3390\u002Fagriculture16182034",{"doi":142,"openalex_id":144,"authors":145,"venue":130,"cited_by_count":36,"oa_url":127,"card":154,"direction":61,"ingested_from":63},"W7213887432",[146,149,151],{"name":147,"orcid":148},"Weiguo Wang","https:\u002F\u002Forcid.org\u002F0009-0003-4028-9363",{"name":150,"orcid":9},"Noboru Noguchi",{"name":152,"orcid":153},"Liangliang Yang","https:\u002F\u002Forcid.org\u002F0000-0002-5055-3987",{"tldr":155,"method":156,"finding":157,"direction":61,"opportunity":158},"提出Rice-STNet时空深度学习框架，用多时相无人机RGB影像估算水稻株高。","CNN提取空间特征，Time2Vec编码时间，GRU建模时序依赖，两季稻田数据验","R²达0.97、RMSE 1.97cm，优于随机森林、SVR、纯CNN及点云方法。","可迁移至其他作物与多源遥感融合，探索轻量化模型及实时田间部署。","2026-09-22T23:30:18.545958Z",{"id":161,"title":162,"url":163,"summary":164,"summary_zh":165,"content":9,"source_name":166,"source_url":163,"published_at":74,"category":12,"cover_url":9,"hotness":13,"is_selected":14,"score":167,"score_detail":168,"sources":173,"tags":175,"search_phrases":178,"slug":181,"view_count":36,"doi":182,"paper":183,"created_at":204},3155,"Time Series Extrinsic Regression for Smart Agriculture: Field‐Level Crop Yield Prediction From Multisource Agricultural Time Series","https:\u002F\u002Fdoi.org\u002F10.1002\u002Fqre.70406","ABSTRACT The growing availability of temporal data in agriculture has created new opportunities for data‐driven decision support systems aimed at improving the efficiency, sustainability, and adaptability of agricultural systems. Remote sensing platforms, weather‐related datasets, and field‐management records provide information on crop development and environmental conditions throughout the growing season. In this context, predictive tasks such as crop yield prediction, irrigation requirement assessment, and nutrient demand estimation can be naturally formulated as Time Series Extrinsic Regression (TSER), a supervised learning task in which a continuous target is predicted from one or more time series. This paper presents TSER within the Smart Agriculture domain and investigates its applicability to field‐level crop yield prediction using multisource agricultural time series. The study adopts a field‐based temporal approach where Sentinel‐2 observations are processed to derive vegetation‐index time series, spatially aggregated at parcel level, and integrated with ERA5‐Land weather‐related variables and static vineyard attributes. Representative neural and non‐neural TSER methods are compared under temporal and spatial generalization scenarios. Results show that TSER provides a suitable supervised learning framework for crop yield prediction from heterogeneous agricultural time series. Conditional LSTM achieves the best overall performance under temporal generalization across vintages, while non‐neural methods remain competitive under spatial generalization to unseen vineyard blocks. These findings suggest that TSER can support field‐level predictive modeling in Smart Agriculture, with model choice depending on the target generalization setting, data availability, and operational constraints.","摘要 农业中时间数据的日益丰富为数据驱动的决策支持系统创造了新的机遇，旨在提高农业系统的效率、可持续性和适应性。遥感平台、气象相关数据集和田间管理记录提供了整个生长季作物发育和环境状况的信息。在此背景下，作物产量预测、灌溉需求评估和养分需求估算等预测任务可以自然地表述为时间序列外在回归（TSER），这是一种监督学习任务，即从一个或多个时间序列中预测连续目标变量。本文在智慧农业领域内介绍TSER，并研究其利用多源农业时间序列进行田块级作物产量预测的适用性。本研究采用基于田块的时间方法，对Sentinel-2观测数据进行处理以导出植被指数时间序列，在地块尺度上进行空间聚合，并与ERA5-Land气象相关变量及静态葡萄园属性相整合。在时间泛化和空间泛化情景下，比较了具有代表性的神经与非神经TSER方法。结果表明，TSER为从异质农业时间序列进行作物产量预测提供了一个合适的监督学习框架。条件LSTM在跨年份的时间泛化下取得了最佳整体性能，而非神经方法在向未见葡萄园地块的空间泛化下仍具有竞争力。