[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"item-586":3,"related-586":70},{"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":27,"secondary_topics":33,"classification":34,"search_phrases":66,"slug":67,"view_count":68,"doi":8,"paper":8,"created_at":69},586,"Deep Learning in Farming: A Systematic Evidence-Weighted Review of Applications, Validation Gaps, and Emerging Frontiers","https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273","Foggia 大学、Barid 大学研究团队开展 PRISMA 2020 流程的系统综述,从 18731 篇文献中筛选 56 篇一手研究(42 篇带定量结果),重点覆盖精准作物管理、精准畜牧、土壤水管理、自主农业系统四类操作域。综述指出:控制条件数据集上的高性能在外部、时序或跨地点验证时常常难以保持。以一项广为引用的基准为例,病害分类准确率在保留实验室图像上 >99%,但在同类的实地采集图像上仅 31.4%。综述建议下一个阶段的农业 AI 应以野外真实性、可复现性、部署成熟度与实际效用而非孤立基准分评判。",null,"## 1. Introduction\n\n### 1.1. The Global Imperative for Agricultural Innovation\n\nAgriculture is a central pillar of the worldwide food supply, yet it faces mounting, compounding pressures: population growth, climatic instability, resource scarcity, labour constraints, and the need to lower environmental externalities. Recent international assessments stress that agrifood systems must become more productive, resilient, inclusive and sustainable, so as to meet food demand while cutting social and environmental costs [[1](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#B1-BDCC-10-00273),[2](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#B2-BDCC-10-00273),[3](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#B3-BDCC-10-00273)]. Rural depopulation and an ageing farming workforce compound these pressures further, making labour shortages especially acute for critical, time-sensitive operations such as harvesting, disease inspection, animal monitoring, and irrigation management [[3](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#B3-BDCC-10-00273),[4](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#B4-BDCC-10-00273),[5](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#B5-BDCC-10-00273)]. DL-based automation, decision support, and robotic perception respond directly to this combination of demands: raising productivity while sustaining farm operational capacity under conditions of reduced and ageing rural labour.\n\n### 1.2. Agriculture 4.0 and Deep Learning as Its Foundational Technology\n\nTo address these challenges, agriculture is undergoing a technological transformation often described as Agriculture 4.0 or Smart Agriculture, characterised by the integration of IoT, robotics, big-data analytics, and AI across the agricultural value chain [[6](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#B6-BDCC-10-00273),[7](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#B7-BDCC-10-00273),[8](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#B8-BDCC-10-00273)]. Large-scale sensor deployments across fields, machinery, drones, and satellites generate heterogeneous data streams, including imagery, soil measurements, moisture data, and weather records [[9](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#B9-BDCC-10-00273),[10](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#B10-BDCC-10-00273)], with the goal not merely of automating existing practices but of using these streams to support more precise interventions, better resource allocation, higher productivity, and a lower environmental impact [[6](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#B6-BDCC-10-00273),[7](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#B7-BDCC-10-00273)].\n\nDeep Learning (DL), a branch of machine learning grounded in multi-layer neural networks, has become the main technology enabling this shift [[6](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#B6-BDCC-10-00273),[11](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#B11-BDCC-10-00273)], owing to its capacity to learn hierarchical representations directly from large amounts of raw, often unstructured data [[12](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#B12-BDCC-10-00273),[13](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#B13-BDCC-10-00273)], precisely the kind of heterogeneous, multi-scale data that cameras, drones, satellites, animal sensors, and soil probes produce [[9](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#B9-BDCC-10-00273),[11](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#B11-BDCC-10-00273)]. Unlike classical machine-learning pipelines, which rely heavily on hand-crafted features, deep neural networks learn task-relevant representations from images, time series, and multimodal records with reduced manual feature engineering [[14](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#B14-BDCC-10-00273),[15](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#B15-BDCC-10-00273)]. This capacity is particularly valuable in farming, where the same biological target appears differently across crops, cultivars, growth stages, lighting conditions, soil backgrounds, disease severities, and management regimes.\n\nHigh benchmark performance should not, however, be read as automatic readiness for farm deployment: models trained on curated datasets routinely overfit to narrow imaging conditions or local management practices. For this reason, the present review treats DL not as a catalogue of architectures, but as a family of methods whose value rests on validation realism, transferability, interpretability, and operational feasibility, evaluated jointly with the agronomic and socio-technical conditions in which it is meant to operate.