[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"item-909":3,"related-909":48},{"id":4,"title":5,"url":6,"summary":7,"summary_zh":8,"content":9,"source_name":10,"source_url":8,"published_at":8,"category":11,"cover_url":8,"hotness":12,"is_selected":13,"score":14,"score_detail":15,"sources":22,"tags":24,"primary_topic":8,"secondary_topics":30,"classification":31,"search_phrases":43,"slug":46,"view_count":20,"doi":8,"paper":8,"created_at":47},909,"基于DINOv3动态LoRA微调的牧场多组分生物量估算","https:\u002F\u002Fwww.mdpi.com\u002F1424-8220\u002F26\u002F16\u002F5285","美国东北大学Khoury计算机学院与Nace AI联合提出基于DINOv3 Vision Transformer的参数高效多输出回归框架，从顶视高分辨率牧场图像预测五类生物量组分（干绿、干死、干三叶、绿干物质、总干生物量）。采用动态深度感知LoRA适配策略：浅层（编码通用视觉）极少扰动，深层（任务特定）强适配；同时将矩形图像拆为两半并通过对比对齐损失处理几何差异。在CSIRO Image2Biomass基准上交叉验证加权R²达0.81。",null,"## 1. Introduction\n\nPasture biomass estimation is a critical task in livestock production systems and precision agriculture. Accurate knowledge of the quantity and composition of available feed directly influences decisions on stocking rates, rotational grazing schedules, and pasture recovery management. Grazing systems cover approximately 25% of the global land surface [[1](https:\u002F\u002Fwww.mdpi.com\u002F1424-8220\u002F26\u002F16\u002F5285#B1-sensors-26-05285)], and suboptimal feed management leads to economic losses and environmental degradation, including soil erosion, biodiversity loss, and carbon depletion.\n\nExisting methods for biomass quantification each present significant limitations. The destructive clip-and-weigh technique, while accurate, is labor-intensive and infeasible at scale. Non-destructive proximal sensors such as rising plate meters and capacitance probes offer faster readings but suffer from poor reliability under heterogeneous vegetation conditions and cannot differentiate biomass by species or condition [[1](https:\u002F\u002Fwww.mdpi.com\u002F1424-8220\u002F26\u002F16\u002F5285#B1-sensors-26-05285)]. Satellite-based remote sensing enables broad-scale monitoring but lacks the spatial resolution to capture fine-grained compositional information and still requires manual calibration.\n\nThe specific problem addressed in this paper is the simultaneous estimation of five pasture biomass components from high-resolution top-view RGB images under conditions of limited labeled data, non-standard image geometry, and significant distributional variation across geographic regions and seasons. This multi-output regression task is complicated by heavily right-skewed target distributions, zero-inflated clover measurements, and a training set of only 357 annotated images spanning 19 sites across four Australian states.\n\nLarge pretrained Vision Transformers (ViTs) such as DINOv3 [[2](https:\u002F\u002Fwww.mdpi.com\u002F1424-8220\u002F26\u002F16\u002F5285#B2-sensors-26-05285)] offer powerful visual representations learned from billions of images. However, deploying these models on small downstream datasets presents a fundamental tension: full fine-tuning updates hundreds of millions of parameters from fewer than 400 labeled examples, making overfitting nearly inevitable, while keeping the backbone entirely frozen forgoes the opportunity to adapt representations to domain-specific patterns such as vegetation density, canopy texture, and species-level discrimination. This tension is especially acute for ViTs, whose self-attention layers learn highly distributed representations that are sensitive to even small perturbations in early layers.\n\nLow-Rank Adaptation (LoRA) [[3](https:\u002F\u002Fwww.mdpi.com\u002F1424-8220\u002F26\u002F16\u002F5285#B3-sensors-26-05285)] offers a principled middle ground by constraining weight updates to low-rank subspaces, dramatically reducing trainable parameter count while preserving the pretrained structure. However, standard LoRA applies a uniform rank across all layers, treating early and late transformer blocks identically. This ignores a well-established property of deep networks: lower layers encode generic, broadly transferable features (edges, textures, color statistics), while deeper layers encode abstract, task-specific representations [[4](https:\u002F\u002Fwww.mdpi.com\u002F1424-8220\u002F26\u002F16\u002F5285#B4-sensors-26-05285)]. In a low-data regime, uniform adaptation allocates capacity to early layers that may need little change and may under-adapt late layers where domain-specific patterns must be learned.\n\nThe central question of this paper is: how should adaptation capacity be distributed across transformer layers when fine-tuning a large pretrained ViT under severe data scarcity? We propose a dynamic, depth-aware LoRA schedule in which the rank r and scaling factor $\\alpha$𝛼 grow exponentially from the first to the last transformer layer. This provides smooth, continuous control over the degree of adaptation: near-zero perturbation of early layers that encode reusable visual primitives, and progressively stronger specialization in deeper layers where task-relevant compositional patterns must be formed. We observe that this design choice yields a consistent improvement in weighted $R^{2}$𝑅 2 over uniform LoRA, and that it helps preserve pretrained feature quality that otherwise degrades under full fine-tuning and aggressive uniform adaptation in low-data settings.