这些发现表明，TSER能够支持智慧农业中的田块级预测建模，模型选择取决于目标泛化设置、数据可用性和操作约束。","Quality and Reliability Engineering International",72,{"impact":169,"substance":170,"depth":171,"authority":78,"freshness":13,"relevant":22,"comment":172},12,20,17,"将时间序列外在回归引入田块级产量预测，方法新颖、结论有对比依据，但属学术论文，产业影响有限。",[174],{"name":166,"url":163},[27,28,29,176,177],"作物产量预测","时间序列",[179,180],"Sentinel-2 作物产量预测","TSER 智慧农业","Sentinel-2作物产量预测-3155","10.1002\u002Fqre.70406",{"doi":182,"openalex_id":184,"authors":185,"venue":166,"cited_by_count":36,"oa_url":163,"card":198,"direction":203,"ingested_from":63},"W7213984793",[186,189,192,194,196],{"name":187,"orcid":188},"Amalia Vanacore","https:\u002F\u002Forcid.org\u002F0000-0002-0320-5541",{"name":190,"orcid":191},"Armando Ciardiello","https:\u002F\u002Forcid.org\u002F0009-0005-7272-9663",{"name":193,"orcid":9},"Gennaro Pio Auricchio",{"name":195,"orcid":9},"Luigi Uccello",{"name":197,"orcid":9},"Annalisa Izzo",{"tldr":199,"method":200,"finding":201,"direction":61,"opportunity":202},"将田块级作物产量预测建模为时间序列外回归，比较多源农业时间序列上的神经与非神经方法。","Sentinel-2植被指数、ERA5-Land气象与葡萄园静态属性，对比LST","条件LSTM在跨年份时间泛化最优，非神经方法在空间泛化到新地块时仍有竞争力。","可探索多作物、多区域下TSER模型迁移性与不确定性量化，及与农事决策系统耦合。","智慧农业 \u002F 农业物联网","2026-09-22T23:30:10.381662Z",{"id":206,"title":207,"url":208,"summary":209,"summary_zh":210,"content":9,"source_name":211,"source_url":208,"published_at":212,"category":12,"cover_url":9,"hotness":13,"is_selected":14,"score":167,"score_detail":213,"sources":216,"tags":218,"search_phrases":220,"slug":223,"view_count":36,"doi":224,"paper":225,"created_at":239},3009,"AgriMAC: An Attention Based Multimodal Deep Clustering Framework for Rice Health Assessment","https:\u002F\u002Fdoi.org\u002F10.22266\u002Fijies2026.1031.06","Rice is Indonesia's staple crop, yet its productivity has declined in recent years because pest and disease outbreaks remain difficult to detect at an early stage.Existing precision agriculture approaches commonly process Internet of Things (IoT) sensor data and remote sensing imagery independently and often rely on supervised learning, requiring large amounts of labeled data.Meanwhile, multispectral drone imagery producing the Normalized Difference Vegetation Index (NDVI) provides richer information on crop physiological conditions than RGB-based vegetation indices.This study proposes Agricultural Multimodal Attention Clustering (AgriMAC), an unsupervised framework that integrates UAV derived NDVI imagery, 7-in-1 IoT soil sensor measurements, and historical weather data from the Open-Meteo API for rice field condition monitoring.Each modality is encoded using a dedicated autoencoder and fused through an entropy-regularized attention mechanism before Deep Embedded Clustering is performed.To reduce the influence of crop growth stage, the IoT representation is residualized using growth-phase statistics estimated exclusively from the training fold, enabling the discovered clusters to represent within-phase agronomic conditions rather than crop age.Experiments conducted under a grouped leave-one-field-out protocol produced a Silhouette Score of 0.465 ± 0.048, a Davies Bouldin Index of 0.807 ± 0.036, and a Calinski Harabasz Index of 195 ± 27.The learned groups also showed low normalized mutual information with growth phase (0.079 ± 0.043) and near chance phase decodability (balanced accuracy = 0.554 ± 0.042), indicating that they are only weakly associated with crop growth stage.The learned attention weights identified IoT soil measurements (0.570 ± 0.024) as the dominant modality, while NDVI imagery (0.210 ± 0.014) and weather information (0.220 ± 0.014) provided