\n\n### 1.3. Scope and Structure of the Review\n\nThis review provides a systematic and analytical synthesis of deep learning in agriculture, with an emphasis on literature published between January 2020 and September 2025. The term farming is used here to delimit the review to the production-oriented part of agriculture, including crop production, livestock production, soil and water management, and autonomous field or barn operations. Broader agri-food applications such as plant breeding, biotechnology, microbiology, seed technology, industrial food processing, post-harvest packaging, retail logistics, and traceability are acknowledged as adjacent areas but are not treated as the principal object of this review, so as to preserve analytical depth and avoid merging operational farming problems with biologically and commercially distinct stages of the agri-food chain. Its objectives are to identify the main areas of progress, highlight methodological and deployment-related weaknesses, and outline the most promising directions for future research. [Section 2](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#sec2-BDCC-10-00273) presents the review methodology. [Section 3](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#sec3-BDCC-10-00273) synthesises the literature across four application domains and highlights cross-domain methodological patterns. [Section 4](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#sec4-BDCC-10-00273) discusses emergent technical directions that may help address current limitations. [Section 5](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#sec5-BDCC-10-00273) summarises the main conclusions, and [Section 6](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#sec6-BDCC-10-00273) provides a roadmap for future research and practice.\n\n### 1.4. Positioning with Respect to Existing Surveys and Systematic Reviews\n\nMultiple surveys have investigated the role of artificial intelligence and deep learning in agriculture, from varied conceptual frameworks. Ref. [[6](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#B6-BDCC-10-00273)] produced one of the earliest structured syntheses of deep learning in agriculture, concentrating on convolutional architectures and vision tasks documented through approximately 2017. Ref. [[11](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#B11-BDCC-10-00273)] addressed farm automation using both machine learning and deep learning, emphasising robotic and sensor-equipped infrastructure, while Ref. [[7](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#B7-BDCC-10-00273)] reviewed machine-learning methods for precision agriculture more broadly, positioning deep learning as one algorithmic family among others. Ref. [[16](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#B16-BDCC-10-00273)] focused specifically on deep learning and agricultural advancement, identifying constraints and potential developments. Adjacent contributions have examined complementary parts of the agri-food digitalisation workflow: Ref. [[17](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#B17-BDCC-10-00273)] surveyed machine-learning applications spanning yield modelling, disease identification, and supply-chain logistics; Ref. [[18](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#B18-BDCC-10-00273)] examined how machine learning and, more narrowly, deep learning support sustainable supply-chain performance; and specialised reviews have covered deep learning for dense visual scene interpretation [[19](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#B19-BDCC-10-00273)], satellite and airborne image classification [[9](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#B9-BDCC-10-00273),[20](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#B20-BDCC-10-00273)], and mask-based segmentation for farming imagery [[10](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#B10-BDCC-10-00273)]. Recently, there have also been a number of survey papers that emerged: Ref. [[21](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#B21-BDCC-10-00273)] provided a survey of 129 papers in five domains of DL application in agriculture but without the realism constraint of deployment; Ref. [[22](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#B22-BDCC-10-00273)] conducted a systematic review of machine learning and computer vision in precision agriculture focusing on innovative methods; Ref. [[23](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#B23-BDCC-10-00273)] employed a bibliometric and keyword network approach to deep learning in crop monitoring; and Ref. [[24](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#B24-BDCC-10-00273)] provided a systematic review of deep learning for lettuce farming alone. The table below contrasts this review with the aforementioned reviews.\n\nThe present review departs from these earlier contributions along four lines: it centres on the 2020–September 2025 window, during which agricultural DL shifted from mostly CNN-based classification toward object detection, segmentation, transformer architectures, self-supervised learning, edge inference, federated learning, and hybrid PBM–DL systems; it follows a PRISMA 2020-oriented protocol with explicit search blocks, screening counts, and a common critical-appraisal framework ([Section 2](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#sec2-BDCC-10-00273)); it tightens the object of analysis onto farming specifically, rather than the whole agri-food chain; and it weights the evidence explicitly, reading reported model performance through dataset representativeness, validation protocol, benchmarking transparency, deployment realism, and reproducibility, so as to separate studies showing algorithmic promise from studies giving firmer evidence of field relevance.