\n\nTo instantiate and evaluate this adaptation strategy, we build a complete pipeline for pasture biomass estimation that additionally addresses two practical challenges. First, the dataset images have a 2:1 aspect ratio that conflicts with the square input assumption of pretrained ViTs; we partition each image into two square halves processed as a dual-view stream with a contrastive alignment loss enforcing feature coherence. Second, we introduce a progressive resolution training strategy in which LoRA adapters are trained and merged at successively higher resolutions ($384 \\rightarrow 512 \\rightarrow 768$384→512→768), which stabilizes the optimization landscape for adapter training. The final system incorporates metadata integration, test time augmentation, and multi backbone ensembling as further supporting components.\n\nThe primary contribution of this work is the depth-aware LoRA adaptation scheme and the accompanying empirical analysis demonstrating its advantages over uniform LoRA, full finetuning, and frozen baselines for adapting large ViTs under data scarcity. As secondary contributions, we introduce the dual-view spatial partitioning strategy for handling rectangular imagery and the progressive resolution training curriculum that stabilizes adapter optimization. The complete framework is evaluated on the CSIRO Image2Biomass benchmark [[1](https:\u002F\u002Fwww.mdpi.com\u002F1424-8220\u002F26\u002F16\u002F5285#B1-sensors-26-05285)], where it achieves a cross-validated weighted $R^{2}$𝑅 2 of 0.81.\n\nThe remainder of this paper is organized as follows. [Section 2](https:\u002F\u002Fwww.mdpi.com\u002F1424-8220\u002F26\u002F16\u002F5285#sec2-sensors-26-05285) reviews related work on biomass estimation, vision transformers, and parameter-efficient finetuning. [Section 3](https:\u002F\u002Fwww.mdpi.com\u002F1424-8220\u002F26\u002F16\u002F5285#sec3-sensors-26-05285) describes the proposed methodology. [Section 4](https:\u002F\u002Fwww.mdpi.com\u002F1424-8220\u002F26\u002F16\u002F5285#sec4-sensors-26-05285) presents the experimental setup. [Section 5](https:\u002F\u002Fwww.mdpi.com\u002F1424-8220\u002F26\u002F16\u002F5285#sec5-sensors-26-05285) reports quantitative results and ablation studies. [Section 6](https:\u002F\u002Fwww.mdpi.com\u002F1424-8220\u002F26\u002F16\u002F5285#sec6-sensors-26-05285) discusses findings and limitations. [Section 7](https:\u002F\u002Fwww.mdpi.com\u002F1424-8220\u002F26\u002F16\u002F5285#sec7-sensors-26-05285) concludes.\n\n## 2. Related Work\n\n### 2.1. Pasture Biomass Estimation\n\nTraditional approaches to pasture biomass estimation fall into three broad categories: destructive sampling, proximal sensing, and remote sensing. Destructive harvesting remains the gold standard for accuracy but is prohibitively expensive and cannot scale beyond small experimental plots [[1](https:\u002F\u002Fwww.mdpi.com\u002F1424-8220\u002F26\u002F16\u002F5285#B1-sensors-26-05285)]. Proximal sensors such as rising plate meters, capacitance probes, and active optical sensors provide faster non destructive readings, but their accuracy degrades under heterogeneous canopy structures and varying moisture conditions [[5](https:\u002F\u002Fwww.mdpi.com\u002F1424-8220\u002F26\u002F16\u002F5285#B5-sensors-26-05285)]. Satellite and UAV-based remote sensing offers spatial coverage but typically operates at coarse resolution and cannot separate biomass by species or phenological state [[6](https:\u002F\u002Fwww.mdpi.com\u002F1424-8220\u002F26\u002F16\u002F5285#B6-sensors-26-05285)].\n\nMore recently, image-based approaches using convolutional neural networks have been explored for biomass estimation from UAV-RGB imagery [[7](https:\u002F\u002Fwww.mdpi.com\u002F1424-8220\u002F26\u002F16\u002F5285#B7-sensors-26-05285),[8](https:\u002F\u002Fwww.mdpi.com\u002F1424-8220\u002F26\u002F16\u002F5285#B8-sensors-26-05285)]. These works primarily employ standard architectures such as AlexNet and ResNet on aerial orthoimages, treating biomass prediction as a single output regression task. Semi-supervised methods using synthetic data and approximate labels have also been proposed to mitigate the annotation bottleneck [[9](https:\u002F\u002Fwww.mdpi.com\u002F1424-8220\u002F26\u002F16\u002F5285#B9-sensors-26-05285)]. More recently, generative AI assisted preprocessing has been proposed as a complementary route to the same problem: Rana et al. [[10](https:\u002F\u002Fwww.mdpi.com\u002F1424-8220\u002F26\u002F16\u002F5285#B10-sensors-26-05285)] use prompt-guided generative image reconstruction to convert visually heterogeneous vegetation imagery into annotation ready representations that preserve plant morphology while reducing background complexity, facilitating the creation of higher-quality training data for downstream models. This preprocessing stage direction is complementary to the adaptation stage contribution of the present work. However, prior work has largely focused on total biomass rather than component level estimation, and few studies have leveraged modern foundation models or addressed the specific challenges of non standard image geometry in pasture imagery.