[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"item-113":3,"related-113":69},{"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":24,"tags":26,"primary_topic":31,"secondary_topics":37,"classification":38,"search_phrases":62,"slug":65,"view_count":66,"doi":67,"paper":8,"created_at":68},113,"A deep learning perception framework for farm digital twins: weed species classification, semantic segmentation, and gradient-based visual interpretability","https:\u002F\u002Fwww.frontiersin.org\u002Fjournals\u002Fsustainable-food-systems\u002Farticles\u002F10.3389\u002Ffsufs.2026.1841305\u002Ffull","Anusree Manoj, Aiswarya S. Kumar, S. Remya, Somula Ramasubbareddy, Yong Yun Cho提出一种面向农业数字孪生系统的深度学习感知框架,涵盖杂草物种分类、语义分割和基于梯度的视觉可解释性分析。",null,"## Abstract\n\nThe emergence of Farm Digital Twins (Farm-DT) as a transformative paradigm in smart agriculture demands robust, real-time perception modules capable of continuous plant-level monitoring, predictive analytics, and automated decision support. A critical bottleneck in operationalising Farm-DTs is the absence of interpretable, species-level weed identification engines that can feed spatially precise weed-distribution data into the virtual farm replica for simulation, yield forecasting, and optimized herbicide scheduling. This paper presents a tri-component deep learning framework designed explicitly as a perception and interpretability layer for Farm Digital Twin architectures, evaluated on a five-species balanced subset of the Moving Fields Weed Dataset (MFWD)—a publicly available benchmark of 94K high-resolution images of 28 weed species. Eleven ImageNet pre-trained architectures spanning convolutional neural networks (CNNs) and vision transformers were benchmarked under a unified stratified 80\u002F10\u002F10 holdout protocol. Swin Transformer v1 achieved the highest test accuracy of 97.3% [_F_ 1 = 0.964; 95% CI: (95.9, 98.4)], and EfficientNetV2-S reached 95.5% [_F_ 1 =0.956; 95% CI: (93.8, 96.9)]; pairwise McNemar tests confirm these advantages are statistically significant (_p_\u003C 0.01, Bonferroni-corrected). Gradient-weighted Class Activation Mapping (Grad-CAM) applied to the top CNN models confirmed that 94.3% of high-activation pixels overlap with ground-truth foliage annotations, providing quantitative validation that classification decisions are driven by botanically meaningful morphological features. The SegFormer-B3 semantic segmentation module achieved mIoU = 0.8961, enabling precise pixel-level weed delineation that directly populates the DT spatial model for variable-rate herbicide application and robotic weeding simulation. Together, these three components map directly onto the Farm-DT architecture: the classifier feeds the species inventory, the segmentation module populates the spatial weed model, and Grad-CAM provides the trust layer required for agronomist acceptance of DT-automated decisions—collectively advancing real-time monitoring and decision-support for Digital Twin-driven urban and peri-urban agriculture.\n\n## 1 Introduction\n\nThe convergence of accelerated urbanization, climate change, and food security pressures has elevated urban agriculture (UA) from a marginal practice to a cornerstone of sustainable urban development. Within this landscape, Digital Twin (DT) technology has emerged as a transformative enabler: by creating a dynamic, continuously updated virtual replica of a physical farm system, DTs allow stakeholders to simulate, forecast, and optimize the full food production lifecycle within a risk-free environment. A fundamental prerequisite for any functional Farm Digital Twin is a high-fidelity, real-time perception layer that can translate physical field conditions—plant health, species composition, pest pressure—into structured data streams feeding the virtual model (; ).\n\nWeed management represents one of the most pressing challenges in this context. Weeds compete directly with crops for light, water, and nutrients, and species-level accurate weed identification is the essential first step toward effective integrated pest management within a DT framework (). Precise species identification permits the DT to simulate the consequences of specific herbicide interventions, predict resistance development (), and schedule variable-rate applications—all without physical trial and error. Yet current approaches suffer from two critical limitations that prevent their integration into operational Farm-DT architectures: (1) most models lack the interpretability required for agronomist trust and regulatory acceptance; and (2) the majority focus on classification alone, without providing the pixel-level spatial data that the DT spatial model requires for site-specific robotic intervention ().\n\nThe emergence of deep learning, particularly the use of CNNs and, to some extent, transformer inspired architectures, has significantly changed automated plant phenotyping and pest detection. These capabilities make it possible to perform feature extraction hierarchically directly from the high-dimensional image data (; ). CNNs excel at capturing fine-grained spatial structures and complex plant patterns through locally receptive fields. On the other hand, vision transformers apply self-attention to understand long-range spatial dependencies and global contextual relationships, thereby surpassing purely convolutional representations in capturing global context (). Yet, even with those changes, two issues continue to limit large-scale adoption: (1) models typically lack sufficient interpretability for agronomists to verify that classification decisions are grounded in actual plant morphology rather than data artifacts; and (2) Most of the methods focus on either classification or segmentation and do not provide the level of pixel-wise spatial detail required by robotic weeding devices and variable-rate sprayer systems.\n\nTo address these challenges, we develop a tri-component deep learning framework with three distinct objectives. The first goal is to develop a multi-class species classifier capable of distinguishing morphologically similar weed species under varied acquisition conditions The second goal is to use explainable artificial intelligence (XAI) techniques to validate model decisions and check whether the features that the model has learned correspond to the botanical features of the plants. The third goal is to accurately generate pixel-level weed delineations by semantic segmentation, which in turn allows for site-specific robotic intervention. To achieve these goals, we systematically evaluate eleven ImageNet pre-trained architectures: GoogLeNet, DenseNet-121, AlexNet, ConvNeXt-Tiny, ShuffleNetV2, SqueezeNet, Inceptionv3, ResNet-50, EfficientNetV2-S, MobileNetV2, and Swin Transformer v1. We chose these models to cover a wide variety of architectural paradigms, such as residual networks, lightweight mobile-optimized networks, attention-based transformers, and inception modules with multi-scale feature extraction designs.\n\nExperiments were conducted using the publicly available Moving Fields Weed Dataset (MFWD) (), which consists of 94, 321 high-resolution RGB images covering 28 weed species commonly appearing in German maize and sorghum environments. These images were captured in a fully automated, high-throughput indoor phenotyping facility with controlled, uniform illumination. A stratified sampling process was carried out leading to a five-species balanced subset (PULDY, match, cheal, galap, SOLNI) with a total of 7, 954 images to reduce classification bias and ensure fair class representation. Class-discriminative heatmaps were created through Gradient-weighted Class Activation Mapping (Grad-CAM) () applied to the top-performing CNNs (EfficientNetV2-S, Inception v3), and quantitative analysis confirmed that 94.3% of high-activation pixels correspond to ground-truth foliage regions. The SegFormer-B3 model (), a transformer-based encoder–decoder architecture that paradoxically melds hierarchically multi-scale feature extraction with a very simple mlp decoder, was trained for semantic segmentation on 7,954 semi-automatically annotated binary masks, initially generated through hsv color-space thresholding and subsequently verified by domain experts. A mean Intersection over Union (mIoU) of 0.8961 was achieved.