\n\n### 1.5. Main Contributions of This Review\n\nThis review makes four main contributions. First, it applies a transparent review methodology that allows readers to assess the strength of evidence of individual studies rather than treating all reported results as equivalent. Second, it synthesises the literature across four agricultural domains using common comparative dimensions. Third, it explicitly addresses the laboratory-to-field gap together with benchmarking, socio-economic, and governance challenges. Fourth, it provides a near-term research agenda, explicitly ordered by the strength of evidence currently available for each direction ([Section 5.2](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#sec5dot2-BDCC-10-00273)), focused on realistic benchmarks, multimodal and self-supervised learning, system-level evaluation, and techno-economic assessment.\n\n### 1.6. Analytical Framework and Originality of the Review\n\nThe originality of this review does not rest on a claim that agricultural deep learning has remained unreviewed. [Section 1.4](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#sec1dot4-BDCC-10-00273) and [Table 1](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#table_body_display_BDCC-10-00273-t001) show clearly that it has not. Its contribution is rather to re-read the recent literature through an evidence-weighted and deployment-oriented framework, interpreting the empirical studies discussed along five analytical dimensions. Data realism points to the degree to which the dataset mirrors field variability.Validation strength points to the presence of external, temporal or cross-site testing. Model–task suitability asks whether the chosen architecture fits the structure of the agricultural problem. Deployment readiness looks for evidence of latency, edge feasibility, robustness and system integration. Practical value weighs agronomic interpretability, economic feasibility and usefulness to farmers or agricultural professionals. These five dimensions are not merely descriptive: [Section 2](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#sec2-BDCC-10-00273) operationalises them directly as the risk-of-bias rubric in [Table 2](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#table_body_display_BDCC-10-00273-t002), so the same criteria introduced here are the ones actually applied to every included study. Through this framework, the review steps past a descriptive list of applications, separating studies showing narrow benchmark success from those offering sturdier evidence of operational relevance, and judging the field’s recent growth in CNN-based vision, object detection, segmentation, transformers, self-supervised learning, federated learning, edge AI, and process-informed hybrid models against one practical question: are the reported systems reproducible, robust, interpretable, and economically plausible inside real farming environments?\n\n**Table 1.** Positioning of the present review relative to representative prior surveys of deep learning in agriculture.\n\n**Table 2.** Risk-of-bias\u002Fcritical-appraisal rubric applied to the 56 included studies, operationalising the five analytical dimensions of [Section 1.6](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#sec1dot6-BDCC-10-00273). Authors’ synthesis.\n\n## 2. Materials and Methods\n\nA systematic literature search, structured around a PRISMA 2020-oriented protocol [[25](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#B25-BDCC-10-00273)], was used to identify recent empirical studies and relevant contextual reviews concerning deep learning in farming. Searches covered Scopus, Web of Science, and IEEE Xplore; Google Scholar supported backward and forward citation searching. The principal publication window was 1 January 2020 to 30 September 2025. Earlier studies were consulted only when they provided foundational technical or historical context and were not treated as part of the recent empirical evidence base. Farming and agricultural-production terms were combined with descriptors of deep learning methods. Production-stage terms such as “sowing”, “harvesting”, “robotic harvesting”, “poultry”, “precision livestock farming”, and “animal welfare” were included to reduce vocabulary bias. A representative Scopus query is:\n\n> TITLE (“deep learning” OR “CNN” OR “RNN” OR “LSTM” OR “Transformer” OR “GNN” OR “self-supervised learning” OR “neural network”) AND TITLE-ABS-KEY (“agriculture” OR “crop” OR “farm” OR “livestock” OR “irrigation” OR “phenotyping” OR “robotics”) AND PUBYEAR > 2019 AND PUBYEAR \u003C 2026 AND (LIMIT-TO (SUBJAREA, “COMP”) OR LIMIT-TO (SUBJAREA, “ENGI”) OR LIMIT-TO (SUBJAREA, “AGRI”)) AND (LIMIT-TO (DOCTYPE, “ar”) OR LIMIT-TO (DOCTYPE, “cp”))\n\nThe query was adapted to the syntax and indexing fields of each database while preserving three conceptual blocks: (i) deep-learning terminology, (ii) farming and agricultural-production terminology, and (iii) task-specific terminology. Google Scholar was not used as a primary database because its ranking and coverage are difficult to reproduce; it was instead used only for backward and forward snowballing from already-eligible papers. The final search update was performed in September 2025, and online-first material was considered only when its peer-reviewed publication status could be verified by that date. The completed PRISMA 2020 Checklist is provided in the [Supplementary Materials](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#app1-BDCC-10-00273). We intentionally opted for using traditional bibliographic databases rather than a scientific search or summarisation based on LLMs (Elicit, Consensus, and others). There were three main reasons behind our choice. First, reproducibility: Boolean search in Scopus, Web of Science, and IEEE Xplore can be repeated independently by any other researcher and will produce consistent results, while an LLM search is highly dependent on specific wording of prompts, model version used, and heuristic ranking\u002Fsummarisation algorithm of the tool itself, which is all non-reproducible. Second, transparent coverage: bibliographic databases provide