[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"item-3672":3,"related-3672":53},{"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":52},3672,"Hybrid Attention Transformers for Multi-Spectral Satellite Super-Resolution","https:\u002F\u002Fdoi.org\u002F10.31224\u002F8320","Spaceborne optical imaging missions, such as the European Space Agency's Copernicus Sentinel-2 constellation, provide vital multi-spectral observations worldwide, yet optical aperture diffraction limits native Ground Sampling Distance (GSD) to 10 m across visible and near-infrared (VNIR) bands. Traditional Single-Image Super-Resolution (SISR) algorithms designed for 8-bit photographic imagery introduce severe radiometric distortions that corrupt downstream biophysical canopy analyses. In this letter, we present HAT-Light, an edge-efficient continuous-scale Hybrid Attention Transformer engineered specifically for 4-channel (RGB+NIR), 16-bit Bottom-Of-Atmosphere (BOA) surface reflectance imagery. HAT-Light addresses the limitations of standard self-attention by integrating non-overlapping Window Multi-Head Self-Attention with Depthwise Convolutional Feed-Forward Networks (DW-FFN), effectively recovering translation-equivariant localized inductive biases essential for resolving fine agricultural field parcel boundaries and airport runway geometries. Arbitrary continuous magnification (s in [2.0, 4.0]) is enabled through harmonic sinusoidal Feature-wise Linear Modulation (FiLM) within a single checkpoint. Furthermore, we enforce physical radiance conservation through a composite multi-task loss suite that unites Smooth Charbonnier regression, 2D real Fourier transform (rFFT2) spectral alignment, directional Sobel edge penalties, and a numerically bounded Convex Cosine Spectral Angle Mapper (SAM) loss ensuring FP16 numerical stability. Evaluated across 600 curated Sentinel-2 test patches spanning five distinct biomes, HAT-Light establishes state-of-the-art accuracy (33.48 dB PSNR, 0.9048 SSIM, 1.37 deg SAM, and 1.89 ERGAS), outperforming Bicubic (+1.25 dB) and RCAN (+0.71 dB). Spatial edge transect profiling and agricultural NDVI correlation (R^2 = 0.898) confirm robust biophysical conservation. Under the Wald synthesis protocol on 100 authentic 2.5 m USGS NAIP aerial patches, HAT-Light achieves superior zero-shot transfer (0.8332 SSIM) at real-time edge throughput (78.7 FPS) on an edge GPU.","星载光学成像任务，如欧洲航天局哥白尼哨兵-2星座，在全球范围内提供了重要的多光谱观测数据，然而光学孔径衍射将可见光与近红外（VNIR）波段的原生地面采样距离（GSD）限制在10 m。专为8位摄影图像设计的传统单图像超分辨率（SISR）算法会引入严重的辐射畸变，从而破坏下游的植被生物物理分析。本文提出HAT-Light，一种面向边缘高效推理的连续尺度混合注意力Transformer，专为4通道（RGB+NIR）、16位大气底层（BOA）地表反射率影像设计。HAT-Light通过将非重叠窗口多头自注意力与深度可分离卷积前馈网络（DW-FFN）相结合，解决了标准自注意力的局限性，有效恢复了平移等变的局部归纳偏置，这对于分辨精细农田地块边界和机场跑道几何形态至关重要。通过谐波正弦特征级线性调制（FiLM），在单一检查点内实现了任意连续放大倍数（s∈[2.0, 4.0]）。此外，我们通过复合多任务损失套件强制物理辐射守恒，该套件融合了平滑Charbonnier回归、二维实傅里叶变换（rFFT2）频谱对齐、方向性Sobel边缘惩罚以及数值有界的凸余弦光谱角映射（SAM）损失，确保FP16数值稳定性。