complementary spatial and temporal context.Overall, AgriMAC provides an interpretable and leakage-aware framework for multimodal clustering of rice field conditions.Although its clustering performance is comparable to that of a capacity-matched IoT-only model, it additionally quantifies the contribution of each sensing modality through attention weights and explicitly mitigates the growth-phase confound, making it suitable for field level agronomic condition monitoring and spatial decision support in precision agriculture.","水稻是印度尼西亚的主要作物，但近年来其生产力有所下降，因为病虫害暴发在早期阶段仍难以检测。现有的精准农业方法通常独立处理物联网（IoT）传感器数据和遥感影像，且往往依赖监督学习，需要大量标注数据。与此同时，生成归一化植被指数（NDVI）的多光谱无人机影像比基于RGB的植被指数能提供更丰富的作物生理状况信息。本研究提出农业多模态注意力聚类（AgriMAC），这是一个无监督框架，整合了无人机获取的NDVI影像、七合一IoT土壤传感器测量数据以及来自Open-Meteo API的历史天气数据，用于稻田状况监测。每种模态均使用专用自编码器进行编码，并通过熵正则化注意力机制进行融合，随后执行深度嵌入聚类。为减少作物生长阶段的影响，IoT表征利用仅从训练折估计的生长阶段统计量进行残差化处理，使发现的聚类能够表征阶段内的农艺状况而非作物年龄。在分组留一田块协议下进行的实验产生了0.465 ± 0.048的轮廓系数、0.807 ± 0.036的Davies-Bouldin指数和195 ± 27的Calinski-Harabasz指数。学习到的分组还显示出与生长阶段的低归一化互信息（0.079 ± 0.043）以及接近随机的阶段可解码性（平衡准确率 = 0.554 ± 0.042），表明它们与作物生长阶段仅存在弱关联。学习到的注意力权重将IoT土壤测量（0.570 ± 0.024）识别为主导模态，而NDVI影像（0.210 ± 0.014）和天气信息（0.220 ± 0.014）则提供了互补的空间和时间背景。总体而言，AgriMAC为稻田状况的多模态聚类提供了一个可解释且感知数据泄漏的框架。尽管其聚类性能与容量匹配的仅IoT模型相当，但它还通过注意力权重量化了每种传感模态的贡献，并明确减轻了生长阶段混杂因素，使其适用于田块级农艺状况监测和精准农业中的空间决策支持。","International journal of intelligent engineering and systems","2026-09-19T00:00:00Z",{"impact":169,"substance":214,"depth":171,"authority":78,"freshness":21,"relevant":22,"comment":215},21,"提出无监督多模态注意力聚类框架，融合无人机NDVI、IoT土壤与气象数据评估水稻健康，方法新颖且实验严谨，对精准农业田间监测有参考价值。",[217],{"name":211,"url":208},[27,28,136,219,29],"多模态融合",[221,222],"AgriMAC 水稻 多模态聚类","无人机 NDVI 水稻 病害监测","AgriMAC水稻多模态聚类-3009","10.22266\u002Fijies2026.1031.06",{"doi":224,"openalex_id":226,"authors":227,"venue":211,"cited_by_count":36,"oa_url":208,"card":234,"direction":203,"ingested_from":63},"W7213619014",[228,230,232],{"name":229,"orcid":9},"Nurfadhilah Mardianti Andini",{"name":231,"orcid":9},"Mike Yuliana",{"name":233,"orcid":9},"Moch. Zen Samsono Hadi",{"tldr":235,"method":236,"finding":237,"direction":203,"opportunity":238},"提出无监督多模态聚类框架AgriMAC，融合无人机NDVI、IoT土壤与气象数据评估水稻健康。","各模态自编码器编码，熵正则注意力融合，深度嵌入聚类，按生长阶段残差化。","聚类性能与仅IoT模型相当，但注意力权重可解释模态贡献并弱化生长阶段混淆。","可探索注意力融合机制在更多作物与传感器组合下的泛化性，并引入时序动态聚类。","2026-09-20T23:30:08.419613Z",{"id":241,"title":242,"url":243,"summary":244,"summary_zh":245,"content":9,"source_name":246,"source_url":243,"published_at":212,"category":12,"cover_url":9,"hotness":13,"is_selected":14,"score":247,"score_detail":248,"sources":250,"tags":252,"search_phrases":255,"slug":258,"view_count":36,"doi":259,"paper":260,"created_at":293},3006,"Explainable spectral–image fusion multi-task learning for maize canopy biochemical and structural trait retrieval","https:\u002F\u002Fdoi.org\u002F10.1016\u002Fj.compag.2026.112450","Accurate and interpretable estimation of multiple maize canopy traits is important for proximal crop monitoring and high-throughput phenotyping. This study developed a spectral–image fusion multi-task learning framework (MM-MTL) for the simultaneous retrieval of chlorophyll index (Chl index), leaf area index (LAI), nitrogen balance index (NBI), and anthocyanin index (Anth index). Proximal hyperspectral observations were collected using a Specim IQ camera covering 400–1000 nm with 204 spectral bands. For each observation, an ROI-mean spectral vector was extracted to retain fine-grained canopy reflectance information, while a co-registered pseudo-RGB image derived from visible bands of the same hyperspectral