an explicit count of included\u002Fexcluded articles at each screening step ([Section 2.2](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#sec2dot2-BDCC-10-00273)), and that is what provides audibility of the PRISMA flowchart in [Figure 1](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#fig_body_display_BDCC-10-00273-f001); no general-purpose search tool powered by LLM can provide the same kind of transparent accounting of the search process. Third, reliability: LLM-driven summarisation of scientific text comes with known risk of generating false citations or misattribution of findings - exactly the kind of mistake that a critical evidence synthesis should avoid. However, these tools are a truly valuable addition to the traditional screening process, and we will further discuss them, as well as the application of large language models in agricultural decision-making in [Section 4.4](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#sec4dot4-BDCC-10-00273) and [Section 6](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#sec6-BDCC-10-00273).\n\n**Figure 1.** PRISMA 2020 flow diagram summarising the identification, screening, eligibility assessment, and inclusion of studies (2020–2025) [[25](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#B25-BDCC-10-00273)]. Diagram authored natively in TikZ by the authors from the underlying screening counts, so that it can be revised as plain text rather than as a raster image.\n\nCandidate publications were checked by title, abstract and, where necessary, full text. Bibliographic metadata were verified against the cited publication rather than inferred from citation-key names. For the evidence-weighted synthesis, the extraction fields were application domain, task, sensing modality, dataset type, architecture family, comparison models, evaluation metrics, validation protocol, code or data availability, deployment status, and stated limitations.\n\n### 2.1. Inclusion and Exclusion Criteria\n\nThe core empirical evidence comprised English-language, peer-reviewed journal articles and indexed conference papers published from 2020 through September 2025 that presented and evaluated a deep-learning model for a production-oriented agricultural task. Eligible tasks included crop disease and weed detection, yield or fruit estimation, livestock identification, health and welfare monitoring, soil-property or crop-water assessment, and robotic perception or control. Empirical papers had to specify a task, a dataset or acquisition setting, a model family, and quantitative or operational performance evidence.\n\nReview articles, reports and studies published before 2020 were used only for contextual framing and were not represented as recent empirical evidence. Papers were not used as task-level evidence when they evaluated only non-deep-learning methods, concerned a non-agricultural application, or addressed supply-chain security, industrial processing, breeding or traceability without a direct connection to farming-stage sensing, prediction or action. This distinction is important because contextual sources and primary empirical studies serve different purposes in a critical review. The reasons for full-text exclusions were categorised into three categories, as depicted in [Figure 1](https:\u002F\u002Fwww.mdpi.com\u002F2504-2289\u002F10\u002F8\u002F273#fig_body_display_BDCC-10-00273-f001). Non-eligible content (\n\n$n = 1250$\n) includes records that did not actually involve evaluating a deep-learning model upon further checking of the full text (such as opinion pieces, theoretical works only, or classical machine-learning-based papers classified due to erroneous search terms), those that discussed topics not related to agriculture even though they matched the query, or those that were about breeding, processing, or traceability but did not have any farming stage sensing\u002Fprediction\u002Faction application. Incomplete data (\n\n$n = 254$\n) includes records where the dataset information, model information, or results were not sufficiently available to be extracted. Pu","Big Data and Cognitive Computing 2026, 10(8): 273","2026-08-14T00:00:00Z","论文",10,false,83,{"impact":17,"substance":18,"depth":19,"authority":20,"freshness":21,"relevant":22,"comment":23},18,22,20,14,9,1,"系统综述深度学习在农业中的应用，方法严谨，覆盖全面，对智慧农业研究有较高参考价值。",[25],{"name":10,"url":6},[],{"code":28,"slug":29,"name":30,"path":31,"parent_code":32},"I2","agricultural-ai","农业AI与大模型","\u002Ftopic\u002Fdigital-agriculture\u002Fagricultural-ai","I",[],{"status":35,"locked":14,"version":36,"method":37,"evidence":38},"classified","2026-10-22.2","rules",{"candidates":39,"agricultural_evidence":42},[40,47,54,60],{"code":28,"score":41,"supported":42,"title":43,"abstract":45,"tags":46},8,true,[44],"deep learning",[],[],{"code":48,"score":49,"supported":42,"title":50,"abstract":51,"tags":53},"C",3,[],[52],"畜牧",[],{"code":55,"score":49,"supported":42,"title":56,"abstract":57,"tags":59},"C6",[],[58],"精准畜牧",[],{"code":61,"score":49,"supported":42,"title":62,"abstract":63,"tags":65},"H2",[],[64],"土壤",[],[],"DeepLearninginFarming:ASystematicEvidenc-586",2,"2026-08-15T23:54:33.850636Z",{"total":71,"page":22,"page_size":71,"items":72},6,[73,143,207,259,311,372],{"id":74,"title":75,"url":76,"summary":77,"summary_zh":78,"content":8,"source_name":79,"source_url":76,"published_at":80,"category":12,"cover_url":8,"hotness":13,"is_selected":14,"score":81,"score_detail":82,"sources":86,"tags":88,"primary_topic":94,"secondary_topics":100,"classification":102,"search_phrases":120,"slug":123,"view_count":124,"doi":125,"paper":126,"created_at":142},5592,"Evaluation of Concept-Rank Collapse Audits for Computer Vision Agriculture Disease Classifiers Under Low-Label Regimes: An Interpretability Capacity Bound for Smallholder Fungicide