\n\nDeep learning approaches to pasture and forage biomass estimation remain an active area: Cândido et al. [[11](https:\u002F\u002Fwww.mdpi.com\u002F1424-8220\u002F26\u002F16\u002F5285#B11-sensors-26-05285)] integrate proximal and satellite remote sensing with machine learning for pasture biomass estimation; Azubuike et al. [[12](https:\u002F\u002Fwww.mdpi.com\u002F1424-8220\u002F26\u002F16\u002F5285#B12-sensors-26-05285)] apply transfer learning and stacking ensembles to smartphone imagery across commercial dairy farms; and Mandal [[13](https:\u002F\u002Fwww.mdpi.com\u002F1424-8220\u002F26\u002F16\u002F5285#B13-sensors-26-05285)] recently benchmarks a systematic range of vision foundation model backbones, from EfficientNet-B3 to DINOv3 ViT-Large, together with several cross view fusion mechanisms, on the same CSIRO Image2Biomass dataset used in this work. These studies confirm that CNN- and transformer based deep learning is an active and growing direction for non-destructive biomass estimation, while also underscoring the gap our work addresses: parameter-efficient adaptation of large foundation models under extreme data scarcity (357 training images) and non standard image geometry, conditions that remain largely unaddressed in this recent literature.\n\n### 2.2. Vision Transformers and Self-Supervised Representation Learning\n\nVision Transformers (ViTs) [[14](https:\u002F\u002Fwww.mdpi.com\u002F1424-8220\u002F26\u002F16\u002F5285#B14-sensors-26-05285)] have emerged as a dominant architecture for visual representation learning, particularly when combined with self-supervised pretraining on large scale datasets. The DINO family of methods [[2](https:\u002F\u002Fwww.mdpi.com\u002F1424-8220\u002F26\u002F16\u002F5285#B2-sensors-26-05285),[15](https:\u002F\u002Fwww.mdpi.com\u002F1424-8220\u002F26\u002F16\u002F5285#B15-sensors-26-05285),[16](https:\u002F\u002Fwww.mdpi.com\u002F1424-8220\u002F26\u002F16\u002F5285#B16-sensors-26-05285)] demonstrates that self-supervised ViTs trained with self distillation objectives produce general purpose visual features that transfer effectively to diverse downstream tasks including classification, segmentation, and depth estimation, often without any finetuning. DINOv2 [[15](https:\u002F\u002Fwww.mdpi.com\u002F1424-8220\u002F26\u002F16\u002F5285#B15-sensors-26-05285)] was trained on a curated dataset of 142 million images and showed strong linear probing and k-NN performance. DINOv3 [[2](https:\u002F\u002Fwww.mdpi.com\u002F1424-8220\u002F26\u002F16\u002F5285#B2-sensors-26-05285)] scales this further to 1.7 billion images and a 7 billion parameter teacher model, introduces Gram anchoring to address dense feature degradation, and produces improved dense and global representations across over sixty benchmarks. In dense linear probing of the frozen backbone, DINOv3 reaches 55.9 mIoU on ADE20k and 86.6 mIoU on PASCAL VOC, compared to 49.5 and 83.1 for DINOv2 [[2](https:\u002F\u002Fwww.mdpi.com\u002F1424-8220\u002F26\u002F16\u002F5285#B2-sensors-26-05285)], a gap the authors attribute to Gram anchoring’s correction of patch level feature degradation that otherwise emerges under long self-supervised training schedules.\n\nThe strength of these pretrained features makes them attractive for domains with limited labeled data, such as agricultural imagery. However, directly deploying frozen features may perform poorly when the target domain differs substantially from the pretraining distribution, as is the case for close range pasture photographs that exhibit fine grained texture variation not well represented in standard image datasets. This motivates the need for controlled adaptation strategies that can bridge the domain gap without degrading the pretrained representation.\n\n### 2.3. Parameter-Efficient Fine Tuning and LoRA for Vision Transformers\n\nFull finetuning of large pretrained models on small downstream datasets is computationally expensive and prone to overfitting. This problem is particularly acute for ViTs, whose self attention layers learn globally distributed representations: perturbations to early layers can propagate through the network and degrade overall feature quality [[4](https:\u002F\u002Fwww.mdpi.com\u002F1424-8220\u002F26\u002F16\u002F5285#B4-sensors-26-05285)]. Parameter-efficient finetuning (PEFT) methods address this by introducing a small number of trainable parameters while keeping the majority of the pretrained model frozen.\n\nLoRA [[3](https:\u002F\u002Fwww.mdpi.com\u002F1424-8220\u002F26\u002F16\u002F5285#B3-sensors-26-05285)] constrains weight updates to low-rank decompositions $\\Delta W = B A$Δ 𝑊=𝐵 𝐴, where $B \\in \\mathbb{R}^{d \\times r}$𝐵∈ℝ 𝑑×𝑟, $A \\in \\mathbb{R}^{r \\times k}$𝐴∈ℝ 𝑟×𝑘, and $r \\ll min \\left(\\right. d , k \\left.\\right)$𝑟≪min(𝑑,𝑘). This reduces the parameter count by orders of magnitude while maintaining task performance and introduces no additional inference latency since the low rank updates can be merged into the original weights. Originally proposed for language models, LoRA has since been applied to ViTs for image classification [[3](https:\u002F\u002Fwww.mdpi.com\u002F1424-8220\u002F26\u002F16\u002F5285#B3-sensors-26-05285)], medical imaging [[17](https:\u002F\u002Fwww.mdpi.com\u002F1424-8220\u002F26\u002F16\u002F5285#B17-sensors-26-05285)], and other visual tasks.