cube preserved two-dimensional canopy structure and visible appearance. MM-MTL integrates spectral and image feature extraction, task-specific fusion, and uncertainty-weighted multi-task learning to jointly estimate the four traits. A total of 1,023 valid samples collected from nine field campaigns across the 2024 and 2025 growing seasons were used for model development and evaluation. Under plot-grouped five-fold cross-validation, MM-MTL achieved R 2 values of 0.872, 0.885, 0.722, and 0.807 for Chl index, LAI, NBI, and Anth index, respectively, and consistently outperformed the single-task, single-representation, and conventional regression baselines. Performance decreased under more challenging generalization settings, with R 2 values ranging from 0.543 to 0.680 under leave-one-campaign-out validation and from 0.425 to 0.640 under bidirectional cross-year validation. Ablation and post-hoc analyses further showed that preserving image spatial organization improved prediction, while task-wise routing, input-representation masking, wavelength perturbation, and cross-fold stability analyses revealed trait-dependent use of spectral and spatial information. These results demonstrate that complementary spectral and spatial representations derived from the same hyperspectral observation can improve multi-trait maize canopy retrieval, while transfer across acquisition campaigns, years, and field environments remains an important direction for further improvement.","准确且可解释地估算多个玉米冠层性状，对于近地作物监测和高通量表型分析具有重要意义。本研究开发了一种光谱—图像融合多任务学习框架（MM-MTL），用于同时反演叶绿素指数（Chl index）、叶面积指数（LAI）、氮平衡指数（NBI）和花青素指数（Anth index）。近地高光谱观测使用Specim IQ相机采集，覆盖400–1000 nm，共204个光谱波段。对于每次观测，提取ROI均值光谱向量以保留细粒度冠层反射率信息，同时从同一高光谱立方体的可见光波段生成配准的伪RGB图像，以保留二维冠层结构和可见外观。MM-MTL集成了光谱与图像特征提取、任务特定融合以及不确定性加权多任务学习，以联合估算这四种性状。研究使用2024年和2025年生长季9次田间试验采集的共计1,023个有效样本进行模型开发与评估。在按小区分组的五折交叉验证下，MM-MTL对Chl index、LAI、NBI和Anth index的R²分别为0.872、0.885、0.722和0.807，且持续优于单任务、单表征和传统回归基线。在更具挑战性的泛化设置下，模型性能有所下降，留一试验验证的R²范围为0.543–0.680，双向跨年验证的R²范围为0.425–0.640。消融分析和事后分析进一步表明，保留图像空间组织可提升预测性能，而任务路由、输入表征掩蔽、波长扰动和跨折稳定性分析揭示了光谱与空间信息的性状依赖性利用方式。这些结果表明，从同一高光谱观测中提取的互补光谱与空间表征可改善多性状玉米冠层反演，而跨采集试验、年份和田间环境的迁移仍是未来改进的重要方向。","Computers and Electronics in Agriculture",81,{"impact":77,"substance":18,"depth":77,"authority":20,"freshness":21,"relevant":22,"comment":249},"提出光谱-图像融合多任务学习框架，1023份样本跨两年验证，方法新颖且结论可靠，对作物高通量表型研究有实质参考价值。",[251],{"name":246,"url":243},[27,28,253,29,254],"玉米","高通量表型",[256,257],"玉米冠层 多任务学习 高光谱","Specim IQ 玉米 表型","玉米冠层多任务学习高光谱-3006","10.1016\u002Fj.compag.2026.112450",{"doi":259,"openalex_id":261,"authors":262,"venue":246,"cited_by_count":36,"oa_url":243,"card":288,"direction":61,"ingested_from":63},"W7213658705",[263,265,267,269,271,274,276,279,281,283,285],{"name":264,"orcid":9},"Penglei Zhang",{"name":266,"orcid":9},"Tianbo Hao",{"name":268,"orcid":9},"Zhuoyuan Zhao",{"name":270,"orcid":9},"Hong Sun",{"name":272,"orcid":273},"Yelu Zeng","https:\u002F\u002Forcid.org\u002F0000-0003-4267-1841",{"name":275,"orcid":9},"Zheng Cui",{"name":277,"orcid":278},"Ta Na","https:\u002F\u002Forcid.org\u002F0000-0002-1348-5655",{"name":280,"orcid":9},"Lang Qiao",{"name":282,"orcid":9},"Durval Dourado Neto",{"name":284,"orcid":9},"Feng Yang",{"name":286,"orcid":287},"Jingzhu Wu","https:\u002F\u002Forcid.org\u002F0000-0002-8386-1038",{"tldr":289,"method":290,"finding":291,"direction":61,"opportunity":292},"提出光谱-图像融合多任务学习框架，同时反演玉米冠层四个生化与结构性状。","Specim IQ高光谱与伪RGB融合，不确定性加权多任务学习，1023样本交叉","融合模型精度优于单任务基线，但跨年份与跨环境泛化性能明显下降。","可研究跨年份\u002F跨环境域适应与迁移学习，提升多性状反演泛化能力。","2026-09-20T23:30:01.739142Z"]