Selectivity","https:\u002F\u002Fdoi.org\u002F10.5281\u002Fzenodo.23545193","Concept-based interpretability methods for plant-disease classifiers are typically evaluated by whether their concept activations correlate with human-annotated labels, but this criterion is silent on a prior question: whether the classifier's internal representation retains enough concept-rank structure for any audit to be informative. the present analysis formalises that prior question as an interpretability capacity bound. A classifier is modelled as a concept-rank operator whose singular values govern how much concept-level signal survives the projection from pixel space to class logits, and a collapse audit is defined as the ratio of retained concept-rank energy to total concept energy. Under a low-label regime, where concept annotations are scarce and the audit must be estimated from few labelled examples, the bound predicts that audit reliability degrades as a function of the concept-rank spectrum and the label budget. The analysis is a theoretical and methodological contribution: it derives the bound, states the conditions under which a collapse audit is identifiable, and characterises the operating characteristics of the audit as a decision procedure under a declared variance structure (between-cluster SD = 6.0, residual SD = 4.5, intraclass correlation = 0.64). Power was estimated across 2000 design cells for cluster counts of 8, 16, 24 and 32 and effect sizes of 3, 5 and 8 percentage points; 52 clusters were required for 5 percentage points at 80% power. The bound reframes concept audits as capacity tests rather than fidelity tests, with consequences for how smallholder fungicide-selectivity decisions should be supported by interpretable disease classifiers.","面向植物病害分类器的基于概念的可解释性方法，通常以其概念激活是否与人工标注标签相关作为评价标准，但该标准对一个更先前的问题保持沉默：分类器的内部表征是否保留了足够的概念秩结构，从而使任何审计具有信息量。本分析将该先前问题形式化为一个可解释性容量上界。分类器被建模为概念秩算子，其奇异值支配着从像素空间到类别logits的投影中概念级信号得以保留的程度，并将坍缩审计定义为保留的概念秩能量与总概念能量之比。在低标签情形下，概念标注稀缺且审计必须由少量标注样本估计，该上界预测审计可靠性会随概念秩谱和标签预算的函数而退化。本分析是一项理论和方法论贡献：推导出该上界，陈述坍缩审计可识别的条件，并在声明的方差结构（簇间SD = 6.0，残差SD = 4.5，组内相关系数 = 0.64）下刻画审计作为决策程序的操作特征。在2000个设计单元中，针对簇数为8、16、24和32以及效应量为3、5和8个百分点的情形估计了功效；在80%功效下达到5个百分点需要52个簇。该上界将概念审计重新框定为容量检验而非保真度检验，并对可解释病害分类器应如何支持小农杀菌剂选择性决策具有启示意义。","Zenodo (CERN European Organization for Nuclear Research)","2026-10-21T00:00:00Z",69,{"impact":41,"substance":83,"depth":17,"authority":84,"freshness":13,"relevant":22,"comment":85},21,12,"理论方法论文，提出可解释性容量上界并给出审计功效分析，对智慧农业病害识别可信度有参考价值，但属小众方法学研究、缺乏田间实证，暂不宜进入每日精选。",[87],{"name":79,"url":76},[89,90,91,92,93],"智慧农业","农业人工智能","小农户","植物病害识别","可解释性",{"code":95,"slug":96,"name":97,"path":98,"parent_code":99},"G2","farm-services","社会化服务与规模经营","\u002Ftopic\u002Frural-services\u002Ffarm-services","G",[101],{"code":28,"slug":29,"name":30,"path":31,"parent_code":32},{"status":35,"locked":14,"version":36,"method":37,"evidence":103},{"candidates":104,"agricultural_evidence":42},[105,110,115],{"code":95,"score":84,"supported":42,"title":106,"abstract":108,"tags":109},[107],"smallholder",[107],[91],{"code":28,"score":13,"supported":42,"title":111,"abstract":113,"tags":114},[112],"computer vision",[],[90,92],{"code":116,"score":22,"supported":14,"title":117,"abstract":118,"tags":119},"I1",[],[],[89],[121,122],"农业人工智能 植物病害识别 可解释性 智慧农业","农业人工智能 植物病害识别","农业人工智能植物病害识别可解释性智慧农业-5592",0,"10.5281\u002Fzenodo.23545193",{"doi":125,"openalex_id":127,"authors":128,"venue":79,"cited_by_count":124,"oa_url":76,"card":135,"direction":139,"ingested_from":141},"W7223137323",[129,131,133],{"name":130,"orcid":8},"Akshay Ramgoolam",{"name":132,"orcid":8},"Priya Bundhoo Appadoo",{"name":134,"orcid":8},"Jean-Noël LEspérance",{"tldr":136,"method":137,"finding":138,"direction":139,"opportunity":140},"提出可解释性容量上界，评估低标签下概念秩崩溃审计的可靠性。","将分类器建模为概念秩算子，推导容量上界并做2000格功效模拟。","审计可靠性随概念秩谱与标签预算下降；5个百分点效应需52个聚类达80%功效。","农业人工智能与决策模型","可将容量上界用于小农户杀菌剂选择性决策的可解释模型审计与标签预算设计。","openalex","2026-10-22T11:21:10.425290Z",{"id":144,"title":145,"url":146,"summary":147,"summary_zh":148,"content":8,"source_name":79,"source_url":146,"published_at":80,"category":12,"cover_url":8,"hotness":13,"is_selected":14,"score":81,"score_detail":149,"sources":151,"tags":153,"primary_topic":157,"secondary_topics":163,"classification":165,"search_phrases":189,"slug":192,"view_count":124,"doi":193,"paper":194,"created_at":206},5591,"Hierarchical Conformal Risk Control for Artificial Intelligence Agriculture Irrigation Scheduling Under Unmodeled Infiltration: A Computer Safety-Guarantee Evaluation","https:\u002F\u002Fdoi.org\u002F10.5281\u002Fzenodo.23543665","Irrigation scheduling algorithms that rely on soil-moisture forecasts are vulnerable to unmodeled infiltration, the process by which water bypasses the root zone through preferential pathways that resistance-based soil models cannot represent. The framework is evaluated through a replication-based design study that characterises the operating characteristics of the before-after-control-impact estimator used to audit scheduler performance, under a declared variance structure with between-cluster standard deviation 6.0, residual standard deviation 4.5 and intraclass correlation 0.64. Power was estimated across 2000 design cells spanning cluster counts of 8, 16, 24 and 32 and effect sizes of 3, 5 and 8 percentage points. The analysis indicates that 54 clusters are required to detect a 5-percentage-point effect at 80% power, and that the minimum detectable effect declines steeply as cluster count rises from 8 to 32. The contribution is a formal safety guarantee for irrigation decision support that remains valid when the infiltration model is wrong, together with an audit design that makes the guarantee empirically checkable.","依赖土壤湿度预报的灌溉调度算法容易受到未建模入渗的影响，即水分通过优先路径绕过根区的过程，而基于阻力的土壤模型无法表征这一过程。该框架通过一项基于重复的设计研究进行评估，该研究刻画了用于审计调度器性能的前后对照影响估计量的运行特性，其所依据的声明方差结构为：组间标准差6.0、残差标准差4.5、组内相关系数0.64。