在涵盖五种不同生物群系的600个精选哨兵-2测试图像块上进行评估，HAT-Light确立了最先进的精度（33.48 dB PSNR、0.9048 SSIM、1.37° SAM和1.89 ERGAS），优于双三次插值（+1.25 dB）和RCAN（+0.71 dB）。空间边缘剖面分析和农业NDVI相关性（R²=0.898）证实了稳健的生物物理守恒性。在100个真实2.5 m USGS NAIP航空图像块上按照Wald合成协议进行测试，HAT-Light在边缘GPU上以实时边缘吞吐量（78.7 FPS）实现了优异的零样本迁移性能（0.8332 SSIM）。",null,"OpenAlex","2026-09-26T00:00:00Z","论文",10,true,81,{"impact":17,"substance":18,"depth":19,"authority":20,"freshness":21,"relevant":22,"comment":23},18,22,19,13,9,1,"面向Sentinel-2多光谱16位反射率影像的超分辨率新方法，兼顾辐射保真与NDVI生物物理一致性，对农业遥感监测有实质技术增量。",[25],{"name":10,"url":6},[27,28,29,30,31],"智慧农业","农业人工智能","农业遥感","卫星影像","作物监测",[33,34],"Sentinel-2 超分辨率 农业遥感","HAT-Light 多光谱 卫星影像","Sentinel-2超分辨率农业遥感-3672",0,"10.31224\u002F8320",{"doi":37,"openalex_id":39,"authors":40,"venue":9,"cited_by_count":36,"oa_url":44,"card":45,"direction":49,"ingested_from":51},"W7214467740",[41],{"name":42,"orcid":43},"Naman Sharma","https:\u002F\u002Forcid.org\u002F0009-0001-8443-0985","https:\u002F\u002Fengrxiv.org\u002Fpreprint\u002Fdownload\u002F8320\u002F13499\u002F11736",{"tldr":46,"method":47,"finding":48,"direction":49,"opportunity":50},"提出HAT-Light混合注意力Transformer，实现Sentinel-2多光谱16位影像连续","混合窗口自注意力与深度可分离卷积前馈网络，结合FiLM连续尺度调制和多任务物理约","在600个Sentinel-2测试块上达33.48dB PSNR，NDVI相关性R²=0.898，边","农业遥感与作物表型","可探索超分后多光谱影像对作物长势、病虫害等下游农学任务的定量增益与不确定性传播。","openalex","2026-09-28T23:30:28.027834Z",{"total":54,"page":22,"page_size":54,"items":55},6,[56,89,121,149,182,222],{"id":57,"title":58,"url":59,"summary":60,"summary_zh":9,"content":9,"source_name":61,"source_url":59,"published_at":62,"category":12,"cover_url":9,"hotness":13,"is_selected":63,"score":64,"score_detail":65,"sources":70,"tags":72,"search_phrases":74,"slug":77,"view_count":36,"doi":78,"paper":79,"created_at":88},3524,"A hybrid PCA and 1D-CNN approach with regional priors for accurate agricultural monitoring: a case study with EuroCropsML","https:\u002F\u002Fdoi.org\u002F10.1007\u002Fs11119-026-10449-z","A hybrid PCA and 1D-CNN approach with regional priors for accurate agricultural monitoring: a case study with EuroCropsML。Precision Agriculture","Precision Agriculture","2026-09-25T00:00:00Z",false,70,{"impact":66,"substance":17,"depth":67,"authority":68,"freshness":21,"relevant":22,"comment":69},12,17,14,"方法类论文，提出PCA与一维CNN结合并引入区域先验，在EuroCropsML数据集上验证，对作物遥感监测有参考价值，但属细分方法进展，公共影响有限。",[71],{"name":61,"url":59},[27,28,29,73],"作物分类",[75,76],"EuroCropsML 作物分类","PCA 1D-CNN 农业监测","EuroCropsML作物分类-3524","10.1007\u002Fs11119-026-10449-z",{"doi":78,"openalex_id":80,"authors":81,"venue":61,"cited_by_count":36,"oa_url":9,"card":9,"direction":49,"ingested_from":51},"W7214386719",[82,85],{"name":83,"orcid":84},"Nourelhouda Groun","https:\u002F\u002Forcid.org\u002F0000-0002-8099-0627",{"name":86,"orcid":87},"Kheir-eddine