\n\nThe main contributions of this work are as follows:\n\n*   **Systematic benchmark with statistical validation**. Eleven pre-trained architectures spanning CNNs and vision transformers are evaluated under a unified 80\u002F10\u002F10 stratified holdout protocol with McNemar pairwise significance tests (Bonferroni-corrected), paired _t_-tests across five random seeds, and 95% confidence intervals. Swin Transformer v1 achieved the highest test accuracy of 97.3% and EfficientNetV2-S achieved 95.5%, confirmed as statistically superior to all CNN baselines (_p_\u003C 0.01).\n\n*   **Quantitative XAI validation**. Grad-CAM () applied to EfficientNetV2-S and Inception v3 confirms that 94.3% of high-activation pixels overlap with ground-truth foliage annotations, providing evidence that classification decisions are driven by botanical morphological features rather than background artifacts or spurious correlations.\n\n*   **Integrated segmentation pipeline**. A semi-automated HSV-thresholding and expert-review annotation pipeline produced 7,954 pixel-wise binary masks. SegFormer-B3 () trained on these masks obtains mIoU = 0.8961 (precision = 0.8139; recall = 0.8204) which allows reliable weed-patch delineation for robotic weeding and targeted herbicide application.\n\n*   **Ablation study and efficiency analysis**. Three controlled ablation experiments have been used to measure the independent effects of ImageNet transfer learning (+15%–17% accuracy), data augmentation technique (+4.2% for Swin v1), and segmentation loss mixture (+0.035 mIoU), together with a study of accuracy-efficiency balance to help decide on models that would work within limited resource environments.\n\nIn this work, the classification module serves as the DT species inventory feed, the segmentation module populates the DT spatial model with pixel-level weed maps, and the Grad-CAM module provides the interpretability layer required for agronomist trust in DT-automated decisions.\n\nThe remaining sections are organized as follows. In Section 2, we summarize the state-of-the-art deep learning-based methods for weed detection and semantic segmentation. Sections 3, 4 provide a detailed methodology that includes dataset preparation, model architectures, training configurations, and evaluation protocols. Section 5 compiles a full set of experiments with cross-comparison of different methods, statistical significance, and ablation analysis. Section 6 discusses the limitations of the proposed framework. Section 7 summarizes the findings and outlines future research directions for interpretable precision agriculture systems.\n\n## 2 Literature review\n\nCombining computer vision with deep learning for agricultural weed management is rapidly becoming a major topic that is not only enhancing sustainable farming practices but also limiting the inappropriate use of herbicides. This section surveys image-based weed classification, transformer architectures, semantic segmentation frameworks, and explainable AI techniques, followed by a structured gap analysis ([Table 1](https:\u002F\u002Fwww.frontiersin.org\u002Fjournals\u002Fsustainable-food-systems\u002Farticles\u002F10.3389\u002Ffsufs.2026.1841305\u002Ffull#T1)).\n\nTable 1\n\n| Study | Dataset | Cls. | Best Acc. | C + T | CV | Aug | XAI(Q) | Seg | Stat | Abl | CI | DT-Ready |\n| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |\n|  | Sugarcane | 7 | 96.6% | × | × | ✓ | × | × | × | × | × | × |\n|  | Mixed crop | 2 | 87.2% | × | × | ✓ | × | × | × | × | × | × |\n|  | Custom weed | 10 | 98.6% | ° | ✓ | ✓ | ° | × | × | × | × | ° |\n|  | MFWD-UAV | 2 | 89% _F_ 1 | × | × | ✓ | × | ✓ | × | × | × | × |\n|  | Sorghum | 5 | 99.1% | × | ✓ | ✓ | ° | × | × | ° | × | × |\n|  | Grape leaf | 4 | 99.4% | × | × | ✓ | ° | × | × | × | × | ° |\n|  | Crop\u002Fweed | 6 | 96.1% | ° | × | ✓ | × | × | × | × | × | × |\n|  | Custom | 3 | 94.5% | × | × | ✓ | × | × | × | × | × | × |\n|  | Potato | 2 | 92.4% | × | × | ✓ | × | ✓ | × | × | × | × |\n|  | Straw.\u002Fpea | 2 | 96.8% | × | × | ✓ | × | × | × | × | × | × |\n| **Proposed** | **MFWD** ( | **5+seg** | **97.3%** | ✓ | ° | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ |\n\nStructured comparison of representative weed detection studies across nine evaluation dimensions.\n\nC + T, both CNN and Transformer families evaluated; CV, cross-validation; Aug, data augmentation; XAI(Q), quantitative XAI evaluation; Seg, semantic segmentation component; Stat, statistical significance tests reported; Abl, ablation study included; CI, confidence intervals reported. ✓ = present; × = absent; ° = partial. Bold values denote results for the proposed method, shown for comparison against the reviewed literature.\n\n### 2.1 CNN-based weed classification\n\nCNNs have proven to be more effective than traditional ML techniques by utilizing hand-crafted features to identify weeds (; ). To identify sugarcane weeds, developed a deep learning framework using 5,660 images. Six architectures—DarkNet-53, AlexNet, GoogLeNet, Inception v3, ResNet-50, and Xception—were evaluated across two batch sizes and three training durations at a fixed learning rate of 0.001. DarkNet-53 achieved the best performance with an _F_ 1-score of 0.99 and 96.6% test accuracy at batch size 16 and 20 epochs, classifying 98% of weed cases with high confidence and thereby reducing the need for precision herbicide applicators.\n\nexamined transfer learning strategies for weed–crop discrimination using MobileNet variants (V1–V3) and DenseNet variants (121\u002F169\u002F201). MobileNetV3 achieved higher classification accuracy than all DenseNet variants with substantially lower computational cost, demonstrating the practical advantage of lightweight architectures in mixed-crop scenarios.\n\n### 2.2 Vision transformer architectures\n\nVision transformers utilize multi-head self-attention to obtain spatial relationships over large distances and global contextual relationships that are not available in purely convolutional representations (). used ImageNet pre-trained weights and AutoAugment data augmentation for fine-tuning Swin Transformer v2 resulting in a 98.61% top-1 classification accuracy on a multi-species weed dataset. Grad-CAM++ visualization showed class-discriminative regions appropriate for robotic intervention guidance, illustrating the transformer's validity for fine-grained botanical classification. evaluated various deep learning architectures including transformer variants for crop and weed identification and found that attention-based models surpass CNN baselines on morphologically complex species pairs.\n\n### 2.3 Semantic segmentation for weed mapping\n\nPixel-wise segmentation enables exact spatial localization and is essential for site-specific weed management since robotic weeding and variable-rate herbicide application require it (). built a UNet-like encoder–decoder with a ResNet-34 backbone for classifying weeds at an early growth stage in uav -sorghum images. Multi-scale feature aggregation and skip connections allowed the model to handle occlusion and motion blur, achieving an 89% _F_ 1-score for intra-row weed detection. tried to solve the problem of weed detection in heavily occluded potato field settings and demonstrated that deep learning remains robust under dense canopy conditions. introduced SegFormer, which couples a hierarchical Mix Transformer (mit) encoder with a lightweight all-mlp decoder, achieving competitive segmentation without positional encodings across standard benchmarks.