功效估计覆盖2000个设计单元，涵盖聚类数8、16、24和32以及效应量3、5和8个百分点。分析表明，在80%功效下检测5个百分点效应需要54个聚类，且最小可检测效应随聚类数从8增至32而急剧下降。本文的贡献在于为灌溉决策支持提供了一种正式的安全保证，该保证在入渗模型错误时仍然有效，同时提出了一种使该保证可经实证检验的审计设计。",{"impact":41,"substance":18,"depth":17,"authority":84,"freshness":21,"relevant":22,"comment":150},"该研究提出在入渗模型失配条件下仍有效的灌溉决策安全保证框架，并给出可审计的试验设计，方法新颖、结论可靠，但属细分方法学研究，产业影响有限，适合作为专业精选而非头条。",[152],{"name":79,"url":146},[89,90,154,155,156],"灌溉决策","安全保证","试验设计",{"code":158,"slug":159,"name":160,"path":161,"parent_code":162},"H3","irrigation","水资源与灌溉","\u002Ftopic\u002Fenvironment\u002Firrigation","H",[164],{"code":28,"slug":29,"name":30,"path":31,"parent_code":32},{"status":35,"locked":14,"version":36,"method":37,"evidence":166},{"candidates":167,"agricultural_evidence":42},[168,174,180,185],{"code":158,"score":169,"supported":42,"title":170,"abstract":171,"tags":173},15,[159],[172,159],"灌溉",[154],{"code":28,"score":84,"supported":42,"title":175,"abstract":177,"tags":179},[176],"artificial intelligence",[178],"decision support",[90],{"code":61,"score":71,"supported":42,"title":181,"abstract":182,"tags":184},[],[64,183],"soil",[],{"code":116,"score":22,"supported":14,"title":186,"abstract":187,"tags":188},[],[],[89],[190,191],"灌溉调度 安全保证","农业人工智能 安全保证 智慧农业 灌溉决策","灌溉调度安全保证-5591","10.5281\u002Fzenodo.23543665",{"doi":193,"openalex_id":195,"authors":196,"venue":79,"cited_by_count":124,"oa_url":146,"card":201,"direction":139,"ingested_from":141},"W7223131205",[197,199],{"name":198,"orcid":8},"Nassima Benkhelifa",{"name":200,"orcid":8},"Yacine Aït Ahmed",{"tldr":202,"method":203,"finding":204,"direction":139,"opportunity":205},"提出分层保形风险控制，为未建模入渗下的灌溉调度提供安全保证。","分层保形风险控制与基于重复设计的BACI审计，2000个设计单元功效分析。","检测5个百分点效应需54个集群，最小可检测效应随集群数增加陡降。","可将保形风险控制扩展到其他未建模扰动下的农业决策安全保证与审计设计。","2026-10-22T11:21:10.283561Z",{"id":208,"title":209,"url":210,"summary":211,"summary_zh":212,"content":8,"source_name":79,"source_url":210,"published_at":80,"category":12,"cover_url":8,"hotness":13,"is_selected":14,"score":213,"score_detail":214,"sources":216,"tags":218,"primary_topic":221,"secondary_topics":222,"classification":224,"search_phrases":239,"slug":242,"view_count":124,"doi":243,"paper":244,"created_at":258},5590,"Evaluation of Concept-Probe Temporal-Stability Audits for Computer Vision Agriculture Disease Classifiers Under Seasonal Phenology Drift: An Interpretability Decay Bound for Smallholder Fungicide Timing","https:\u002F\u002Fdoi.org\u002F10.5281\u002Fzenodo.23545234","Concept-probe audits are increasingly used to certify that a crop-disease classifier's internal representations track the pathological concepts a smallholder would act on, but a probe certified at one point in the season may silently decouple from those concepts as phenology shifts the visual statistics of the field. the present analysis treats that decoupling as a formal object. It defines a temporal-stability audit as a sequence of concept-probe accuracies measured across defined phenological windows, and it derives an interpretability decay bound: the rate at which probe accuracy can fall before the fungicide-timing decision it is meant to support becomes unreliable. The bound is expressed as a function of the probe's initial margin, the drift in concept prevalence between windows, and the classifier's sensitivity to the drifted features, and it yields a certified audit interval rather than a single-point certificate. The paper then characterises the operating characteristics of the audit design under a declared variance structure, using a replication-based evaluation of the before-after-control-impact estimator with between-cluster SD = 6.0, residual SD = 4.5 and intraclass correlation = 0.64. Power was estimated across 2000 design cells over cluster counts of 8, 16, 24 and 32 and effect sizes of 3, 5 and 8 percentage points; 54 clusters were required for 5 percentage points at 80% power. The contribution is a decision-theoretic account of when a concept-probe certificate expires, and a design rule for scheduling re-audits across the season.","概念探针审计（concept-probe audit）日益被用于验证作物病害分类器的内部表征是否追踪小农户会据以采取行动的病理学概念，但在生长季某一时点获得认证的探针，可能随着物候变化改变田间的视觉统计特征而悄然与这些概念脱钩。本分析将这种脱钩视为一个形式化对象。它将时间稳定性审计（temporal-stability audit）定义为在既定物候窗口内测量的一系列概念探针准确率，并推导出一个可解释性衰减界限（interpretability decay bound）：即探针准确率在其所支持的杀菌剂施用时机决策变得不可靠之前所能下降的速率。该界限表示为探针初始裕度、窗口间概念流行率漂移以及分类器对漂移特征敏感度的函数，并给出一个经认证的审计区间而非单点认证。随后，本文在声明的方差结构下刻画该审计设计的运行特征，采用基于重复的前后对照影响估计量评估，其中组间标准差 = 6.0、残差标准差 = 4.5、组内相关系数 = 0.64。功效在 2000 个设计单元上估计，聚类数取 8、16、24 和 32，效应量取 3、5 和 8 个百分点；在 80% 功效下达到 5 个百分点需要 54 个聚类。本文的贡献在于从决策论角度阐明概念探针认证何时失效，并给出在整个生长季内安排重新审计的设计规则。",70,{"impact":41,"substance":18,"depth":17,"authority":84,"freshness":13,"relevant":22,"comment":215},"该论文提出概念探针审计的时序稳定性框架与可解释性衰减界，方法新颖、统计设计扎实，但属小众方法论研究，产业影响有限，适合作为专业精选而非头条。",[217],{"name":79,"url":210},[89,90,91,219,220],"作物病害识别","模型审计",{"code":95,"slug":96,"name":97,"path":98,"parent_code":99},[223],{"code":28,"slug":29,"name":30,"path":31,"parent_code":32},{"status":35,"locked":14,"version":36,"method":37,"evidence":225},{"candidates":226,"agricultural_evidence":42},[227,231,235],{"code":95,"score":169,"supported":42,"title":228,"abstract":229,"tags":230},[107],[91,107],[91],{"code":28,"score":13,"supported":42,"title":232,"abstract":233,"tags":234},[112],[],[219,90],{"code":116,"score":22,"supported":14,"title":236,"abstract":237,"tags":238},[],[],[89],[240,241],"概念探针 时序稳定性 作物病害分类器","杀菌剂 施药时机 可解释性衰减","概念探针时序稳定性作物病害分类器-5590","10.5281\u002Fzenodo.23545234",{"doi":243,"openalex_id":245,"authors":246,"venue":79,"cited_by_count":124,"oa_url":210,"card":253,"direction":139,"ingested_from":141},"W7223088030",[247,249,251],{"name":248,"orcid":8},"Abdi-Basid Idriss Farah",{"name":250,"orcid":8},"Amina Abdillahi Guedi",{"name":252,"orcid":8},"Moussa Omar