\n\nHowever, standard LoRA applies a uniform rank across all layers, which does not account for the well documented functional specialization of transformer layers: lower layers capture generic, transferable features while deeper layers capture increasingly task specific representations [[4](https:\u002F\u002Fwww.mdpi.com\u002F1424-8220\u002F26\u002F16\u002F5285#B4-sensors-26-05285),[18](https:\u002F\u002Fwww.mdpi.com\u002F1424-8220\u002F26\u002F16\u002F5285#B18-sensors-26-05285)]. In low data settings, uniform adaptation may allocate unnecessary capacity to early layers where the pretrained representation is already appropriate, while under adapting deeper layers where domain specific patterns must be learned. Recent work has explored layer wise learning rate schedules and selective layer unfreezing as alternatives, but these approaches provide coarse, discrete control compared to the smooth, continuous modulation possible with a rank schedule.\n\nOur work introduces a depth aware LoRA scheme that addresses this gap. By increasing the rank and scaling factor with layer depth, we achieve an allocation of adaptation capacity that is better matched to the layer wise feature hierarchy of ViTs. This approach appears especially beneficial when the training set is extremely small, where each degree of freedom in the adaptation must be carefully managed to avoid overfitting.\n\nDespite the availability of strong pretrained vision models and efficient adaptation techniques, there remains a gap in their application to agricultural biomass estimation under the specific constraints of non standard image geometry, limited data, and multi output regression. Our work addresses this gap by combining DINOv3 with dynamic depth-aware LoRA, dual view spatial partitioning, and progressive resolution training.\n\n## 3. Methodology\n\nThe overall pipeline consists of five components: (1) dual-view spatial partitioning of rectangular images, (2) feature extraction using a DINOv3 backbone with dynamic depth aware LoRA adaptation, (3) contrastive alignment of dual-view embeddings, (4) metadata integration via a dense sub network, and (5) multi output regression with Huber loss. The system is trained using a progressive resolution curriculum and evaluated through five fold stratified group cross validation. An overview of the architecture is shown in [Figure 1](https:\u002F\u002Fwww.mdpi.com\u002F1424-8220\u002F26\u002F16\u002F5285#fig_body_display_sensors-26-05285-f001).\n\n**Figure 1.** Overview of the proposed pipeline. The input image ($2000 \\times 1000$2000×1000) is split into left and right square halves, each patchified and passed through a shared DINOv3 backbone with dynamic depth aware LoRA. As shown in the inset, the LoRA rank and scaling factor follow $r_{ℓ} = 2^{1 + 8 \\left(\\right. ℓ - 1 \\left.\\right) \u002F \\left(\\right. L - 1 \\left.\\right)}$𝑟 ℓ=2 1+8(ℓ−1)\u002F(𝐿−1) and $\\alpha_{ℓ} = 2^{2 + 7 \\left(\\right. ℓ - 1 \\left.\\right) \u002F \\left(\\right. L - 1 \\left.\\right)}$𝛼 ℓ=2 2+7(ℓ−1)\u002F(𝐿−1), ranging from $r = 2 , \\alpha = 4$𝑟=2,𝛼=4 at the first layer to $r = 512 , \\alpha = 512$𝑟=512,𝛼=512 at the deepest layer. The pooled embeddings from both streams are concatenated and mapped to five biomass predictions (Grass, Clover, Dry Dead, GDM, Total).\n\n### 3.1. Dual-View Spatial Partitioning\n\nThe input images have dimensions of approximately $2000 \\times 1000$2000×1000 pixels (width × height), representing a $70 \\times 30$70×30 cm pasture quadrat captured from above. DINOv3 models are pretrained on square images with a fixed patch size (14 or 16 pixels), and directly resizing a 2:1 aspect ratio image to a square format would compress the horizontal axis and distort spatial relationships between vegetation features.\n\nInstead, each image $\\mathcal{I} \\in \\mathbb{R}^{3 \\times H \\times W}$ is partitioned along the width dimension into two equal square halves:\n\n$$\n\\mathcal{I}^{L} = \\mathcal{I} \\left[\\right. : , : , 0 : W \u002F 2 \\left]\\right. , \\mathcal{I}^{R} = \\mathcal{I} \\left[\\right. : , : , W \u002F 2 : W \\left]\\right. .\n$$\n\n(1)\n\nBoth halves are resized to the target resolution (e.g., $768 \\times 768$) and processed through the shared DINOv3 backbone in the same batch. The resulting CLS token embeddings $\\mathbf{f}^{L} , \\mathbf{f}^{R} \\in \\mathbb{R}^{d}$ are concatenated to form the combined visual representation $\\mathbf{f}_{vis} = \\left[\\right. \\mathbf{f}^{L} ; \\mathbf{f}^{R} \\left]\\right. \\in \\mathbb{R}^{2 d}$.\n\n### 3.2. Backbone and Dynamic Depth-Aware LoRA Adaptation\n\nWe use DINOv3 ViT-Large (304 M parameters, $d = 1024$, $L = 24$ layers) and DINOv3 ViT-Huge (632 M parameters, $d = 1280$, $L = 32$ layers) as backbone models. Both are initialized from publicly available self-supervised checkpoints and kept largely frozen during training.\n\n#### 3.2.1. The Case for LoRA over Full Fine-Tuning\n\nWith only 357 training images and backbone sizes exceeding 300 M parameters, full finetuning creates a severely over parameterized optimization problem. In our experiments, full finetuning of even the last four transformer blocks produced unstable training dynamics and degraded validation performance relative to a frozen baseline (CV $R^{2}$ of 0.70 vs. 0.72). This is consistent with the broader observation that self-supervised ViT representations are holistic: each layer’s output depends on the representations learned in all preceding layers, so perturbation of early layers may disrupt downstream feature representations. LoRA constrains updates to low-rank subspaces, providing plasticity for domain adaptation while limiting the risk of catastrophic forgetting of pretrained features.