Otmani","https:\u002F\u002Forcid.org\u002F0000-0002-7487-0928","2026-09-26T23:30:03.186961Z",{"id":90,"title":91,"url":92,"summary":93,"summary_zh":9,"content":9,"source_name":94,"source_url":9,"published_at":95,"category":12,"cover_url":9,"hotness":13,"is_selected":14,"score":96,"score_detail":97,"sources":101,"tags":103,"search_phrases":107,"slug":110,"view_count":36,"doi":9,"paper":111,"created_at":120},3125,"Full-Season Agentic Farm System FAIRY: Event-Driven Multi-Agent Orchestration for Soybean Production（FAIRY 全季节智能体农场系统：大豆生产的事件驱动多智能体编排）","https:\u002F\u002Faiagentstore.ai\u002Fai-agent-news\u002Ftopic\u002Fagriculture-food\u002F2026-08-11","哈尔滨工业大学研究人员发布并部署全栈、事件驱动的智能体引擎 FAIRY 于中国运行中的大豆研究农场。FAIRY 集成传感器、无人机、卫星植被产品、机械 API、作物过程模型与多智能体编排层，执行起垄→播种→灌溉→病虫害防治→收获→干燥→存储工作流，并在 64 垄研究场上跨 100 个全季节场景评估 9 个智能体控制器。这是智能体系统能够在大农业时间尺度和延迟结果下进行推理的最清晰演示之一，将农业中的智能体工作从实验室演示推进到全过程评估。同期 arXiv 推出 HarvestBench 基准将 LLM 驱动智能体置于农场网格世界（拖拉机面临动物选择绕行或碾压），结果显示模型差异巨大、对道德简报高度敏感、避免意愿具有可衡量的价格弹性。","Harbin Institute of Technology \u002F arXiv","2026-09-18T00:00:00Z",89,{"impact":98,"substance":99,"depth":19,"authority":68,"freshness":21,"relevant":22,"comment":100},24,23,"哈工大在真实大豆农场部署全季节事件驱动多智能体系统并配套 HarvestBench 基准，是农业智能体从演示走向全过程评估的标志性进展，专业深度与信息增量俱佳。",[102],{"name":94,"url":92},[27,104,28,29,105,106],"无人农场","多智能体","大豆生产",[108,109],"哈工大 FAIRY 大豆","HarvestBench 智能体 农场","哈工大FAIRY大豆-3125",{"doi":9,"openalex_id":9,"authors":112,"venue":9,"cited_by_count":36,"oa_url":9,"card":113,"direction":117,"ingested_from":119},[],{"tldr":114,"method":115,"finding":116,"direction":117,"opportunity":118},"部署全季节事件驱动多智能体系统FAIRY，在大豆农场完成从起垄到存储的全流程评估。","集成传感器、无人机、卫星、作物模型与机械API，用9个智能体控制器在64垄100","智能体系统能在大农业时间尺度下推理，模型差异大且对道德简报敏感。","农业人工智能与决策模型","可研究多智能体在长周期、延迟反馈农业任务中的鲁棒性与伦理约束机制。","agent","2026-09-22T00:05:38.611717Z",{"id":122,"title":123,"url":124,"summary":125,"summary_zh":9,"content":9,"source_name":126,"source_url":9,"published_at":127,"category":12,"cover_url":9,"hotness":13,"is_selected":63,"score":128,"score_detail":129,"sources":132,"tags":134,"search_phrases":137,"slug":140,"view_count":22,"doi":9,"paper":141,"created_at":148},3048,"基于双路径注意力与多尺度融合的作物病虫害识别网络DPMFNet","https:\u002F\u002Fwww.mdpi.com\u002F1099-4300\u002F28\u002F9\u002F1032","盐城工学院Hong Zhang、Fagen Song等联合江苏开放大学提出DPMFNet轻量级双路径网络，集成空间-通道双注意力（SCDA）与多尺度深度可分离卷积（MDSC）模块，构建AttMDSCBlock残差结构。在PlantVillage与AI Challenger 2018数据集上DPMFNet仅14.24M参数和2.55G FLOPs，跨注意力机制融合局部细节与全局上下文，轻量化金字塔策略自适应整合多分辨率特征，在复杂农田场景下兼顾精度与可部署性，为嵌入式田间设备提供高性价比方案。","MDPI Entropy 28(9):1032","2026-09-20T00:00:00Z",79,{"impact":130,"substance":18,"depth":17,"authority":20,"freshness":13,"relevant":22,"comment":131},16,"轻量级双路径注意力网络在两大公开数据集上兼顾精度与可部署性，对嵌入式田间设备落地有实质参考价值。",[133],{"name":126,"url":124},[27,28,135,31,136],"病虫害识别","轻量化模型",[138,139],"盐城工学院 DPMFNet 病虫害识别","PlantVillage AI Challenger 作物病害","盐城工学院DPMFNet病虫害识别-3048",{"doi":9,"openalex_id":9,"authors":142,"venue":9,"cited_by_count":36,"oa_url":9,"card":143,"direction":117,"ingested_from":119},[],{"tldr":144,"method":145,"finding":146,"direction":117,"opportunity":147},"提出轻量级双路径网络DPMFNet，实现复杂农田场景下的作物病虫害高精度识别。","空间-通道双注意力与多尺度深度可分离卷积，构建AttMDSCBlock残差结构。","仅14.24M参数、2.55G