Youssouf",{"tldr":254,"method":255,"finding":256,"direction":139,"opportunity":257},"提出概念探针时间稳定性审计与可解释性衰减界，判断何时需重新审计。","定义跨物候窗口探针精度序列，推导衰减界，用BACI复制评估统计功效。","5个百分点效应需54个聚类达80%功效，审计证书会随物候漂移过期。","可将衰减界与再审计调度规则嵌入小农户杀菌剂时机决策支持系统，并验证跨作物泛化。","2026-10-22T11:21:08.978348Z",{"id":260,"title":261,"url":262,"summary":263,"summary_zh":264,"content":8,"source_name":79,"source_url":262,"published_at":80,"category":12,"cover_url":8,"hotness":13,"is_selected":14,"score":265,"score_detail":266,"sources":269,"tags":271,"primary_topic":274,"secondary_topics":275,"classification":276,"search_phrases":295,"slug":298,"view_count":124,"doi":299,"paper":300,"created_at":310},5589,"Conformal-Risk Evaluation for Artificial Intelligence Agriculture Spray-Nozzle Fault Detection Under Label-Delay Feedback: A Computer Coverage Audit","https:\u002F\u002Fdoi.org\u002F10.5281\u002Fzenodo.23544107","Spray-nozzle fault detection in precision agriculture is a classification problem in which the label that confirms a fault arrives after the spray pass, so the detector must act on unlabelled evidence and its error rate must be certified before the label exists. Conformal prediction offers distribution-free coverage guarantees, but the guarantee is stated for exchangeable calibration and test data, an assumption that label-delayed feedback violates because the calibration set is always older than the deployment set. the present analysis formalises label-delay feedback as a distribution shift in the conformity score and derives a coverage bound for split-conformal fault detectors under that shift. The bound separates the nominal miscoverage rate into an exchangeable component and a delay-induced component that scales with the total variation distance between the calibration and deployment score distributions. A replication-based evaluation of the ['Intercept', 'exposure', 'group', 'severity'] estimator was conducted: 600 data sets of N = 500 were drawn from a declared data-generating process (intercept 12.0, exposure 0.42, group -3.1, severity 1.35, error SD 4.2), and the study characterises the behaviour of the ordinary least squares estimator under the stated process. The estimator recovered the declared coefficients with estimated bias 0.01122 for the intercept and -0.0002 for exposure, root mean squared error 0.9217 and 0.01614 respectively, and empirical coverage of the nominal 95% intervals of 1.0 for both parameters across 600 replications. The coverage audit shows that the delay-induced term, not the exchangeable term, dominates the risk of a conformal spray-nozzle detector, and that the audit procedure can certify a detector before field labels return.","在精准农业中，喷雾喷嘴故障检测是一个分类问题，其中确认故障的标签在喷雾作业结束后才到达，因此检测器必须基于未标记的证据采取行动，并且其错误率必须在标签存在之前得到认证。保形预测（conformal prediction）提供了无分布覆盖保证，但该保证是针对可交换的校准数据和测试数据而言的，而标签延迟反馈违反了这一假设，因为校准集总是比部署集更旧。本分析将标签延迟反馈形式化为一致性分数中的分布偏移，并推导了在该偏移下分裂保形故障检测器的覆盖界。该界将名义错误覆盖 rate 分解为可交换分量和延迟诱导分量，后者随校准与部署分数分布之间的全变差距离而缩放。对 ['Intercept', 'exposure', 'group', 'severity'] 估计器进行了基于重复的评估：从声明的数据生成过程（截距 12.0，暴露 0.42，分组 -3.1，严重程度 1.35，误差标准差 4.2）中抽取了 600 个 N = 500 的数据集，研究刻画了普通最小二乘估计器在所述过程下的行为。该估计器恢复了声明的系数，截距的估计偏差为 0.01122，暴露为 -0.0002，均方根误差分别为 0.9217 和 0.01614，两个参数在 600 次重复中名义 95% 区间的经验覆盖率为 1.0。覆盖审计表明，延迟诱导项而非可交换项主导了保形喷雾喷嘴检测器的风险，并且该审计程序可以在田间标签返回之前对检测器进行认证。",62,{"impact":41,"substance":17,"depth":267,"authority":21,"freshness":13,"relevant":22,"comment":268},17,"论文提出标签延迟下共形预测覆盖界，方法有新意但属预印本、应用面窄，可作为专业补充而非每日精选头条。",[270],{"name":79,"url":262},[89,90,272,273],"精准施药","病虫害监测",{"code":28,"slug":29,"name":30,"path":31,"parent_code":32},[],{"status":35,"locked":14,"version":36,"method":37,"evidence":277},{"candidates":278,"agricultural_evidence":42},[279,283,290],{"code":28,"score":21,"supported":42,"title":280,"abstract":281,"tags":282},[176],[],[90],{"code":116,"score":284,"supported":42,"title":285,"abstract":286,"tags":289},7,[],[287,288],"精准农业","precision agriculture",[89],{"code":291,"score":22,"supported":14,"title":292,"abstract":293,"tags":294},"F4",[],[],[272],[296,297],"农业人工智能 病虫害监测 智慧农业 精准施药","农业人工智能 病虫害监测","农业人工智能病虫害监测智慧农业精准施药-5589","10.5281\u002Fzenodo.23544107",{"doi":299,"openalex_id":301,"authors":302,"venue":79,"cited_by_count":124,"oa_url":262,"card":305,"direction":139,"ingested_from":141},"W7223116087",[303],{"name":304,"orcid":8},"Achieng' Atim Okello",{"tldr":306,"method":307,"finding":308,"direction":139,"opportunity":309},"将标签延迟反馈建模为共形分数分布偏移，推导覆盖率界并审计喷雾喷嘴故障检测器。","分裂共形预测、总变差距离、600次重复模拟（N=500）评估OLS估计。","延迟诱导项而非可交换项主导检测器风险，审计可在田间标签返回前认证检测器。","可将该覆盖率审计框架扩展到其他标签延迟的农业传感器故障检测，并研究延迟分布估计误差对认证的影响。","2026-10-22T11:21:02.845718Z",{"id":312,"title":313,"url":314,"summary":315,"summary_zh":316,"content":8,"source_name":79,"source_url":314,"published_at":80,"category":12,"cover_url":8,"hotness":13,"is_selected":14,"score":317,"score_detail":318,"sources":320,"tags":322,"primary_topic":326,"secondary_topics":327,"classification":328,"search_phrases":355,"slug":358,"view_count":124,"doi":359,"paper":360,"created_at":371},5588,"Certified Temporal Fusion of Artificial Intelligence Climate Analytics for Crop Failure Early Warning Under Sparse Agricultural Sensors: Error Bounds and Famine Relief Targeting","https:\u002F\u002Fdoi.org\u002F10.5281\u002Fzenodo.23545489","Sparse agricultural sensor networks across sub-Saharan Africa leave crop-failure early-warning systems with too little ground observation to support reliable famine relief targeting. The framework treats sensor sparsity as a missing-data problem with a known observation operator and propagates climate-analytics uncertainty through a temporal Kalman-Buys filter whose gain is constrained by a certified risk envelope. Using NASA POWER monthly reanalysis for Accra, Ghana (2000-2023, 312 observations), the analysis establishes a mean temperature of 26.521 C and a warming trend of 0.1622 C per decade, which calibrates the climate-analytics prior. The derived bounds show that relief-targeting error scales with the square root of sensor density and with the Lipschitz constant of the crop-response function, yielding an explicit sensor-placement rule for a target misclassification rate. The contribution is a certified, assumption-explicit bridge between climate analytics and famine relief targeting that states the conditions under which early warning is trustworthy.","撒哈拉以南非洲稀疏的农业传感器网络使得作物歉收预警系统缺乏足够的地面观测数据，难以支撑可靠的饥荒救援目标定位。