\n\n#### 3.2.2. Why Uniform LoRA Is Insufficient\n\nStandard LoRA applies the same rank r and scaling factor $\\alpha$ to every layer. While this is a reasonable default when data is abundant, it becomes less effective in low data settings for two reasons. First, early layers in DINOv3 encode generic visual primitives (edges, local textures, color statistics) that are already well suited to pasture imagery; adapting these layers consumes trainable capacity that could be better allo","Sensors 2026, 26(16), 5285","论文",10,false,64,{"impact":16,"substance":17,"depth":18,"authority":16,"freshness":19,"relevant":20,"comment":21},12,20,18,2,1,"论文提出动态LoRA方法用于牧场生物量估算，方法新颖且数据详实，但影响范围限于细分领域，时效性较低。",[23],{"name":10,"url":6},[25,26,27,28,29],"农业信息化","精准农业","DINOv3 LoRA","牧场生物量","Vision Transformer Sensors",[],{"status":32,"locked":13,"version":33,"method":34,"evidence":35},"needs_review","2026-10-22.2","rules",{"candidates":36,"agricultural_evidence":13,"reason":42},[37],{"code":38,"score":19,"supported":13,"title":39,"abstract":40,"tags":41},"I1",[],[],[25,26],"农业相关性需要复核",[44,45],"农业信息化 牧场生物量 精准农业 Vision Transformer Sensors","农业信息化 牧场生物量","农业信息化牧场生物量精准农业VisionTransformerSensors-909","2026-08-24T00:06:13.245618Z",{"total":49,"page":20,"page_size":49,"items":50},6,[51,114,237,289,399,451],{"id":52,"title":53,"url":54,"summary":55,"summary_zh":8,"content":8,"source_name":56,"source_url":54,"published_at":57,"category":11,"cover_url":8,"hotness":12,"is_selected":13,"score":58,"score_detail":59,"sources":62,"tags":64,"primary_topic":68,"secondary_topics":73,"classification":74,"search_phrases":90,"slug":93,"view_count":94,"doi":95,"paper":96,"created_at":113},5579,"A Composite Field-Evaluation Framework for Digital Agriculture Maps","https:\u002F\u002Fdoi.org\u002F10.21203\u002Frs.3.rs-11283264\u002Fv1","A Composite Field-Evaluation Framework for Digital Agriculture Maps。Research Square","Research Square","2026-10-21T00:00:00Z",38,{"impact":60,"substance":49,"depth":60,"authority":49,"freshness":12,"relevant":20,"comment":61},8,"预印本提出数字农业地图的田间复合评估框架，方法有一定新意，但摘要信息极少、未经同行评审，公共影响有限，暂不宜进入每日精选。",[63],{"name":56,"url":54},[65,25,66,67],"数字农业","遥感","地图评估",{"code":38,"slug":69,"name":70,"path":71,"parent_code":72},"smart-agriculture","智慧农业总论","\u002Ftopic\u002Fdigital-agriculture\u002Fsmart-agriculture","I",[],{"status":75,"locked":13,"version":33,"method":34,"evidence":76},"classified",{"candidates":77,"agricultural_evidence":80},[78,85],{"code":38,"score":79,"supported":80,"title":81,"abstract":83,"tags":84},13,true,[82],"digital agriculture",[82],[25,65],{"code":86,"score":20,"supported":13,"title":87,"abstract":88,"tags":89},"I3",[],[],[66],[91,92],"Research Square 数字农业","农业信息化 地图评估 数字农业 遥感","ResearchSquare数字农业-5579",0,"10.21203\u002Frs.3.rs-11283264\u002Fv1",{"doi":95,"openalex_id":97,"authors":98,"venue":56,"cited_by_count":94,"oa_url":54,"card":8,"direction":111,"ingested_from":112},"W7223099987",[99,102,105,108],{"name":100,"orcid":101},"Phillip Lanza","https:\u002F\u002Forcid.org\u002F0000-0002-5545-1744",{"name":103,"orcid":104},"David G. Rossiter","https:\u002F\u002Forcid.org\u002F0000-0003-4558-1286",{"name":106,"orcid":107},"Diana Stuart Sinton","https:\u002F\u002Forcid.org\u002F0000-0001-5828-9001",{"name":109,"orcid":110},"Louis Longchamps","https:\u002F\u002Forcid.org\u002F0000-0002-4761-6094","数字乡村与农业信息化","openalex","2026-10-22T11:20:47.463814Z",{"id":115,"title":116,"url":117,"summary":118,"summary_zh":119,"content":8,"source_name":120,"source_url":117,"published_at":57,"category":11,"cover_url":8,"hotness":12,"is_selected":13,"score":121,"score_detail":122,"sources":126,"tags":128,"primary_topic":133,"secondary_topics":138,"classification":150,"search_phrases":211,"slug":214,"view_count":94,"doi":215,"paper":216,"created_at":236},5553,"Artificial Intelligence–Driven Precision Agriculture Systems: Engineering Frameworks for Climate-Resilient Food Production","https:\u002F\u002Fdoi.org\u002F10.53982\u002Fajerd.2026.01.14-j","Climate change, soil issues, water scarcity, and the increasing need for food worldwide mean we need smart farming systems that can manage resources flexibly. Old farming methods used the same inputs and simple rules for decisions. AI can help us manage resources better by adapting and optimizing them as we deal with complex environmental factors. This paper looks closely at AI-powered precision agriculture systems designed to make food production more resilient to changing weather. The authors reviewed research from 2021 to 2026 in agricultural engineering, computer engineering, and environmental systems. The authors examined how machine learning, deep learning, reinforcement learning, computer vision, robotics, and IoT sensors are used for predicting crop yields, spotting diseases, improving irrigation, managing nutrients, and automating field tasks. Our analysis shows that hybrid deep learning models (CNN-LSTM) are better than single models for predicting crop yields over space and