\n\n### 2.4 Explainable AI in agricultural systems\n\nInterpretability is essential for building agronomist confidence and regulatory acceptance, as it allows practitioners to verify that model decisions reflect genuine plant morphology rather than dataset artifacts (). introduced Gradient-weighted Class Activation Mapping (Grad-CAM), which produces class-discriminative localizationmaps by backpropagating class scores through the final convolutional layer without modifying the network architecture. Grad-CAM enhances weakly supervised localization, improves adversarial robustness, and provides spatial explanations for all stakeholders.\n\nproposed an interpretable deep learning method for sorghum weed classification based on a specially designed DenseNet-169 supplemented by Dimensional Convolutional Feature Fusion (dcff) and Deformable Multi-Kernel Channel Fusion (dmkcf) modules for extracting features at different scales. Grad-CAM and Local Interpretable Model-agnostic Explanations (lime) were used for the interpretability at the global and instance levels respectively. The method achieved 99.07% accuracy and 100% precision for broad-leaf weed detection.\n\nKarim et al. employed MobileNetV3-Large with custom dropout-dense layers for real-time grape leaf disease detection on the NVIDIA Jetson Nano edge platform. The model achieved 99.66% training accuracy and 99.42% test accuracy. Grad-CAM was used to localize diseased regions, enabling automated recognition of both real-time and stored images through a custom graphical interface.\n\n### 2.5 Digital twins in agriculture: sensing requirements\n\nFarm Digital Twins require a real-time perception layer that converts field imagery into structured data streams for simulation and decision support. Deep learning-based weed detection fulfils this role by providing species inventory, spatial weed maps, and confidence scores that the DT simulation engine consumes directly. The absence of integrated, interpretable, and spatially precise weed detection frameworks represents a key gap in current Farm-DT implementations, which this study addresses.\n\n### 2.6 Structured comparison and research gaps\n\n[Table 1](https:\u002F\u002Fwww.frontiersin.org\u002Fjournals\u002Fsustainable-food-systems\u002Farticles\u002F10.3389\u002Ffsufs.2026.1841305\u002Ffull#T1) shows a detailed comparison, covering nine dimensions, of major weed detection studies. There are still four significant issues that the existing literature has not addressed.\n\n*   Most of the current research focuses on assessing traditional CNN families (such as ResNet, VGG, Inception) and in many cases do not perform systematic comparisons between these and recent transformer-based models (Swin, EfficientNetV2, ConvNeXt) on the same benchmark under exactly the same experimental conditions.\n\n*   Performance evaluations are limited to scalar measures (accuracy, _F_ 1-score) without any quantitative analysis that would verify that classification decisions actually rely on biologically significant characteristics instead of background correlations or dataset artifacts.\n\n*   Classification and segmentation are handled separately; no end-to-end pipelines incorporating both modalities for full weed management decision support.\n\n*   Most studies employ laboratory-acquired or synthetically augmented datasets that fail to capture the phenotypic diversity and imaging variability present in operational agricultural environments.\n\nThis study addresses each of the four research gaps identified in the literature, as follows: (1) A thorough performance comparison of eleven CNN and transformer-based models with McNemar significance tests, paired t-tests, and 95% confidence intervals; (2) Quantitative Grad-CAM results show that 94.3% of the top-activated pixels are indeed ground-truth foliage areas; (3) an integrated classification-segmentation model capable of providing species identification together with spatial localization at the pixel level; (4) the highly realistic MFWD dataset collected in indoor phenotyping laboratory conditions (); and (5) a three-experiment ablation study that quantified the independent contributions of transfer learning, data augmentation, and loss function design.\n\n## 3 Materials\n\n### 3.1 Moving fields weed dataset (MFWD)\n\n#### 3.1.1 Dataset characterization\n\nAll experiments use the publicly available **Moving Fields Weed Dataset (****MFWD****)** (), comprising 94,321 high-resolution rgb images of **28 weed species** common in German maize and sorghum production systems (DOI: [10.14459\u002F2023md1717366](https:\u002F\u002Fdoi.org\u002F10.14459\u002F2023md1717366)). Images were acquired in a _fully automated, high-throughput indoor phenotyping facility_—not under natural field conditions— using a conveyor-mounted imaging platform that photographs individual microplot trays (40 × 30 × 22 cm) under controlled, uniform illumination at high temporal resolution. This setup tracks more than 5,000 individual plants across all developmental stages, from cotyledon emergence to reproductive phases, providin","Frontiers in Sustainable Food Systems 2026-07-08","2026-07-08T01:00:00Z","论文",10,false,75,{"impact":17,"substance":18,"depth":19,"authority":20,"freshness":21,"relevant":22,"comment":23},20,22,18,13,2,1,"提出面向农田数字孪生的深度学习感知框架，方法新颖、数据详实，但发表于2026年7月8日，时效性较低。",[25],{"name":10,"url":6},[27,28,29,30],"深度学习","农业数字孪生","杂草识别","语义分割",{"code":32,"slug":33,"name":34,"path":35,"parent_code":36},"I2","agricultural-ai","农业AI与大模型","\u002Ftopic\u002Fdigital-agriculture\u002Fagricultural-ai","I",[],{"status":39,"locked":14,"version":40,"method":41,"evidence":42},"classified","2026-10-22.2","rules",{"candidates":43,"agricultural_evidence":45},[44,50,57],{"code":32,"score":20,"supported":45,"title":46,"abstract":48,"tags":49},true,[47],"deep learning",[27],[27,30],{"code":51,"score":52,"supported":45,"title":53,"abstract":54,"tags":56},"I1",3,[],[55],"数字孪生",[],{"code":58,"score":22,"supported":14,"title":59,"abstract":60,"tags":61},"F4",[],[],[29],[63,64],"农业数字孪生 杂草识别 深度学习 语义分割","农业数字孪生 杂草识别","农业数字孪生杂草识别深度学习语义分割-113",14,"10.3389\u002Ffsufs.2026.1841305\u002Ffull","2026-07-31T00:00:56.059130Z",{"total":70,"page":22,"page_size":70,"items":71},6,[72,144,207,264,307,369],{"id":73,"title":74,"url":75,"summary":76,"summary_zh":77,"content":8,"source_name":78,"source_url":75,"published_at":79,"category":12,"cover_url":8,"hotness":13,"is_selected":14,"score":80,"score_detail":81,"sources":86,"tags":88,"primary_topic":92,"secondary_topics":97,"classification":99,"search_phrases":123,"slug":126,"view_count":84,"doi":127,"paper":128,"created_at":143},3618,"Multi-Target Remote Sensing Segmentation with Cross-Scale Attention for Urban-Ecological Intelligence","https:\u002F\u002Fdoi.org\u002F10.56201\u002Fijgem.vol.12.no1.2026.pg131.154","Remote sensing image segmentation is essential to extract valuable information from satellite and aerial images to achieve significant applications such as urban planning and ecological monitoring. Yet, it is hard to accurately segment diverse and complicated features because of the constraints of conventional approaches and the requirement to process variable object sizes and complicated boundaries. This research aims to address the above difficulties through the use and assessment of advanced deep learning models specifically UNet, SegNet, and TransU-Net in multi target semantic segmentation. These networks have been applied in single-object segmentation tasks with a focus on structures and roads, while the UNet architecture was additionally utilized for multi-object segmentation comprising buildings, woods, grasslands, water bodies, and cultivated land. Using appropriate remote sensing datasets, the accuracy of the models was thoroughly evaluated using common evaluation metrics, including pixel accuracy, Intersection over Union (IoU), and F1-score. The experimental outcomes illustrate the capabilities of these deep learning methods to provide accurate identification of essential features. Notably, the maximum attainable accuracy for single-object building segmentation was 96.85%, whereas the overall accuracy for multi-object segmentation with the UNet was 84.2%. The experimental results show these deep learning algorithms can effectively separate different targets from remote sensing imagery, thus making them fundamental tools for geospatial analysis and related fields.","遥感图像分割对于从卫星和航空图像中提取有价值信息，以实现城市规划和生态监测等重要应用至关重要。