FLOPs，在PlantVillage与AI Challenger ","可探索真实田间多病虫害并发与边缘设备实时推理的轻量化自适应识别研究。","2026-09-21T00:04:39.305757Z",{"id":150,"title":151,"url":152,"summary":153,"summary_zh":154,"content":9,"source_name":155,"source_url":152,"published_at":95,"category":12,"cover_url":9,"hotness":13,"is_selected":63,"score":15,"score_detail":156,"sources":158,"tags":160,"search_phrases":163,"slug":166,"view_count":36,"doi":167,"paper":168,"created_at":181},2944,"Rule-driven functional zoning index prediction for sustainable agricultural landscapes design based on multimodal fusion deep learning","https:\u002F\u002Fdoi.org\u002F10.1038\u002Fs41598-026-72129-2","Abstract This study develops a rule-based functional zoning index system for agricultural landscapes that integrates multi-source variables, including remote sensing spectral information, landscape structure, topographic conditions, crop configuration, and land cover. To enable automated prediction of the proposed index system, a Multi-Modal Landscape Fusion Network (MMLF-Net) model is developed. By integrating multi-source data, including remote sensing images, topographic features, crop structures, and land cover, the model constructs an end-to-end AL functional zoning system. Experimental verification is carried out in the Fresno region of California, the United States of America. The results show that MMLF-Net effectively improves the accuracy of functional identification and the stability of zoning, achieving an overall accuracy of 87.34% and a Kappa coefficient of 0.84. Among all functional types, the F1 scores for intensive production and natural fallow zones exceed 0.90, demonstrating the advantages of multimodal feature fusion in functional identification. Further landscape pattern analysis reveals several key structural characteristics, including overextended production-ecology interfaces, fragmented ecological patches, and complex morphologies within composite management zones. These findings provide a quantitative basis for ecological restoration layout and agricultural pollution prevention and control. This study aims to offer a practical spatial decision-making tool for the sustainable development of regional agriculture.","摘要 本研究构建了一套基于规则的农业景观功能分区指标体系，该体系整合了多源变量，包括遥感光谱信息、景观结构、地形条件、作物配置和土地覆盖。为实现对所提指标体系的自动化预测，本研究开发了多模态景观融合网络（MMLF-Net）模型。通过整合遥感影像、地形特征、作物结构和土地覆盖等多源数据，该模型构建了端到端的农业景观功能分区系统。实验验证在美国加利福尼亚州弗雷斯诺地区开展。结果表明，MMLF-Net有效提升了功能识别的精度和分区的稳定性，总体精度达到87.34%，Kappa系数为0.84。在所有功能类型中，集约生产区和自然休耕区的F1分数均超过0.90，证明了多模态特征融合在功能识别中的优势。进一步的景观格局分析揭示了若干关键结构特征，包括生产-生态界面过度延伸、生态斑块破碎化以及复合管理区内形态复杂等。这些发现为生态修复布局和农业污染防治提供了定量依据。