time. Reinforcement learning algorithms are also better at adjusting irrigation cycles when weather is unpredictable. Using edge computing for system design reduces delays and improves real-time decision-making compared to traditional cloud-based systems. However, there are still big challenges with understanding how AI models work, having enough diverse data, ensuring cybersecurity, setting common standards, and making these systems affordable, especially in developing countries. To tackle these problems, we suggest a framework that combines AI analysis, distributed sensor networks, adaptive controls, and climate risk assessments. The authors provided critical insights into essential research gaps and outline a technological roadmap towards achieving sustainable, resilient and intelligent agricultural systems.","气候变化、土壤问题、水资源短缺以及全球日益增长的粮食需求，意味着我们需要能够灵活管理资源的智能农业系统。传统农业方法对相同的投入采用固定模式，并依赖简单规则进行决策。人工智能可以帮助我们更好地管理资源，通过适应和优化资源利用来应对复杂的环境因素。本文深入探讨了人工智能驱动的精准农业系统，旨在使粮食生产对不断变化的天气更具韧性。作者综述了2021年至2026年间农业工程、计算机工程和环境系统领域的研究。作者考察了机器学习、深度学习、强化学习、计算机视觉、机器人技术及物联网传感器如何被用于作物产量预测、病害识别、灌溉优化、养分管理和田间作业自动化。我们的分析表明，混合深度学习模型（CNN-LSTM）在时空作物产量预测方面优于单一模型。在天气不可预测的情况下，强化学习算法在调整灌溉周期方面也表现更优。与传统基于云的系统相比，采用边缘计算进行系统设计可降低延迟并改善实时决策能力。然而，在理解人工智能模型的工作原理、获取足够多样化的数据、确保网络安全、制定通用标准以及使这些系统具备可负担性方面，仍存在重大挑战，尤其是在发展中国家。为应对这些问题，我们提出了一个融合人工智能分析、分布式传感器网络、自适应控制和气候风险评估的框架。作者对关键研究空白提供了重要见解，并勾勒了实现可持续、韧性和智能农业系统的技术路线图。","ABUAD Journal of Engineering Research and Development (AJERD)",80,{"impact":18,"substance":123,"depth":18,"authority":79,"freshness":124,"relevant":20,"comment":125},22,9,"系统综述AI驱动精准农业的工程框架与技术路线，方法梳理扎实、结论有参考价值，但属综述类论文，产业落地影响有限。",[127],{"name":120,"url":117},[129,130,131,26,132],"智慧农业","农业人工智能","物联网","气候韧性",{"code":134,"slug":135,"name":136,"path":137,"parent_code":72},"I2","agricultural-ai","农业AI与大模型","\u002Ftopic\u002Fdigital-agriculture\u002Fagricultural-ai",[139,144],{"code":140,"slug":141,"name":142,"path":143,"parent_code":72},"I5","agricultural-iot","物联网传感与边缘计算","\u002Ftopic\u002Fdigital-agriculture\u002Fagricultural-iot",{"code":145,"slug":146,"name":147,"path":148,"parent_code":149},"H3","irrigation","水资源与灌溉","\u002Ftopic\u002Fenvironment\u002Firrigation","H",{"status":75,"locked":13,"version":33,"method":34,"evidence":151},{"candidates":152,"agricultural_evidence":80},[153,164,170,177,183,191,198,205],{"code":134,"score":18,"supported":80,"title":154,"abstract":156,"tags":163},[155],"artificial intelligence",[157,158,159,160,161,162],"深度学习","机器学习","计算机视觉","machine learning","deep learning","computer vision",[130],{"code":38,"score":165,"supported":80,"title":166,"abstract":168,"tags":169},16,[167],"precision agriculture",[26,167],[129,26],{"code":140,"score":12,"supported":80,"title":171,"abstract":172,"tags":176},[],[131,173,174,175],"边缘计算","IoT","edge computing",[131],{"code":145,"score":124,"supported":80,"title":178,"abstract":179,"tags":182},[],[180,181,146],"灌溉","水资源",[],{"code":184,"score":185,"supported":80,"title":186,"abstract":187,"tags":190},"H7",7,[],[188,189],"气候变化","climate change",[132],{"code":192,"score":49,"supported":80,"title":193,"abstract":194,"tags":197},"H2",[],[195,196],"土壤","soil",[],{"code":199,"score":200,"supported":80,"title":201,"abstract":202,"tags":204},"A7",3,[],[203],"间作",[],{"code":206,"score":200,"supported":80,"title":207,"abstract":208,"tags":210},"F3",[],[209],"养分管理",[],[212,213],"AI 精准农业 气候韧性","CNN-LSTM 作物产量预测","AI精准农业气候韧性-5553","10.53982\u002Fajerd.2026.01.14-j",{"doi":215,"openalex_id":217,"authors":218,"venue":120,"cited_by_count":94,"oa_url":117,"card":229,"direction":235,"ingested_from":112},"W7223121456",[219,222,224,226],{"name":220,"orcid":221},"Olumayowa J. Iyanda","https:\u002F\u002Forcid.org\u002F0000-0002-5340-9884",{"name":223,"orcid":8},"Oluwadare OYEBODE",{"name":225,"orcid":8},"Iretiayo Adelaiye",{"name":227,"orcid":228},"Ayomide Afolabi","https:\u002F\u002Forcid.org\u002F0000-0002-5171-886X",{"tldr":230,"method":231,"finding":232,"direction":233,"opportunity":234},"综述2021-2026年AI驱动精准农业系统，提出气候韧性粮食生产工程框架。","综述机器学习、深度学习、强化学习、计算机视觉、机器人与物联网在精准农业中的应用。","CNN-LSTM混合模型和强化学习在产量预测与灌溉调度上优于单一方法，边缘计算提升实时性。","农业人工智能与决策模型","可研究边缘计算与强化学习结合的轻量级灌溉决策系统，并探索发展中国家低成本部署与标准化路径。","智慧农业 \u002F 农业物联网","2026-10-22T11:20:18.151376Z",{"id":238,"title":239,"url":240,"summary":241,"summary_zh":8,"content":8,"source_name":242,"source_url":240,"published_at":243,"category":11,"cover_url":8,"hotness":12,"is_selected":13,"score":244,"score_detail":245,"sources":249,"tags":251,"primary_topic":254,"secondary_topics":255,"classification":256,"search_phrases":268,"slug":271,"view_count":94,"doi":272,"paper":273,"created_at":288},5539,"Robust and transparent crop-row detection from RGB field imagery for vision-based agricultural guidance systems","https:\u002F\u002Fdoi.org\u002F10.1007\u002Fs11119-026-10473-z","Robust and transparent crop-row detection from RGB field imagery for vision-based agricultural guidance systems。