然而，由于传统方法的局限以及处理可变目标尺寸和复杂边界的需要，准确分割多样且复杂的地物十分困难。本研究旨在通过使用和评估先进的深度学习模型，特别是UNet、SegNet和TransU-Net在多目标语义分割中的应用，来解决上述难题。这些网络已被应用于单目标分割任务，重点关注建筑物和道路，而UNet架构还被额外用于多目标分割，包括建筑物、林地、草地、水体和耕地。利用适当的遥感数据集，通过常用评价指标对模型的精度进行了全面评估，包括像素精度、交并比（IoU）和F1分数。实验结果表明了这些深度学习方法在准确识别关键地物方面的能力。值得注意的是，单目标建筑物分割的最高可达精度为96.85%，而使用UNet进行多目标分割的总体精度为84.2%。实验结果表明，这些深度学习算法能够有效地从遥感图像中分离出不同目标，从而使其成为地理空间分析及相关领域的基础工具。","IIARD INTERNATIONAL JOURNAL OF GEOGRAPHY AND ENVIRONMENTAL MANAGEMENT","2026-09-25T00:00:00Z",63,{"impact":82,"substance":19,"depth":83,"authority":13,"freshness":84,"relevant":22,"comment":85},12,15,8,"将UNet、SegNet、TransU-Net用于多目标遥感语义分割，建筑单目标精度96.85%、多目标84.2%，方法常规但结论对农业与生态遥感监测有参考价值。",[87],{"name":78,"url":75},[89,27,30,90,91],"农业人工智能","遥感","生态监测",{"code":93,"slug":94,"name":95,"path":96,"parent_code":36},"I3","remote-sensing","遥感与农情监测","\u002Ftopic\u002Fdigital-agriculture\u002Fremote-sensing",[98],{"code":32,"slug":33,"name":34,"path":35,"parent_code":36},{"status":39,"locked":14,"version":40,"method":41,"evidence":100},{"candidates":101,"agricultural_evidence":45},[102,107,112,118],{"code":93,"score":83,"supported":45,"title":103,"abstract":105,"tags":106},[104],"remote sensing",[90,104],[90],{"code":32,"score":108,"supported":45,"title":109,"abstract":110,"tags":111},9,[],[27,47],[89,27,30],{"code":113,"score":52,"supported":45,"title":114,"abstract":115,"tags":117},"I4",[],[116],"空间分析",[],{"code":119,"score":22,"supported":14,"title":120,"abstract":121,"tags":122},"H6",[],[],[91],[124,125],"UNet 遥感 多目标分割","TransU-Net 城市生态 遥感","UNet遥感多目标分割-3618","10.56201\u002Fijgem.vol.12.no1.2026.pg131.154",{"doi":127,"openalex_id":129,"authors":130,"venue":78,"cited_by_count":133,"oa_url":134,"card":135,"direction":141,"ingested_from":142},"W7214298981",[131],{"name":132,"orcid":8},"Chibueze Favour Aririguzo",0,"https:\u002F\u002Fiiardjournals.org\u002Fget\u002FIJGEM\u002FVOL. 12 NO. 1 2026\u002FMulti-Target Remote Sensing Segmentation 131-154.pdf",{"tldr":136,"method":137,"finding":138,"direction":139,"opportunity":140},"评估UNet、SegNet和TransU-Net在多目标遥感语义分割中的性能，用于城市与生态监测。","使用UNet、SegNet、TransU-Net模型，在遥感数据集上以像素精度、","单目标建筑分割最高精度96.85%，UNet多目标分割总体精度84.2%。","农业遥感与作物表型","可探索跨尺度注意力机制提升多目标分割精度，并迁移至耕地与作物精细分类。","智慧农业 \u002F 农业物联网","openalex","2026-09-27T23:30:28.366758Z",{"id":145,"title":146,"url":147,"summary":148,"summary_zh":149,"content":8,"source_name":150,"source_url":147,"published_at":151,"category":12,"cover_url":8,"hotness":13,"is_selected":14,"score":152,"score_detail":153,"sources":156,"tags":158,"primary_topic":161,"secondary_topics":162,"classification":163,"search_phrases":184,"slug":187,"view_count":52,"doi":188,"paper":189,"created_at":206},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",67,{"impact":82,"substance":19,"depth":154,"authority":20,"freshness":84,"relevant":22,"comment":155},16,"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。",[157],{"name":150,"url":147},[159,89,27,29,160],"智慧农业","精准农业",{"code":32,"slug":33,"name":34,"path":35,"parent_code":36},[],{"status":39,"locked":14,"version":40,"method":41,"evidence":164},{"candidates":165,"agricultural_evidence":45},[166,175,180],{"code":32,"score":167,"supported":45,"title":168,"abstract":169,"tags":174},19,[47],[170,27,171,172,47,173],"目标检测","机器学习","machine learning","YOLO",[89,27],{"code":51,"score":154,"supported":45,"title":176,"abstract":178,"tags":179},[177],"precision agriculture",[160,177],[159,160],{"code":58,"score":22,"supported":14,"title":181,"abstract":182,"tags":183},[],[],[29],[185,186],"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":188,"openalex_id":190,"authors":191,"venue":150,"cited_by_count":133,"oa_url":147,"card":200,"direction":204,"ingested_from":142},"W7213558223",[192,194,196,198],{"name":193,"orcid":8},"Njoku Camillus Ekene",{"name":195,"orcid":8},"Francis A. Okoye",{"name":197,"orcid":8},"Ebere Uzoka Chidi",{"name":199,"orcid":8},"OGBU MARY NNENNA",{"tldr":201,"method":202,"finding":203,"direction":204,"opportunity":205},"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。","农业人工智能与决策模型","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":208,"title":209,"url":210,"summary":211,"summary_zh":212,"content":8,"source_name":213,"source_url":210,"published_at":214,"category":12,"cover_url":8,"hotness":13,"is_selected":14,"score":15,"score_detail":215,"sources":217,"tags":219,"primary_topic":220,"secondary_topics":221,"classification":222,"search_phrases":244,"slug":247,"view_count":52,"doi":248,"paper":249,"created_at":263},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":83,"substance":17,"depth":19,"authority":66,"freshness":84,"relevant":22,"comment":216},"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。",[218],{"name":213,"url":210},[159,89,27,29,160],{"code":32,"slug":33,"name":34,"path":35,"parent_code":36},[],{"status":39,"locked":14,"version":40,"method":41,"evidence":223},{"candidates":224,"agricultural_evidence":45},[225,229,233,240],{"code":51,"score":154,"supported":45,"title":226,"abstract":227,"tags":228},[177],[160,177],[159,160],{"code":32,"score":154,"supported":45,"title":230,"abstract":231,"tags":232},[47],[27,47],[89,27],{"code":234,"score":66,"supported":45,"title":235,"abstract":237,"tags":239},"I7",[236],"bibliometric",[238,236],"文献计量",[],{"code":58,"score":22,"supported":14,"title":241,"abstract":242,"tags":243},[],[],[29],[245,246],"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":248,"openalex_id":250,"authors":251,"venue":213,"cited_by_count":133,"oa_url":8,"card":258,"direction":204,"ingested_from":142},"W7212352052",[252,255],{"name":253,"orcid":254},"Furkan Ulaş","https:\u002F\u002Forcid.org\u002F0009-0002-3052-4457",{"name":256,"orcid":257},"Muhammad Aasım","https:\u002F\u002Forcid.org\u002F0000-0002-8524-9029",{"tldr":259,"method":260,"finding":261,"direction":204,"opportunity":262},"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",{"id":265,"title":266,"url":267,"summary":268,"summary_zh":8,"content":269,"source_name":270,"source_url":8,"published_at":271,"category":12,"cover_url":8,"hotness":13,"is_selected":14,"score":272,"score_detail":273,"sources":276,"tags":278,"primary_topic":282,"secondary_topics":283,"classification":285,"search_phrases":301,"slug":304,"view_count":70,"doi":305,"paper":8,"created_at":306},413,"深度学习支持下地块尺度农田宜机化遥感评价方法","http:\u002F\u002Fwww.tcsae.org\u002Fcn\u002Farticle\u002Fdoi\u002F10.11975\u002Fj.issn.1002-6819.202510247","《农业工程学报》2026年第42卷第9期刊载华中师范大学等单位杨子恒、胡琼、吴尚岐、徐保东、蔡志文论文。DOI: 10.11975\u002Fj.issn.1002-6819.202510247。研究基于多源中、高分辨率遥感数据和深度学习语义分割算法,自动提取农田地块与道路等基础设施的空间分布信息,结合数字高程模型与水旱分布数据,从地块形状、规模、平整度以及道路通达度4个维度建立综合评价体系,以当阳市为研究区域。结果:语义分割模型农田地块与道路面积准确度分别达到0.92和0.79,边界匹配度0.85和0.86;研究区宜机化良好农田面积占比达67.3%,丘陵山区乡镇普遍面临道路通达度不足与地块平整度较差问题。","**摘要:**\n\n农业机械化是推进农业现代化与保障国家粮食安全的关键支撑，而农田宜机化水平是决定农机作业效率的重要基础。