本研究旨在为区域农业可持续发展提供实用的空间决策工具。","Scientific Reports",{"impact":17,"substance":18,"depth":17,"authority":68,"freshness":21,"relevant":22,"comment":157},"多模态深度学习用于农业景观功能分区，方法新颖、指标可靠，对农业空间决策有参考价值。",[159],{"name":155,"url":152},[27,28,29,161,162],"多模态融合","功能分区",[164,165],"MMLF-Net 农业景观 功能分区","Fresno 农业景观 多模态","MMLF-Net农业景观功能分区-2944","10.1038\u002Fs41598-026-72129-2",{"doi":167,"openalex_id":169,"authors":170,"venue":155,"cited_by_count":36,"oa_url":152,"card":176,"direction":49,"ingested_from":51},"W7213550895",[171,173],{"name":172,"orcid":9},"Ruifen Wen",{"name":174,"orcid":175},"Juan Du","https:\u002F\u002Forcid.org\u002F0000-0002-7422-8767",{"tldr":177,"method":178,"finding":179,"direction":49,"opportunity":180},"构建规则驱动农业景观功能分区指数，并用多模态融合深度学习实现自动预测。","多模态融合网络MMLF-Net，融合遥感、地形、作物结构与土地覆盖数据。","MMLF-Net总体精度87.34%、Kappa 0.84，集约生产与自然休耕区F1超0.90。","可探索规则驱动指数与可解释深度学习的耦合，并迁移到不同农业景观区验证泛化性。","2026-09-19T23:30:32.908587Z",{"id":183,"title":184,"url":185,"summary":186,"summary_zh":187,"content":9,"source_name":188,"source_url":185,"published_at":95,"category":12,"cover_url":9,"hotness":13,"is_selected":63,"score":15,"score_detail":189,"sources":193,"tags":195,"search_phrases":198,"slug":201,"view_count":202,"doi":203,"paper":204,"created_at":221},2906,"UAV remote sensing for crop lodging monitoring: A comprehensive review","https:\u002F\u002Fdoi.org\u002F10.1016\u002Fj.compag.2026.112446","Global agriculture is facing increasing pressure to improve productivity while minimizing losses caused by extreme climate events. Crop lodging poses a major threat to food security by reducing crop yield, deteriorating grain quality, and increasing harvesting costs. As a flexible, high-resolution remote sensing platform, unmanned aerial vehicles (UAVs) have emerged as an effective solution for eliminating the gap between labor-intensive ground surveys and low-resolution satellite observations. This review systematically synthesizes the current state of research on UAV-based crop lodging monitoring through a comprehensive analysis of peer-reviewed literature. It covers the complete technical workflow, from lodging mechanisms to lodging quantification. The capabilities of diverse sensing modalities, including RGB, multispectral imagery (MSI), hyperspectral imagery (HSI), thermal infrared (TIR), light detection and ranging (LiDAR), and synthetic aperture radar (SAR), are systematically evaluated for detecting lodging-induced variations in color, texture, spectral characteristics, canopy temperature, structural attributes, and radar backscatter. In addition, recent advances in analytical approaches are reviewed, encompassing traditional statistical analyses, machine learning techniques, and state-of-the-art deep learning models. Furthermore, the review summarizes UAV-based lodging assessments across a wide range of crop species and examines methods for quantifying key lodging parameters, including lodging severity, lodging type, and affected area. By identifying current research gaps and emerging opportunities, this review outlines future research directions and provides valuable guidance for advancing accurate, efficient, and real-time crop lodging monitoring in smart agriculture.","全球农业正面临日益增大的压力，既要提高生产力，又要尽量减少极端气候事件造成的损失。