Precision Agriculture","Precision Agriculture","2026-10-22T00:00:00Z",79,{"impact":165,"substance":246,"depth":18,"authority":247,"freshness":12,"relevant":20,"comment":248},21,14,"核心期刊新论文，提出鲁棒且可解释的RGB作物行检测方法，对视觉导航农机具参考价值明确，但属细分技术进展，未达重大突破层级。",[250],{"name":242,"url":240},[129,130,26,252,253],"机器视觉","作物行检测",{"code":38,"slug":69,"name":70,"path":71,"parent_code":72},[],{"status":75,"locked":13,"version":33,"method":34,"evidence":257},{"candidates":258,"agricultural_evidence":80},[259,264],{"code":38,"score":260,"supported":80,"title":261,"abstract":262,"tags":263},5,[],[167],[129,26],{"code":134,"score":19,"supported":13,"title":265,"abstract":266,"tags":267},[],[],[130,252],[269,270],"RGB 作物行检测","农业人工智能 作物行检测 智慧农业 机器视觉","RGB作物行检测-5539","10.1007\u002Fs11119-026-10473-z",{"doi":272,"openalex_id":274,"authors":275,"venue":242,"cited_by_count":94,"oa_url":8,"card":8,"direction":8,"ingested_from":112},"W7223117828",[276,279,282,285],{"name":277,"orcid":278},"Esther Gascó","https:\u002F\u002Forcid.org\u002F0000-0003-1798-4483",{"name":280,"orcid":281},"Clara I. López-González","https:\u002F\u002Forcid.org\u002F0000-0002-2755-1692",{"name":283,"orcid":284},"Eva Besada-Portas","https:\u002F\u002Forcid.org\u002F0000-0001-6129-4653",{"name":286,"orcid":287},"Gonzalo Pájares","https:\u002F\u002Forcid.org\u002F0000-0003-0915-6282","2026-10-22T11:20:09.678587Z",{"id":290,"title":291,"url":292,"summary":293,"summary_zh":294,"content":8,"source_name":295,"source_url":292,"published_at":296,"category":11,"cover_url":8,"hotness":12,"is_selected":13,"score":297,"score_detail":298,"sources":300,"tags":302,"primary_topic":307,"secondary_topics":313,"classification":314,"search_phrases":361,"slug":364,"view_count":94,"doi":365,"paper":366,"created_at":398},5480,"Sustainable Plant Disease Management: Current Advances, Challenges, and Future Directions","https:\u002F\u002Fdoi.org\u002F10.3390\u002Fsu182010661","Sustainable plant disease management is essential for maintaining agricultural productivity, environmental health, and global food security. Climate change has increased crop vulnerability to emerging and re-emerging plant diseases, posing significant challenges for agricultural sustainability. Although synthetic chemical pesticides have long been effective for disease control, their extensive use is increasingly questioned because of environmental contamination, pathogen resistance, and potential risks to human and ecosystem health. Consequently, plant disease management (PDM) is shifting toward more sustainable and integrated approaches. Future strategies are expected to incorporate multiple disciplines, including: (i) biological control and microbiome engineering to enhance natural plant defense mechanisms; (ii) advanced molecular and genomic technologies, such as CRISPR\u002FCas systems, RNA interference (RNAi), and epigenome editing, to develop durable disease resistance; (iii) smart delivery systems for the precise application of antimicrobial compounds and defense elicitors; (iv) nanotechnology-based platforms for the targeted delivery of nutrients, pesticides, and biocontrol agents, improving efficacy while reducing chemical inputs; (v) digital and precision agriculture technologies, including remote sensing, unmanned aerial vehicles (UAVs), and artificial intelligence (AI)-based diagnostics, to facilitate real-time disease surveillance and targeted interventions; and (vi) agroecological practices such as crop diversification, soil health restoration, and climate-smart farming to enhance agroecosystem resilience. These innovations aim not only to suppress plant pathogens but also to build sustainable, adaptive, and resilient agricultural systems that address future environmental and biological challenges.","可持续的植物病害管理对于维持农业生产力、环境健康和全球粮食安全至关重要。气候变化加剧了作物对新发和再发植物病害的脆弱性，给农业可持续性带来了重大挑战。尽管合成化学农药长期以来在病害防控中行之有效，但其广泛使用正日益受到质疑，原因包括环境污染、病原菌抗药性以及对人类和生态系统健康的潜在风险。因此，植物病害管理（plant disease management, PDM）正转向更加可持续和综合的途径。未来的策略预计将融合多个学科，包括：(i) 生物防治和微生物组工程，以增强植物天然防御机制；(ii) 先进的分子和基因组技术，如CRISPR\u002FCas系统、RNA干扰（RNAi）和表观基因组编辑，以培育持久抗病性；(iii) 用于精准施用抗菌化合物和防御诱导子的智能递送系统；(iv) 基于纳米技术的平台，用于靶向递送养分、农药和生物防治剂，在提高功效的同时减少化学投入；(v) 数字和精准农业技术，包括遥感、无人机（UAVs）和基于人工智能（AI）的诊断，以促进实时病害监测和靶向干预；以及(vi) 农业生态实践，如作物多样化、土壤健康恢复和气候智慧型农业，以增强农业生态系统的韧性。这些创新不仅旨在抑制植物病原菌，还旨在构建可持续、适应性强且具有韧性的农业系统，以应对未来的环境和生物挑战。","Sustainability","2026-10-20T00:00:00Z",81,{"impact":123,"substance":17,"depth":18,"authority":79,"freshness":60,"relevant":20,"comment":299},"系统综述可持续植物病害管理六大技术方向，涵盖AI诊断、纳米递送与基因编辑，对智慧农业与绿色防控具参考价值。",[301],{"name":295,"url":292},[129,26,303,304,305,306],"基因编辑","植物病害","生物防治","纳米技术",{"code":308,"slug":309,"name":310,"path":311,"parent_code":312},"F4","crop-protection","农药与绿色防控","\u002Ftopic\u002Finputs-equipment\u002Fcrop-protection","F",[],{"status":75,"locked":13,"version":33,"method":34,"evidence":315},{"candidates":316,"agricultural_evidence":80},[317,321,326,333,337,341,346,352,356],{"code":38,"score":60,"supported":80,"title":318,"abstract":319,"tags":320},[],[26,167],[129,26],{"code":308,"score":185,"supported":80,"title":322,"abstract":323,"tags":325},[],[324,305],"农药",[305],{"code":327,"score":49,"supported":80,"title":328,"abstract":329,"tags":332},"A8",[],[330,331],"粮食安全","food security",[],{"code":192,"score":49,"supported":80,"title":334,"abstract":335,"tags":336},[],[195,196],[],{"code":184,"score":49,"supported":80,"title":338,"abstract":339,"tags":340},[],[188,189],[],{"code":86,"score":49,"supported":80,"title":342,"abstract":343,"tags":345},[],[66,344],"remote sensing",[],{"code":347,"score":200,"supported":80,"title":348,"abstract":349,"tags":351},"H6",[],[350],"农业生态",[],{"code":134,"score":200,"supported":80,"title":353,"abstract":354,"tags":355},[],[155],[],{"code":357,"score":20,"supported":13,"title":358,"abstract":359,"tags":360},"F2",[],[],[303],[362,363],"CRISPR