然而，现有研究多集中于宜机化的必要性论证与成本效益分析，缺乏一套系统的精细化评价方法，难以科学性与有针对性的指导实施大区域农田宜机化改造工作。该研究基于多源中、高分辨率遥感数据和深度学习语义分割算法，自动提取农田地块与道路等基础设施的空间分布信息，为农田宜机化提供关键数据基础，同时结合数字高程模型与水旱分布数据，从地块形状、规模、平整度以及道路通达度4个维度建立综合评价体系，并以当阳市为研究区域，实现地块尺度农田宜机化水平精细化评估。结果表明：1）语义分割模型在地块提取与道路识别中表现良好，农田地块与道路的面积准确度分别达到0.92和0.79，边界匹配度分别达到0.85和0.86；共识别出农田地块46.8万个，道路总长度9 747.23 km；2）研究区农田宜机化水平空间分布差异显著，高宜机化区集中分布于东南部平原，在中部平原与低丘地带呈分散分布，西部丘陵区的宜机化水平较低；研究区宜机化分级结果显示，宜机化良好的农田面积占比达67.3%，乡镇尺度归因分析表明，除草埠湖镇外，地块形状与规模是各乡镇宜机化水平的共性制约因素，丘陵山区乡镇还普遍面临道路通达度不足与地块平整度较差的问题。该研究构建的评价方法能够有效、精细地表征地块尺度的农田宜机化水平，为大范围农田宜机化动态监测与差异化改造策略制定提供数据支持和决策依据。\n\n**Abstract:**\n\nThe adaptability level of farmland to agricultural machinery (i.e., the degree to which farmland facilitates agricultural machinery) can determine the efficiency and cost of mechanized operations in modern agriculture. However, existing research has focused mostly on the machinery adaptability or cost-benefit analysis after farmland consolidation. It is lacking in a systematic and refined evaluation of national food security. It is often required to implement the high adaptability of the large-scale farmland machinery. This study aims to automatically extract the spatial distribution of farmland parcels and roads using multi-source medium- and high-resolution remote sensing data and the deep learning semantic segmentation network HRNet (High-Resolution Network). A multi-task learning was employed to simultaneously predict the farmland, parcel boundaries, and roads. Parcel boundaries were also refined using conditional probability post-processing. Thereby, fine-scale data were provided to assess the adaptability level of farmland machinery. A 30 m digital elevation model (DEM) was combined with the paddy\u002Fdryland distribution data from a random forest classifier. An evaluation system was established from four dimensions: parcel shape, parcel size, surface evenness, and road accessibility. Seven indicators were obtained to determine the weight of each indicator. Criteria Importance Through Intercriteria Correlation (CRITIC) and the parcel-level Farmland Machinery Adaptability Index (FMAI) were subsequently calculated to validate in Dangyang City, Hubei Province. The results show that the HRNet semantic segmentation model performed well in both parcel and road extraction, with an area F1 score of 0.92 for farmland and 0.79 for roads, as well as an edge F1-score (F1-edge) of 0.85 for parcel boundaries and 0.86 for roads. A total of 468 000 farmland parcels and 9747.23 km of roads were identified in the study area. The spatial distribution of farmland machinery adaptability exhibited significant heterogeneity: High-adaptability areas (FMAI > 77.3) were concentrated in the southeastern Juzhang River alluvial plain in the central plains and low hilly areas, while the western hilly and mountainous regions generally shared the low adaptability levels. The grading results indicated that the area proportion of grades 1 and 2 (good adaptability) accounted for 67.3% of the total farmland area, but their parcel number proportion was only 38.6%, indicating the fragmented parcels generally had the lower adaptability grades. Township-scale attribution analysis revealed that the irregular parcel shape and small parcel size were common constraints to machinery adaptability over all townships, except for Caobuhu Town. The additional constraints were attributed to the insufficient road accessibility (road accessibility score 62-70) and low surface evenness (evenness score 78-85) in hilly and mountainous townships (e.g., Wangdian and Miaoqian). The evaluation can effectively and finely characterize the parcel-level farmland machinery adaptability and its limiting factors. The finding can provide data support and decision-making for dynamic monitoring of large-scale machinery adaptability, differentiated consolidation, and high-standard construction in farmland.\n\n[![Image 1](http:\u002F\u002Fwww.tcsae.org\u002FfileNYGCXB\u002Fjournal\u002Farticle\u002Fnygcxb\u002F2026\u002F9\u002Fshy-202510247-1.jpg)](http:\u002F\u002Fwww.tcsae.org\u002FfileNYGCXB\u002Fjournal\u002Farticle\u002Fnygcxb\u002F2026\u002F9\u002Fshy-202510247-1.jpg \"Figure1.Remote sensing imagery and samples of the study area\")\n\n**图 1** 研究区域遥感影像与样本\n\n**Figure 1.** Remote sensing imagery and samples of the study area\n\n[![Image 2](http:\u002F\u002Fwww.tcsae.org\u002FfileNYGCXB\u002Fjournal\u002Farticle\u002Fnygcxb\u002F2026\u002F9\u002Fshy-202510247-2.jpg)](http:\u002F\u002Fwww.tcsae.org\u002FfileNYGCXB\u002Fjournal\u002Farticle\u002Fnygcxb\u002F2026\u002F9\u002Fshy-202510247-2.jpg \"Figure2.Digital elevation model, slope and distribution of paddy and dryland\")\n\n**图 2** 数字高程模型、坡度与水田旱地分布数据\n\n**Figure 2.** Digital elevation model, slope and distribution of paddy and dryland\n\n[![Image 3](http:\u002F\u002Fwww.tcsae.org\u002FfileNYGCXB\u002Fjournal\u002Farticle\u002Fnygcxb\u002F2026\u002F9\u002Fshy-202510247-3.jpg)](http:\u002F\u002Fwww.tcsae.org\u002FfileNYGCXB\u002Fjournal\u002Farticle\u002Fnygcxb\u002F2026\u002F9\u002Fshy-202510247-3.jpg \"Figure3.Architecture of the HRNet\")\n\n**图 3** 高分辨率特征网络结构\n\n**Figure 3.** Architecture of the HRNet\n\n[![Image 4](http:\u002F\u002Fwww.tcsae.org\u002FfileNYGCXB\u002Fjournal\u002Farticle\u002Fnygcxb\u002F2026\u002F9\u002Fshy-202510247-4.jpg)](http:\u002F\u002Fwww.tcsae.org\u002FfileNYGCXB\u002Fjournal\u002Farticle\u002Fnygcxb\u002F2026\u002F9\u002Fshy-202510247-4.jpg \"Figure4.Extraction results of parcels and roads in the study area\")\n\n**图 4** 研究区地块与道路提取结果\n\n**Figure 4.** Extraction results of parcels and roads in the study area\n\n[![Image 5](http:\u002F\u002Fwww.tcsae.org\u002FfileNYGCXB\u002Fjournal\u002Farticle\u002Fnygcxb\u002F2026\u002F9\u002Fshy-202510247-5.jpg)](http:\u002F\u002Fwww.tcsae.org\u002FfileNYGCXB\u002Fjournal\u002Farticle\u002Fnygcxb\u002F2026\u002F9\u002Fshy-202510247-5.jpg \"Figure5.Spatial distribution of farmland machinery adaptability index in the study area\")\n\n**图 5** 研究区农田宜机化指数空间分布\n\n**Figure 5.** Spatial distribution of farmland machinery adaptability index in the study area\n\n[![Image 6](http:\u002F\u002Fwww.tcsae.org\u002FfileNYGCXB\u002Fjournal\u002Farticle\u002Fnygcxb\u002F2026\u002F9\u002Fshy-202510247-6.jpg)](http:\u002F\u002Fwww.tcsae.org\u002FfileNYGCXB\u002Fjournal\u002Farticle\u002Fnygcxb\u002F2026\u002F9\u002Fshy-202510247-6.jpg \"Figure6.Area distribution of different farmland machinery adaptability grades in various townships\")\n\n**图 6** 各乡镇不同农田宜机化等级的面积分布\n\n**Figure 6.** Area distribution of different farmland machinery adaptability grades in various townships\n\n[![Image 7](http:\u002F\u002Fwww.tcsae.org\u002FfileNYGCXB\u002Fjournal\u002Farticle\u002Fnygcxb\u002F2026\u002F9\u002Fshy-202510247-7.jpg)](http:\u002F\u002Fwww.tcsae.org\u002FfileNYGCXB\u002Fjournal\u002Farticle\u002Fnygcxb\u002F2026\u002F9\u002Fshy-202510247-7.jpg \"Figure7.Area-weighted mean values of farmland machinery adaptability indicators across townships\")\n\n**图 7** 各乡镇农田宜机化评价指标面积加权均值对比\n\n**Figure 7.** Area-weighted mean values of farmland machinery adaptability indicators across townships\n\n[![Image 8](http:\u002F\u002Fwww.tcsae.org\u002FfileNYGCXB\u002Fjournal\u002Farticle\u002Fnygcxb\u002F2026\u002F9\u002Fshy-202510247-8.jpg)](http:\u002F\u002Fwww.tcsae.org\u002FfileNYGCXB\u002Fjournal\u002Farticle\u002Fnygcxb\u002F2026\u002F9\u002Fshy-202510247-8.jpg \"Figure8.Changes in the mean values of farmland machinery adaptability indicators under different scale errors\")\n\n**图 8** 不同比例误差下的宜机化评价指标均值变化\n\n**Figure 8.