该框架将传感器稀疏性视为具有已知观测算子的缺失数据问题，并通过时间Kalman-Buys滤波器传播气候分析不确定性，该滤波器的增益受经认证的风险包络约束。利用NASA POWER对加纳阿克拉的月度再分析数据（2000—2023年，312个观测值），分析确定了平均气温为26.521°C、每十年0.1622°C的增暖趋势，据此校准了气候分析先验。推导出的界限表明，救援目标定位误差随传感器密度的平方根以及作物响应函数的Lipschitz常数缩放，从而为目标误分类率给出了明确的传感器布设规则。该研究的贡献在于构建了一个经认证的、假设明确的气候分析与饥荒救援目标定位之间的桥梁，阐明了早期预警可信赖的条件。",74,{"impact":84,"substance":18,"depth":17,"authority":84,"freshness":13,"relevant":22,"comment":319},"该研究提出一种带误差界认证的时序融合方法，用于稀疏传感器条件下的作物歉收预警与饥荒救援瞄准，方法新颖、结论明确，但案例局限于加纳阿克拉，产业影响有限，适合作为技术前沿条目收录。",[321],{"name":79,"url":314},[89,90,323,324,325],"粮食安全","遥感","气候分析",{"code":28,"slug":29,"name":30,"path":31,"parent_code":32},[],{"status":35,"locked":14,"version":36,"method":37,"evidence":329},{"candidates":330,"agricultural_evidence":42},[331,335,341,346,350],{"code":28,"score":21,"supported":42,"title":332,"abstract":333,"tags":334},[176],[],[90],{"code":336,"score":49,"supported":42,"title":337,"abstract":338,"tags":340},"I5",[],[339],"农业传感",[],{"code":342,"score":22,"supported":14,"title":343,"abstract":344,"tags":345},"A8",[],[],[323],{"code":116,"score":22,"supported":14,"title":347,"abstract":348,"tags":349},[],[],[89],{"code":351,"score":22,"supported":14,"title":352,"abstract":353,"tags":354},"I3",[],[],[324],[356,357],"NASA POWER 加纳 气候分析","作物歉收 早期预警 传感器","NASAPOWER加纳气候分析-5588","10.5281\u002Fzenodo.23545489",{"doi":359,"openalex_id":361,"authors":362,"venue":79,"cited_by_count":124,"oa_url":314,"card":365,"direction":139,"ingested_from":141},"W7223111766",[363],{"name":364,"orcid":8},"Fatima-Zahra El Khattabi",{"tldr":366,"method":367,"finding":368,"direction":369,"opportunity":370},"提出带误差界的时间融合框架，在稀疏传感器下用气候分析预警作物歉收。","卡尔曼-布西滤波+NASA POWER再分析数据，风险包络约束增益。","救济瞄准误差随传感器密度平方根和作物响应Lipschitz常数缩放，给出布点规则。","农业遥感与作物表型","可将该认证误差界框架扩展到多作物多区域，结合卫星遥感弥补地面传感器稀疏。","2026-10-22T11:21:02.682497Z",{"id":373,"title":374,"url":375,"summary":376,"summary_zh":377,"content":8,"source_name":79,"source_url":375,"published_at":80,"category":12,"cover_url":8,"hotness":13,"is_selected":14,"score":378,"score_detail":379,"sources":381,"tags":383,"primary_topic":387,"secondary_topics":388,"classification":389,"search_phrases":405,"slug":408,"view_count":124,"doi":409,"paper":410,"created_at":424},5587,"Calibration-Shift Inference for Computer Vision Agriculture Yield Models Under Retrofit Sensor Replacement: An Artificial Intelligence Post-Deployment Recalibration Guarantee","https:\u002F\u002Fdoi.org\u002F10.5281\u002Fzenodo.23544195","Computer vision yield models deployed in agricultural fields are routinely paired with retrofit sensor packages that replace or supplement the original imaging hardware, and the resulting calibration shift can silently invalidate the model's predictive guarantees. The analysis characterises the operating characteristics of the design under a declared variance structure with between-cluster standard deviation 6.0, residual standard deviation 4.5 and intraclass correlation 0.64, using 2000 replications per design cell across cluster counts of 8, 16, 24 and 32 and effect sizes of 3, 5 and 8 percentage points. The design attains 80% power at a 5 percentage point effect with 54 clusters. The paper states a formal guarantee: under stated assumptions on the stability of the calibration map, the before-after-control-impact estimator yields a consistent and asymptotically normal estimate of the calibration shift, and the associated confidence interval covers the true shift at the nominal rate. The contribution is a design-theoretic account of how retrofit-induced calibration shift can be detected and bounded without observational data, together with the sample-size requirements that follow.","部署在农田中的计算机视觉产量模型通常与改装传感器组件配合使用，这些组件用于替换或补充原始成像硬件，由此产生的校准偏移可能悄然使模型的预测保证失效。该分析在声明的方差结构下刻画了设计的工作特性，其中组间标准差为6.0、残差标准差为4.5、组内相关系数为0.64，每个设计单元采用2000次重复，聚类数分别为8、16、24和32，效应量为3、5和8个百分点。该设计在54个聚类、5个百分点效应下达到80%的检验力。论文给出了一项形式化保证：在关于校准映射稳定性的既定假设下，前后对照影响（before-after-control-impact）估计量给出校准偏移的一致且渐近正态的估计，且相关置信区间以名义覆盖率覆盖真实偏移。其贡献在于从设计理论角度说明了如何在没有观测数据的情况下检测并界定改装引起的校准偏移，以及随之而来的样本量要求。",65,{"impact":41,"substance":17,"depth":267,"authority":84,"freshness":13,"relevant":22,"comment":380},"方法学扎实、统计设计严谨的预印本论文，但偏理论设计层面，缺乏田间实证数据，产业影响有限，适合作为技术参考而非每日精选头条。",[382],{"name":79,"url":375},[89,90,384,385,386],"计算机视觉","产量预测","传感器校准",{"code":28,"slug":29,"name":30,"path":31,"parent_code":32},[],{"status":35,"locked":14,"version":36,"method":37,"evidence":390},{"candidates":391,"agricultural_evidence":42},[392,397,401],{"code":28,"score":393,"supported":42,"title":394,"abstract":395,"tags":396},24,[176,112],[384,112],[90,384],{"code":116,"score":22,"supported":14,"title":398,"abstract":399,"tags":400},[],[],[89],{"code":351,"score":22,"supported":14,"title":402,"abstract":403,"tags":404},[],[],[385],[406,407],"计算机视觉 产量预测 传感器校准","农业人工智能 传感器校准 计算机视觉 产量预测","计算机视觉产量预测传感器校准-5587","10.5281\u002Fzenodo.23544195",{"doi":409,"openalex_id":411,"authors":412,"venue":79,"cited_by_count":124,"oa_url":375,"card":419,"direction":139,"ingested_from":141},"W7223104380",[413,415,417],{"name":414,"orcid":8},"Tigist Alemu Gebremariam",{"name":416,"orcid":8},"Hawi Tesfaye Desta",{"name":418,"orcid":8},"Mulugeta Bekele Wolde",{"tldr":420,"method":421,"finding":422,"direction":139,"opportunity":423},"提出改造传感器替换后视觉产量模型校准漂移的检测与置信区间保证方法。","用前后对照影响估计量，基于方差结构做2000次重复的模拟设计分析。","5个百分点效应下54个集群可达80%检验力，估计量一致且渐近正态。","可将该设计理论扩展到多传感器异构替换与在线校准监测，填补真实田间验证空白。","2026-10-22T11:21:01.554995Z"]