RNAi 作物抗病","基因编辑 智慧农业 植物病害 生物防治","CRISPRRNAi作物抗病-5480","10.3390\u002Fsu182010661",{"doi":365,"openalex_id":367,"authors":368,"venue":295,"cited_by_count":94,"oa_url":292,"card":393,"direction":235,"ingested_from":112},"W7223044776",[369,372,375,378,381,384,387,390],{"name":370,"orcid":371},"Karim M. Hassan","https:\u002F\u002Forcid.org\u002F0000-0002-1241-2307",{"name":373,"orcid":374},"Basma T. Abd-Elhalim","https:\u002F\u002Forcid.org\u002F0000-0002-3009-332X",{"name":376,"orcid":377},"Mohamed K. Abou El-Nasr","https:\u002F\u002Forcid.org\u002F0000-0003-1451-833X",{"name":379,"orcid":380},"Essam Yosef Abdul-Hafeez","https:\u002F\u002Forcid.org\u002F0000-0003-4744-4672",{"name":382,"orcid":383},"Ahmed N. Abdelhamid","https:\u002F\u002Forcid.org\u002F0000-0002-1326-7885",{"name":385,"orcid":386},"Mostafa Abdelkader","https:\u002F\u002Forcid.org\u002F0000-0003-2339-5087",{"name":388,"orcid":389},"Mohamed Abd El-Hamed Nasser","https:\u002F\u002Forcid.org\u002F0000-0002-0328-2969",{"name":391,"orcid":392},"Mahmoud A. A. Ali","https:\u002F\u002Forcid.org\u002F0000-0002-9062-0048",{"tldr":394,"method":395,"finding":396,"direction":235,"opportunity":397},"综述植物病害可持续管理的多学科前沿进展、挑战与未来方向。","文献综述，整合生物防治、基因编辑、纳米递送、数字农业与生态农艺。","未来病害管理需融合生物、分子、纳米与数字技术构建韧性农业系统。","遥感、无人机与AI诊断的实时病害监测闭环仍缺田间验证与低成本落地研究。","2026-10-21T23:30:09.242484Z",{"id":400,"title":401,"url":402,"summary":403,"summary_zh":8,"content":8,"source_name":404,"source_url":8,"published_at":405,"category":11,"cover_url":8,"hotness":12,"is_selected":13,"score":406,"score_detail":407,"sources":410,"tags":412,"primary_topic":415,"secondary_topics":418,"classification":419,"search_phrases":438,"slug":441,"view_count":94,"doi":8,"paper":442,"created_at":450},5462,"Architecture diverse TabNet ensemble for crop type classification using harmonized multisource soil data（基于多源土壤数据融合的架构多样性TabNet集成作物类型分类）","https:\u002F\u002Fwww.nature.com\u002Farticles\u002Fs41598-026-75366-7","对18,237条多源土壤记录、7类作物构建TabNet集成框架。TPE优化使对数损失由0.3325降至0.2566，集成后准确率0.9053、宏F1 0.9025，优于XGBoost的0.8724。作者称尚未做外部验证。","Scientific Reports 2026-10-18","2026-10-18T00:00:00Z",70,{"impact":16,"substance":17,"depth":408,"authority":79,"freshness":60,"relevant":20,"comment":409},17,"方法新颖、数据规模可观且指标有实质提升，但缺乏外部验证、属细分方法研究，公共影响有限，可作候选而非必选。",[411],{"name":404,"url":402},[130,158,26,413,414],"土壤数据","作物分类",{"code":192,"slug":196,"name":416,"path":417,"parent_code":149},"土壤健康与改良","\u002Ftopic\u002Fenvironment\u002Fsoil",[],{"status":75,"locked":13,"version":33,"method":34,"evidence":420},{"candidates":421,"agricultural_evidence":80},[422,426,430,434],{"code":192,"score":17,"supported":80,"title":423,"abstract":424,"tags":425},[195,196],[195],[413],{"code":134,"score":19,"supported":13,"title":427,"abstract":428,"tags":429},[],[],[130,158],{"code":38,"score":20,"supported":13,"title":431,"abstract":432,"tags":433},[],[],[26],{"code":86,"score":20,"supported":13,"title":435,"abstract":436,"tags":437},[],[],[414],[439,440],"TabNet 作物类型分类","多源土壤数据 作物分类","TabNet作物类型分类-5462",{"doi":8,"openalex_id":8,"authors":443,"venue":8,"cited_by_count":94,"oa_url":8,"card":444,"direction":233,"ingested_from":449},[],{"tldr":445,"method":446,"finding":447,"direction":233,"opportunity":448},"构建架构多样性TabNet集成，用多源土壤数据分类7类作物。","18,237条多源土壤记录，TPE调参的TabNet集成，对比XGBoost。","集成后准确率0.9053、宏F1 0.9025，优于XGBoost的0.8724。","尚未外部验证，可跨区域迁移验证并融合遥感时序特征提升泛化。","agent","2026-10-21T00:10:08.428584Z",{"id":452,"title":453,"url":454,"summary":455,"summary_zh":8,"content":8,"source_name":456,"source_url":454,"published_at":457,"category":11,"cover_url":8,"hotness":458,"is_selected":13,"score":459,"score_detail":460,"sources":462,"tags":466,"primary_topic":468,"secondary_topics":469,"classification":470,"search_phrases":477,"slug":480,"view_count":94,"doi":481,"paper":482,"created_at":494},5396,"Understanding the Digital Agriculture Landscape in Central Africa: 2000-2026 Systematic Review.","https:\u002F\u002Fdoi.org\u002F10.13140\u002Frg.2.2.22601.71520","Understanding the Digital Agriculture Landscape in Central Africa: 2000-2026 Systematic Review.。OpenAlex","OpenAlex","2026-10-19T00:00:00Z",25,37,{"impact":60,"substance":49,"depth":60,"authority":49,"freshness":124,"relevant":20,"comment":461},"中非数字农业系统综述，主题相关但摘要信息极少，仅题录级元数据，增量与权威性有限，不建议进入每日精选。",[463,464],{"name":456,"url":454},{"name":456,"url":465},"https:\u002F\u002Fdoi.org\u002F10.13140\u002Frg.2.2.16146.41922",[65,129,25,467],"中非",{"code":38,"slug":69,"name":70,"path":71,"parent_code":72},[],{"status":75,"locked":13,"version":33,"method":34,"evidence":471},{"candidates":472,"agricultural_evidence":80},[473],{"code":38,"score":247,"supported":80,"title":474,"abstract":475,"tags":476},[82],[82],[25,65,129],[478,479],"中非 数字农业","农业信息化 数字农业 智慧农业 中非","中非数字农业-5396","10.13140\u002Frg.2.2.22601.71520",{"doi":481,"openalex_id":483,"authors":484,"venue":8,"cited_by_count":94,"oa_url":454,"card":8,"direction":111,"ingested_from":112},"W7222930115",[485,487,489,492],{"name":486,"orcid":8},"Anani Ogou",{"name":488,"orcid":8},"Henry Chotangui",{"name":490,"orcid":491},"Honoré Beyegue-Djonko","https:\u002F\u002Forcid.org\u002F0009-0004-6447-5889",{"name":493,"orcid":8},"Christopher Mubeteneh Tankou","2026-10-20T23:30:22.542913Z"]