** Changes in the mean values of farmland machinery adaptability indicators under different scale errors\n\n[![Image 9](http:\u002F\u002Fwww.tcsae.org\u002FfileNYGCXB\u002Fjournal\u002Farticle\u002Fnygcxb\u002F2026\u002F9\u002Fshy-202510247-9.jpg)](http:\u002F\u002Fwww.tcsae.org\u002FfileNYGCXB\u002Fjournal\u002Farticle\u002Fnygcxb\u002F2026\u002F9\u002Fshy-202510247-9.jpg \"Figure9.Changes in the share of farmland parcels with different mechanization levels under different scale errors\")\n\n**图 9** 不同比例误差下不同宜机化等级的农田地块数量占比变化\n\n**Figure 9.** Changes in the share of farmland parcels with different mechanization levels under different scale errors\n\n**表 1** 精度评价指标\n\n**Table 1** Accuracy assessment metrics\n\n指标类别\n\nMetrics category 评价指标\n\nAssessment metrics 计算公式\n\nFormulation 适用对象\n\nObject 说明\n\nExplanation\n面积准确度\n\nArea accuracy 精确率Precision$ {\\mathrm{Precision}}=\\dfrac{{\\mathrm{TP}}}{{\\mathrm{TP+FP}}} $农田\n道路 TP、TN、FP和FN分别表示准确识别的目标像素数量、准确识别的非目标像素数量、目标错分的像素数量以及非目标错分的像元数量\n召回率Recall$ {\\mathrm{Recall}}=\\dfrac{{\\mathrm{TP}}}{{\\mathrm{TP+FN}}} $\n面积综合得分F1$ F1=\\dfrac{2\\text{×}{\\mathrm{Precision}}\\text{×}{\\mathrm{Recall}}}{{\\mathrm{Precision}}+{\\mathrm{Recall}}} $\n边界匹配度\n\nEdge accuracy 完整度Com$ \\mathrm{Com}=\\dfrac{L_{\\mathrm{pred}}\\cap L\\mathrm{_{truth}}}{L_{\\mathrm{truth}}} $地块边界\n道路 _L_ pred 与 _L_ truth 分别表示模型提取的线状地物和真实的线状地物，∩表示两者在彼此缓冲区内的重叠长度，本研究缓冲区大小设置为2.5 m\n精准度Corr$ \\mathrm{Corr}=\\dfrac{L_{\\mathrm{truth}}\\cap L_{\\mathrm{pred}}}{L_{\\mathrm{pred\\mathrm{ }}}} $\n边界综合得分F1-edge$ F1-\\text{edge}=\\dfrac{2\\text{×}\\mathrm{Com}\\text{×}\\mathrm{Corr}}{\\mathrm{Com}+\\mathrm{Corr}} $\n\n**表 2** 农田宜机化评价指标赋分标准\n\n**Table 2** Scoring standards of farmland machinery adaptability indicators\n\n一级指标\n\nPrimary indicators 二级指标\n\nSecondary indicators 赋分Score\n10 20 30 40 50 60 70 80 90 100\n地块形状\n\nParcel shape 地块长宽比-水田 PAR-P(1.0, 1.5](1.5, 2.0](2.0, 2.5](2.5, 3.0]>5.0(3.0, 5.0]\n地块长宽比-旱地 PAR-D(1.0, 1.5](1.5, 2.0](2.0, 2.5](2.5, 3.0](3.0, 10.0](10.0, 20.0]>20.0\n地块宽度 PW\u002Fm[0, 5](5, 10](10, 15](15, 20](20, 25]>25\n地块与最小包围盒面积比\n\nPMBAR\u003C0.1[0.1, 0.2)[0.2, 0.3)[0.3, 0.4)[0.4, 0.5)[0.5, 0.6)[0.6, 0.7)[0.7, 0.8)[0.8, 0.9)>0.9\n地块规模\n\nParcel size 面积-水田 Area-P\u002Fhm 2(0, 0.03](0.03, 0.06](0.06, 0.13](0.13, 0.20](0.20, 0.33]>0.33\n面积-旱地 Area-D\u002Fhm 2(0, 0.03](0.03, 0.06](0.06, 0.33](0.33,0.66](0.66, 1.33](1.33, 3.33]>3.33\n地块平整度\n\nSurface levelness 坡度 Slope\u002F(°)>25[25, 15)[15, 10)[10, 6)[6, 2)[2, 0]\n农田道路通达度\n\nFarmland road accessibility 地块与道路距离 RD\u002Fm>100[50, 100)[20, 50)[10, 20)[5, 10)\u003C5\n道路宽度 RW\u002Fm 0(0, 1.5](1.5, 2.0](2.0, 3.0](3.0, 4.0](4.0, 5.0](5.0, 6.0]>6.0\n\n**表 3** 模型精度评价\n\n**Table 3** Model accuracy evaluation\n\n指标\n\nMetrics 面积准确度Area accuracy 边界匹配度Edge accuracy\nPrecision Recall F1 Com Corr F1-edge\n农田 0.908 0.946 0.927---\n地块边界---0.897 0.828 0.859\n道路 0.811 0.775 0.792 0.848 0.878 0.862\n\n**表 4** 农田宜机化指标权重\n\n**Table 4** Weights for farmland machinery adaptability indicators\n\n指标\n\nIndicator 地块形状Parcel shape 地块规模\n\nParcel size 地块平整度\n\nSurface levelness 农田道路通达度\n\nFarmland road accessibility\n地块长宽比\n\nPAR 地块宽度\n\nPBW 地块最小包围盒面积比\n\nPMBAR 最近道路距离\n\nRD 道路宽度\n\nRW\n主观权重Objective weight 0.083 0.083 0.083 0.250 0.250 0.125 0.125\n客观权重Subjective weight 0.177 0.134 0.097 0.141 0.097 0.186 0.168\n综合权重Comprehensive weight 0.107 0.081 0.058 0.256 0.176 0.169 0.153\n\n**表 5** 研究区农田宜机化分级统计结果\n\n**Table 5** Grading statistics result of farmland machinery adaptability in the study area\n\n宜机化等级\n\nGrading 宜机化指数\n\nFMAI 数量统计\n\nNumber statistics 面积统计\n\nArea statistics 宜机化指标统计\n\nMachinery adaptability indicators statistics\n农田地块数\n\nNumber of \n\nparcels 数量占比\n\nPercentage of \n\ntotal number\u002F%农田面积\n\nFarmland area\u002Fhm2 面积占比\n\nPercentage of \n\ntotal number\u002F%地块形状均值\n\nMean of\n\nP-shape 地块规模均值\n\nMean of\n\nP-size 地块平整度均值\n\nMean of SL 农田道路通\n\n达度均值\n\nMean of FRA\n一级 77.3～100 73297 15.65 24869.00 42.40 76.37 74.98 94.06 85.82\n二级 67.3～77.3 107234 22.90 14581.13 24.86 68.43 56.32 85.65 81.03\n三级 57.3～67.3 108130 23.09 9448.59 16.11 64.92 44.87 84.66 61.54\n四级 47.1～57.3 103400 22.08 7495.62 12.78 64.58 43.70 81.61 35.75\n五级 20.7～47.1 76243 16.28 2257.48 3.85 57.30 25.67 74.53 24.70\n合计 468304 100 58651.82 100 66.20 48.82 84.04 58.11\n\n[1]陈印军, 易小燕, 方琳娜, 等. 中国耕地资源与粮食增产潜力分析[J]. 中国农业科学, 2016, 49(6): 1117-1131. doi: [10.3864\u002Fj.issn.0578-1752.2016.06.008](https:\u002F\u002Fdx.doi.org\u002F10.3864\u002Fj.issn.0578-1752.2016.06.008)\nCHEN Yinjun, YI Xiaoyan, FANG Linna, et al. Analysis of cultivated land and grain production potential in China[J]. Scientia Agricultura Sinica, 2016, 49(6): 1117-1131. (in Chinese with English abstract) doi: [10.3864\u002Fj.issn.0578-1752.2016.06.008](https:\u002F\u002Fdx.doi.org\u002F10.3864\u002Fj.issn.0578-1752.2016.06.008)\n[2]卢成会, 祝坤. 农村劳动力老龄化趋势问题研究[J\u002FOL]. 中国农业资源与区划, 1-11[2025-09-10]. https:\u002F\u002Flink.cnki.net\u002Furlid\u002F11.3513.S.20250924.1520.006. \nLU Chenghui, ZHU Kun. Research on the problem of the aging trend of rural labor force[J\u002FOL]. Chinese Journal of Agricultural Resources and Regional Planning,1-11[2025-09-10]. https:\u002F\u002Flink.cnki.net\u002Furlid\u002F11.3513.S.20250924.1520.006. (in Chinese with English abstract)\n[3]胡志全, 朱殿霄, 辛岭, 等. 基于产业要素年代差距分析的农业现代化水平国际比较研究[J]. 中国农业科学, 2018, 51(7): 1412-1420. \nHU Zhiquan, ZHU Dianxiao, XIN Ling, et al. Comparison study on the level of international agricultural modernization based on the method of generation gap of industry elements[J]. Scientia Agricultura Sinica, 2018, 51(7): 1412-1420. (in Chinese with English abstract)\n[4]张洪程, 胡雅杰, 戴其根, 等. 中国大田作物栽培学前沿与创新方向探讨[J]. 中国农业科学, 2022, 55(22): 4373-4382. doi: [10.3864\u002Fj.issn.0578-1752.2022.22.004](https:\u002F\u002Fdx.doi.org\u002F10.3864\u002Fj.issn.0578-1752.2022.22.004)\nZHANG Hongcheng, HU Yajie, DAI Qigen, et al. Discussions on frontiers and directions of scientific and technological innovation in China’s field crop cultivation[J]. Scientia Agricultura Sinica, 2022, 55(22): 4373-4382. (in Chinese with English abstract) doi: [10.3864\u002Fj.issn.0578-1752.2022.22.004](https:\u002F\u002Fdx.doi.org\u002F10.3864\u002Fj.issn.0578-1752.2022.22.004)\n[5]DENG O, RAN J, HUANG S, et al. Managing fragmented croplands for environmental and economic benefits in China[J]. Nature Food, 2024, 5(3): 230-240. doi: [10.1038\u002Fs43016-024-00938-7](https:\u002F\u002Fdx.doi.org\u002F10.1038\u002Fs43016-024-00938-7)\n[6]魏鹏, 余敦, 胡宜之, 等. 规划管控约束下耕地细碎化评价与整治分区[J]. 农业工程学报, 2024, 40(3): 283-293. \nWEI Peng, YU Dun, HU Yizhi, et al. Evaluating cultivated land fragmentation and remediation zoning under planning management[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2024, 40(3): 283-293. (in Chinese with English abstract)\n[7]邹利林, 梁一凡, 林程, 等. 基于“田块-农户”尺度的耕地细碎化评价及影响因素探测[J]. 农业工程学报, 2025, 41(22): 306-315. \nZOU Lilin, LIANG Yifan, LIN Cheng, et al. Multidimensional evaluation and driving factors detection of cultivated land fragmentation based on plot-household scale[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2025, 41(22): 306-315. (in Chinese with English abstract)\n[8]廖宜涛, 施彬彬, 王传奇, 等. 南方丘陵山地油菜和小麦播种机械化发展现状与趋势[J]. 农业工程学报, 2025, 41(1): 12-26. doi: [10.11975\u002Fj.issn.1002-6819.202406151](https:\u002F\u002Fdx.doi.org\u002F10.11975\u002Fj.issn.1002-6819.202406151)\nLIAO Yitao, SHI Binbin, WANG Chuanqi, et al. Development status and trend of mechanization of rapeseed and wheat sowing in hilly and mountainous region of southern China[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2025, 41(1): 12-26. (in Chinese with English abstract) doi: [10.11975\u002Fj.issn.1002-6819.202406151](https:\u002F\u002Fdx.doi.org\u002F10.11975\u002Fj.issn.1002-6819.202406151)\n[9]周应恒, 胡凌啸. 对我国农田基础设施建设的思考[J]. 