作物倒伏通过降低作物产量、恶化籽粒品质并增加收获成本，对粮食安全构成重大威胁。作为一种灵活的高分辨率遥感平台，无人机（UAV）已成为弥合劳动密集型地面调查与低分辨率卫星观测之间差距的有效解决方案。本文通过系统分析同行评议文献，全面综述了基于无人机的作物倒伏监测研究现状，涵盖从倒伏机理到倒伏量化的完整技术流程。系统评估了多种传感模态的能力，包括RGB影像、多光谱影像（MSI）、高光谱影像（HSI）、热红外（TIR）、激光雷达（LiDAR）和合成孔径雷达（SAR），用于检测倒伏引起的颜色、纹理、光谱特征、冠层温度、结构属性和雷达后向散射变化。此外，综述了分析方法的最新进展，涵盖传统统计分析、机器学习技术和前沿深度学习模型。同时，本文总结了针对多种作物类型的基于无人机的倒伏评估研究，并探讨了关键倒伏参数的量化方法，包括倒伏严重程度、倒伏类型和受影响面积。通过识别当前研究空白和新兴机遇，本文展望了未来研究方向，为推进智慧农业中准确、高效、实时的作物倒伏监测提供了有价值的指导。","Computers and Electronics in Agriculture",{"impact":17,"substance":18,"depth":17,"authority":190,"freshness":191,"relevant":22,"comment":192},15,8,"核心期刊发表的无人机遥感作物倒伏监测综述，系统梳理多模态传感与深度学习分析路径，对智慧农业精准监测有实质参考价值。",[194],{"name":188,"url":185},[27,196,28,29,197],"无人机","作物倒伏",[199,200],"无人机 遥感 作物倒伏","UAV 倒伏 监测","无人机遥感作物倒伏-2906",2,"10.1016\u002Fj.compag.2026.112446",{"doi":203,"openalex_id":205,"authors":206,"venue":188,"cited_by_count":36,"oa_url":185,"card":215,"direction":220,"ingested_from":51},"W7213542473",[207,210,212],{"name":208,"orcid":209},"Dashuai Wang","https:\u002F\u002Forcid.org\u002F0000-0002-3159-7175",{"name":211,"orcid":9},"Changxing Geng",{"name":213,"orcid":214},"Xiaoguang Liu","https:\u002F\u002Forcid.org\u002F0000-0002-0935-3094",{"tldr":216,"method":217,"finding":218,"direction":49,"opportunity":219},"综述无人机遥感在作物倒伏监测中的技术流程、传感器与分析模型。","系统综述RGB、多光谱、高光谱、热红外、LiDAR、SAR及机器学习深度学习方法","多模态遥感结合深度学习可实现倒伏程度、类型与面积的精准量化。","多源数据融合与实时轻量化模型是倒伏监测的潜在突破方向。","智慧农业 \u002F 农业物联网","2026-09-19T23:30:01.740139Z",{"id":223,"title":224,"url":225,"summary":226,"summary_zh":9,"content":9,"source_name":227,"source_url":9,"published_at":228,"category":12,"cover_url":9,"hotness":13,"is_selected":63,"score":229,"score_detail":230,"sources":232,"tags":234,"search_phrases":237,"slug":240,"view_count":36,"doi":9,"paper":241,"created_at":248},2901,"AgriScope:面向农业图像的像素级多模态理解统一框架,arXiv 2609.20325(预印本)","https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.20325","Boudiaf、Alanssari、Hussain、Javed提出AgriScope,一个统一的像素级多模态农业图像理解框架,联合支持图像级、区域级、像素级理解,可实现接地描述生成、指代表达分割、多轮多模态交互等任务。集成生物专用语义表征、密集空间表征与像素解码;引入大规模像素级农业多模态指令调优数据集AgriGround,包含50万+图像和1100万+指令跟随样本,涵盖植物病害分析、作物与杂草识别、昆虫识别、细粒度植物理解。实验表明AgriScope在多项农业视觉语言任务上有效。","arXiv (preprint)","2026-09-17T00:00:00Z",75,{"impact":17,"substance":18,"depth":17,"authority":191,"freshness":21,"relevant":22,"comment":231},"提出统一像素级农业多模态理解框架并开源50万图像、1100万指令样本的大规模数据集，方法新颖、数据规模突出，但为arXiv预印本、未经同行评审，权威性有限，值得作为前沿技术动态精选。",[233],{"name":227,"url":225},[27,28,29,235,236],"植物病害识别","多模态大模型",[238,239],"AgriScope 农业图像 多模态","AgriGround 像素级 农业数据集","AgriScope农业图像多模态-2901",{"doi":9,"openalex_id":9,"authors":242,"venue":9,"cited_by_count":36,"oa_url":9,"card":243,"direction":117,"ingested_from":119},[],{"tldr":244,"method":245,"finding":246,"direction":117,"opportunity":247},"提出AgriScope统一框架，实现农业图像像素级多模态理解与多任务交互。","构建AgriGround数据集（50万+图像、1100万+指令样本），融合语义与","AgriScope在接地描述、指代分割、多轮交互等农业视觉语言任务上有效。","可探索像素级多模态模型在田间实时病害诊断与精准施药决策中的落地与轻量化。","2026-09-19T00:06:08.678379Z"]