华中农业大学学报(社会科学版), 2016(2): 23-29, 135. doi: [10.13300\u002Fj.cnki.hnwkxb.2016.02.004](https:\u002F\u002Fdx.doi.org\u002F10.13300\u002Fj.cnki.hnwkxb.2016.02.004)\nZHOU Yingheng, HU Lingxiao. Some considerations on farmland infrastructure construction[J]. Journal of Huazhong Agricultural University (Social Sciences Edition), 2016(2): 23-29, 135. (in Chinese with English abstract) doi: [10.13300\u002Fj.cnki.hnwkxb.2016.02.004](https:\u002F\u002Fdx.doi.org\u002F10.13300\u002Fj.cnki.hnwkxb.2016.02.004)\n[10]张宗毅, 李玉珠, 张萌, 等. 基于农田宜机化视角的农用柴油技术效率提升路径[J]. 农业工程学报, 2024, 40(19): 62-71. doi: [10.11975\u002Fj.issn.1002-6819.202404019](https:\u002F\u002Fdx.doi.org\u002F10.11975\u002Fj.issn.1002-6819.202404019)\nZHANG Zongyi, LI Yuzhu, ZHANG Meng, et al. Pathways for enhancing the technological efficiency of agricultural diesel from the perspective of the level of farmland suitability for agricultural machinery operations[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2024, 40(19): 62-71. (in Chinese with English abstract) doi: [10.11975\u002Fj.issn.1002-6819.202404019](https:\u002F\u002Fdx.doi.org\u002F10.11975\u002Fj.issn.1002-6819.202404019)\n[11]毕红杰. 基于AHP对吉林省粮食综合生产能力评价与分析[J]. 中国农机化学报, 2015, 36(6): 323-327. doi: [10.13733\u002Fj.jcam.issn.2095-5553.2015.06.076](https:\u002F\u002Fdx.doi.org\u002F10.13733\u002Fj.jcam.issn.2095-5553.2015.06.076)\nBI Hongjie. Evaluation and analysis of grain comprehensive production capacity in Jilin province based on AHP[J]. Journal of Chinese Agricultural Mechanization, 2015, 36(6): 323-327. (in Chinese with English abstract) doi: [10.13733\u002Fj.jcam.issn.2095-5553.2015.06.076](https:\u002F\u002Fdx.doi.org\u002F10.13733\u002Fj.jcam.issn.2095-5553.2015.06.076)\n[12]袁鹏, 张宗毅, 李洪波. 农田宜机化水平与农户农机购置行为——基于全国丘陵山区10省4905户农户的数据[J]. 农业技术经济, 2024(10): 91-105. \nYUAN Peng, ZHANG Zongyi, LI Hongbo. Mechanization suitability of farmland and farmers agricultural machinery acquisition behavior: Based on data from 4905 farming households in 10 provinces in the hilly mountainous regions of China[J]. Journal of Agrotechnical Economics, 2024(10): 91-105. (in Chinese with English abstract)\n[13]陈乐为, 张宗毅, 李洪波, 等. 改地降本: 农田宜机化视角下的农机化成本效率分析——基于空间计量模型的实证研究[J]. 自然资源学报, 2024, 39(9): 2188-2205. doi: [10.31497\u002Fzrzyxb.20240911](https:\u002F\u002Fdx.doi.org\u002F10.31497\u002Fzrzyxb.20240911)\nCHEN Lewei, ZHANG Zongyi, LI Hongbo, et al. Farmland transformation to reduce costs: An empirical analysis of the cost efficiency of agricultural mechanization from the perspective of the level of farmland suitability for agricultural machinery operations based on spatial econometric models[J]. Journal of Natural Resources, 2024, 39(9): 2188-2205. (in Chinese with English abstract) doi: [10.31497\u002Fzrzyxb.20240911](https:\u002F\u002Fdx.doi.org\u002F10.31497\u002Fzrzyxb.20240911)\n[14]牛坡, 夏飞, 尤显平, 等. 重庆市宜机化土地整治效益研究[J]. 农业工程, 2021, 11(8): 86-89. \nNIU Po, XIA Fei, YOU Xianping, et al. Benefits of land consolidation for suitable large-medium farm mechanization of Chongqing[J]. Agricultural Engineering, 2021, 11(8): 86-89. (in Chinese with English abstract)\n[15]陈梅, 曹中华. 丘陵山区宜机化土地整治项目综合效益评价研究——基于改进AHP—模糊综合评价法[J]. 中国农机化学报, 2019, 40(1): 145-151. doi: [10.13733\u002Fj.jcam.issn.2095-5553.2019.01.27](https:\u002F\u002Fdx.doi.org\u002F10.13733\u002Fj.jcam.issn.2095-5553.2019.01.27)\nCHEN Mei, CAO Zhonghua. Research on comprehensive benefits evaluation of land consolidation in hilly and mountainous areas based on method of improved AHP-Fuzzy comprehensive evaluation[J]. Journal of Chinese Agricultural Mechanization, 2019, 40(1): 145-151. (in Chinese with English abstract) doi: [10.13733\u002Fj.jcam.issn.2095-5553.2019.01.27](https:\u002F\u002Fdx.doi.org\u002F10.13733\u002Fj.jcam.issn.2095-5553.2019.01.27)\n[16]林建平, 黄坤, 邓爱珍, 等. 耦合耕地综合质量和空间连通性的基本农田划定研究[J]. 农业机械学报, 2024, 55(7): 221-231, 251. \nLIN Jianping, HUANG Kun, DENG Aizhen, et al. Basic farmland demarcation coupling comprehensive quality and spatial connectivity of cultivated land[J]. Transactions of the Chinese Society for Agricultural Machinery, 2024, 55(7): 221-231, 251. (in Chinese with English abstract)\n[17]吴尚岐, 胡琼, 蔡志文, 等. 基于多源遥感和多分支时空特征融合网络的高标准农田建设地块形态变化[J]. 农业工程学报, 2025, 41(21): 277-286. \nWU Shangqi, HU Qiong, CAI Zhiwen, et al. Morphological variation in well-facilitated farmland parcels after construction using multi-source remote sensing and dual branch spatiotemporal fusion network[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2025, 41(21): 277-286. (in Chinese with English abstract)\n[18]刘光盛, 赵乐松, 程迎轩, 等. 基于限制因子的粤北丘陵山区耕地宜机化整治分区[J]. 农业工程学报, 2021, 37(12): 262-270. doi: [10.11975\u002Fj.issn.1002-6819.2021.12.030](https:\u002F\u002Fdx.doi.org\u002F10.11975\u002Fj.issn.1002-6819.2021.12.030)\nLIU Guangsheng, ZHAO Lesong, CHENG Yingxuan, et al. Land consolidation zoning of Northern Guangdong for suitable mechanization transformation in hilly and mountainous areas based on limiting factors[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2021, 37(12): 262-270. (in Chinese with English abstract) doi: [10.11975\u002Fj.issn.1002-6819.2021.12.030](https:\u002F\u002Fdx.doi.org\u002F10.11975\u002Fj.issn.1002-6819.2021.12.030)\n[19]董光龙, 赵轩, 刘金花, 等. 基于耕地质量评价与空间集聚特征的基本农田划定研究[J]. 农业机械学报, 2020, 51(2): 133-142. doi: [10.6041\u002Fj.issn.1000-1298.2020.02.015](https:\u002F\u002Fdx.doi.org\u002F10.6041\u002Fj.issn.1000-1298.2020.02.015)\nDONG Guanglong, ZHAO Xuan, LIU Jinhua, et al. Prime cropland assignment based on cultivated land quality evaluation and spatial cluster pattern[J]. Transactions of the Chinese Society for Agricultural Machinery, 2020, 51(2): 133-142. (in Chinese with English abstract) doi: [10.6041\u002Fj.issn.1000-1298.2020.02.015](https:\u002F\u002Fdx.doi.org\u002F10.6041\u002Fj.issn.1000-1298.2020.02.015)\n[20]刘彦文, 刘成武, 何宗宜, 等. 基于像元尺度耕地质量局部空间自相关的基本农田划定[J]. 农业机械学报, 2019, 50(5): 260-268, 319. \nLIU Yanwen, LIU Chengwu, HE Zongyi, et al. Delineation of basic farmland based on local spatial autocorrelation analysis of cultivated land quality in pixel scale[J]. Transactions of the Chinese Society for Agricultural Machinery, 2019, 50(5): 260-268, 319. (in Chin","农业工程学报 2026 42(9) 2026-08-09","2026-08-09T02:00:00Z",79,{"impact":19,"substance":18,"depth":19,"authority":66,"freshness":274,"relevant":22,"comment":275},7,"核心期刊论文，提出基于深度学习的农田宜机化遥感评价方法，数据详实，对农业现代化有实际指导意义。",[277],{"name":270,"url":267},[279,27,30,280,281],"数字农业","农田宜机化","遥感评价",{"code":32,"slug":33,"name":34,"path":35,"parent_code":36},[284],{"code":93,"slug":94,"name":95,"path":96,"parent_code":36},{"status":39,"locked":14,"version":40,"method":41,"evidence":286},{"candidates":287,"agricultural_evidence":45},[288,292,297],{"code":32,"score":20,"supported":45,"title":289,"abstract":290,"tags":291},[27],[27],[27,30],{"code":93,"score":293,"supported":45,"title":294,"abstract":295,"tags":296},11,[90],[90],[],{"code":51,"score":22,"supported":14,"title":298,"abstract":299,"tags":300},[],[],[279],[302,303],"农田宜机化 数字农业 深度学习 语义分割","农田宜机化 数字农业","农田宜机化数字农业深度学习语义分割-413","10.11975\u002Fj.issn.1002-6819.202510247","2026-08-11T23:57:03.136894Z",{"id":308,"title":309,"url":310,"summary":311,"summary_zh":312,"content":8,"source_name":313,"source_url":310,"published_at":314,"category":12,"cover_url":8,"hotness":13,"is_selected":14,"score":315,"score_detail":316,"sources":318,"tags":320,"primary_topic":323,"secondary_topics":329,"classification":331,"search_phrases":349,"slug":352,"view_count":133,"doi":353,"paper":354,"created_at":368},5387,"LIGHTWEIGHT FEATURE FUSED ATTENTION DRIVEN MULTI LEVEL HYBRID MOBILENET V2 FRAMEWORK FOR EARLY TOMATO LEAF DISEASE DETECTION","https:\u002F\u002Fdoi.org\u002F10.62441\u002Factainnovations.vi.1918","It is crucial to find and treat tomato crop diseases early and accurately to improve crop efficiency and make sure there is no scarcity of crop. However, computerized analysis is provocative due to lighting differences, compound backgrounds, and graphic likenesses among infection regions. Research work proposed a lightweight Attention Driven Multi Level Hybrid MobileNet-V2 algorithm that fuses the handcrafted texture structures with Deep Learning (DL) features, then image preprocessing is incorporated using Gaussian, Bilateral, and Guided filters to minimize noise by preserving the disease lesions. LAB color model along with Contrast Limited Adaptive Histogram Equalization (CLAHE) are used to improve the contrast. Otsu thresholding is involved to effectively separate the leaf segment from diseased parts. Local Binary Pattern (LBP) and Gray Level Co-occurrence Matrix (GLCM) descriptors are used to obtain fine-grained features, while high level features are gained by pretrained MobileNet-V2 model. Features fusion is fed to the fully connected layers of deep learning model for training and validation purpose. Results have revealed that the suggested model attains better classification performance as compared to existing models, making it an effective explanation for real time tomato leaf disease detection for smart agriculture applications.","早期准确发现并治疗番茄作物病害，对于提高作物效率、确保作物不短缺至关重要。然而，由于光照差异、复杂背景以及感染区域之间的图形相似性，计算机化分析具有挑战性。研究提出了一种轻量级注意力驱动多层混合MobileNet-V2算法，该算法将手工纹理结构与深度学习（DL）特征相融合，然后结合高斯、双边和引导滤波器进行图像预处理，以在保留病害病变的同时最小化噪声。采用LAB颜色模型和对比度受限自适应直方图均衡化（CLAHE）来提高对比度。引入Otsu阈值分割以有效分离叶片片段与病害部分。使用局部二值模式（LBP）和灰度共生矩阵（GLCM）描述符获取细粒度特征，同时通过预训练的MobileNet-V2模型获取高层特征。特征融合后输入深度学习模型的全连接层进行训练和验证。结果表明，与现有模型相比，所提出的模型取得了更好的分类性能，使其成为智慧农业应用中实时番茄叶片病害检测的有效解决方案。","Acta Innovations","2026-10-18T00:00:00Z",64,{"impact":82,"substance":19,"depth":154,"authority":13,"freshness":84,"relevant":22,"comment":317},"提出融合手工纹理特征与MobileNet-V2的轻量番茄叶病检测方法，有一定方法新颖性，但属单篇论文、应用范围有限，未达每日精选门槛。",[319],{"name":313,"url":310},[159,89,27,321,322],"番茄病害","图像识别",{"code":324,"slug":325,"name":326,"path":327,"parent_code":328},"B1","vegetables","蔬菜","\u002Ftopic\u002Fhorticulture\u002Fvegetables","B",[330],{"code":32,"slug":33,"name":34,"path":35,"parent_code":36},{"status":39,"locked":14,"version":40,"method":41,"evidence":332},{"candidates":333,"agricultural_evidence":45},[334,340,344],{"code":324,"score":83,"supported":45,"title":335,"abstract":337,"tags":339},[336],"tomato",[338,336],"番茄",[321],{"code":32,"score":108,"supported":45,"title":341,"abstract":342,"tags":343},[],[27,47],[89,322,27],{"code":51,"score":274,"supported":45,"title":345,"abstract":346,"tags":348},[],[159,347],"smart agriculture",[159],[350,351],"MobileNet-V2 番茄叶片病害","番茄叶病 深度学习 检测","MobileNet-V2番茄叶片病害-5387","10.62441\u002Factainnovations.vi.1918",{"doi":353,"openalex_id":355,"authors":356,"venue":313,"cited_by_count":133,"oa_url":362,"card":363,"direction":141,"ingested_from":142},"W7222977034",[357,359],{"name":358,"orcid":8},"Mrs. Vijetalaxmi A Hegde",{"name":360,"orcid":361},"Abhay Ramakant Kasetwar","https:\u002F\u002Forcid.org\u002F0000-0002-5800-8764","https:\u002F\u002Factainnovations.com\u002Findex.php\u002Fpub\u002Farticle\u002Fdownload\u002F1918\u002F981",{"tldr":364,"method":365,"finding":366,"direction":204,"opportunity":367},"提出轻量注意力多级混合MobileNet-V2框架，融合手工纹理与深度学习特征实现番茄叶片病害早期检","高斯\u002F双边\u002F引导滤波去噪，LAB+CLAHE增强，Otsu分割，LBP与GLCM","所提模型分类性能优于现有模型，可有效用于智慧农业实时番茄叶片病害检测。","可探索多作物跨域泛化、田间复杂光照下轻量模型部署与可解释性，及病害早期分级检测。","2026-10-20T23:30:10.424511Z",{"id":370,"title":371,"url":372,"summary":373,"summary_zh":374,"content":8,"source_name":375,"source_url":372,"published_at":314,"category":12,"cover_url":8,"hotness":13,"is_selected":14,"score":376,"score_detail":377,"sources":379,"tags":381,"primary_topic":384,"secondary_topics":385,"classification":386,"search_phrases":406,"slug":409,"view_count":133,"doi":410,"paper":411,"created_at":419},5385,"A Comprehensive Review of Deep Learning and Computer Vision Techniques for Fruit Quality Assessment and Classification","https:\u002F\u002Fdoi.org\u002F10.25258\u002Fijddt.16.75s.151","Fruit image analysis has become an essential research area in precision agriculture, enabling automated fruit classification, quality grading, maturity estimation, disease diagnosis, yield prediction, and intelligent harvesting through advances in computer vision and artificial intelligence (AI).Recent progress in imaging technologies, large-scale benchmark datasets, and deep learning has substantially improved the accuracy, robustness, and scalability of automated fruit analysis systems.This survey presents a comprehensive and systematic review of image processing and AI techniques for fruit image analysis.It covers the complete image analysis pipeline, including image acquisition, preprocessing, enhancement, color space transformation, segmentation, feature extraction, and handcrafted texture descriptors.Furthermore, conventional machine learning algorithms are reviewed alongside state-of-the-art deep learning architectures, including Convolutional Neural Networks (CNNs), Vision Transformers (ViTs), and Vision Mamba models, for applications such as fruit recognition, quality assessment, disease detection, ripeness estimation, and object detection.Publicly available benchmark datasets, evaluation metrics, and performance comparison strategies are also summarized to facilitate objective assessment of existing approaches.In addition, the survey discusses key research challenges, including illumination variations, occlusion, complex orchard environments, limited annotated datasets, computational complexity, and domain adaptation, while highlighting emerging research trends such as self-supervised learning, multimodal learning, explainable AI, lightweight deep networks, foundation models, and edge intelligence for real-time deployment.By providing a unified review of methodologies, datasets, evaluation protocols, challenges, and future research directions, this survey offers a comprehensive reference for researchers and practitioners developing intelligent and scalable fruit image analysis systems for precision agriculture.","水果图像分析已成为精准农业中的一个重要研究领域，借助计算机视觉和人工智能（AI）的进步，实现了果实自动分类、品质分级、成熟度估计、病害诊断、产量预测和智能采收。成像技术、大规模基准数据集和深度学习的最新进展显著提升了自动化水果分析系统的准确性、鲁棒性和可扩展性。本综述对用于水果图像分析的图像处理和AI技术进行了全面而系统的回顾。内容涵盖完整的图像分析流程，包括图像采集、预处理、增强、颜色空间变换、分割、特征提取和手工纹理描述符。此外，综述了传统机器学习算法以及最先进的深度学习架构，包括卷积神经网络（CNN）、视觉Transformer（ViT）和视觉Mamba模型，用于果实识别、品质评估、病害检测、成熟度估计和目标检测等应用。同时总结了公开可用的基准数据集、评价指标和性能比较策略，以便对现有方法进行客观评估。此外，本综述讨论了关键研究挑战，包括光照变化、遮挡、复杂果园环境、标注数据集有限、计算复杂度和域适应，并重点介绍了新兴研究趋势，如自监督学习、多模态学习、可解释AI、轻量级深度网络、基础模型和面向实时部署的边缘智能。通过提供方法、数据集、评价协议、挑战和未来研究方向的统一综述，本综述为开发面向精准农业的智能且可扩展的水果图像分析系统的研究人员和实践者提供了全面的参考。","International Journal of Drug Delivery Technology",56,{"impact":84,"substance":19,"depth":154,"authority":70,"freshness":84,"relevant":22,"comment":378},"系统性综述覆盖CNN、ViT、Vision Mamba等前沿方法，信息量足，但期刊权威性一般且偏学术综述，公共价值有限，适合主题聚合而非每日精选头条。",[380],{"name":375,"url":372},[159,89,382,27,383],"计算机视觉","水果品质检测",{"code":32,"slug":33,"name":34,"path":35,"parent_code":36},[],{"status":39,"locked":14,"version":40,"method":41,"evidence":387},{"candidates":388,"agricultural_evidence":45},[389,396,400],{"code":32,"score":390,"supported":45,"title":391,"abstract":393,"tags":395},28,[47,392],"computer vision",[170,27,171,382,172,47,394,392],"artificial intelligence",[89,27,382],{"code":51,"score":274,"supported":45,"title":397,"abstract":398,"tags":399},[],[160,177],[159],{"code":401,"score":52,"supported":45,"title":402,"abstract":403,"tags":405},"B2",[],[404],"orchard",[],[407,408],"水果品质 计算机视觉 深度学习","农业人工智能 水果品质检测 计算机视觉 智慧农业","水果品质计算机视觉深度学习-5385","10.25258\u002Fijddt.16.75s.151",{"doi":410,"openalex_id":412,"authors":413,"venue":375,"cited_by_count":133,"oa_url":372,"card":414,"direction":141,"ingested_from":142},"W7222892831",[],{"tldr":415,"method":416,"finding":417,"direction":139,"opportunity":418},"系统综述深度学习与计算机视觉在水果品质评估与分级中的全流程方法。","综述图像采集到分割、特征提取及CNN、ViT、Vision Mamba等模型与公","深度学习显著提升水果分析精度与鲁棒性，但光照、遮挡和小样本仍是主要瓶颈。","可探索自监督与轻量化模型在复杂果园遮挡、小样本场景下的边缘实时部署。","2026-10-20T23:30:10.066042Z"]