Search Results (3,533)

Cropland abandonment is a critical challenge for food security and the sustainable use of land resources, especially in the hilly and mountainous regions of southern China. To characterize the spatiotemporal evolution and associated-factor patterns of cropland abandonment in these regions, this study uses Ji’an City, Jiangxi Province, as a case study and develops an integrated framework for abandonment identification and monitoring, predictive-association analysis, relative risk assessment, and zoning management. Using Landsat remote sensing imagery from 1990 to 2024, crop phenological features, and multisource topographic, soil, climatic, ecological, and socioeconomic data, we identified and evaluated long-term cropland abandonment. The results show that, after incorporating phenological features and temporal filtering, annual land-use classification evaluated with spatially separated testing samples achieved a mean overall accuracy of 95.27% and a mean Kappa coefficient of 0.931, and the overall accuracy of cropland abandonment identification reached 91.58%, with a Kappa coefficient of 0.83. Cropland abandonment in Ji’an City showed an overall accelerating trend, with the abandonment rate exceeding 15% during 2020–2021. Spatially, abandonment was more intensive in the west than in the east and more severe in mountainous areas than in plains. Paddy field abandonment was the dominant form of cropland abandonment across the city, whereas dryland abandonment was more limited in extent but showed stronger interannual fluctuations. The predictive-association analysis indicated that tree cover consistently showed high predictive importance for both paddy fields and drylands. Paddy field predictions were more strongly associated with topographic position, surface roughness, and soil conservation, whereas dryland predictions showed stronger associations with aridity and population density. From 1993 to 2023, the model-estimated relative abandonment risk increased, and high-risk areas expanded from deep mountainous regions to low-hill areas. By coupling relative risk levels with associated-factor patterns, we divided paddy fields and drylands into four management zones: core protection zones, dynamic warning zones, integrated improvement zones, and ecological fallow zones. Differentiated management strategies were then proposed for each zone. This study provides methodological support and decision-making evidence for cropland abandonment monitoring, cropland protection, and differentiated land management in the hilly and mountainous regions of southern China. Full article

Intelligent Transportation Systems (ITS) increasingly combine sensing, communication, computation, data platforms, and mobility services. This article presents a structured, literature-based mapping review of ITS applications from a standards-oriented perspective. The study analyzes 42 studies organized into five thematic groups and evaluated through 63 article–standard assessments using selected ITU-T Recommendations as an analytical lens. The rubric used in this work examined whether each study reported, or allowed reviewers to infer, evidence on architecture, data handling, interoperability, security and privacy, deployment assumptions, and digital-twin capabilities. Partial alignment was the most frequent outcome, accounting for 25 of 63 article–standard assessments (39.7%). At group level, satisfactory or optimal alignment occurred in 5 of 8 assessments (62.5%) in the digital-twin group and in 3 of 13 (23.1%) in the Big Data group; in the latter, 6 of 13 assessments (46.2%) showed limited or no alignment. Stronger alignment was usually found when studies described architectures, data flows, sensing mechanisms, service workflows, physical–virtual modeling, or system-management components relevant to the Recommendation, and weaker alignment when they focused mainly on algorithms, datasets, prediction accuracy, authentication, or secure dissemination without sufficient detail on interfaces, data governance, gateway roles, deployment conditions, or platform integration. The review proposes a five-dimension standards-facing reporting checklist addressing interoperability, data lifecycle and governance, security and privacy, operational readiness, and standards-facing evidence. It supports traceable reporting through explicit evidence-status categories and locations, and can be implemented as a Standards and Interoperability Reporting Statement (SIRS) for authors, reviewers, and editors. Overall, the findings show that standards-oriented assessment depends not only on technical performance but also on explicit and traceable integration evidence, while the proposed reporting profile provides a practical mechanism for making such evidence more systematically visible in future ITS studies. Full article

Financial return prediction increasingly relies on both financial text and structured market covariates, but adapting large language models across financial institutions remains difficult because raw data cannot be centralized and clients often exhibit heterogeneous, non-stationary market signals. This paper studies data-local federated financial return prediction with a frozen LLM, aiming to share useful cross-client adaptation while preserving client-specific predictive behavior. We propose a federated routed-adapter framework in which the server maintains a pool of lightweight adapter prototypes, each client selects a personalized mixture of these prototypes through projected directional routing, and local residual adapters are learned on private client data around the selected mixture. The server then maps uploaded residual updates back to the shared prototype space through an exact least-norm decomposition for communication-efficient aggregation. The framework keeps raw financial data and client-private prediction heads local, while uploaded residuals remain model updates and should not be interpreted as a formal privacy guarantee without additional mechanisms such as secure aggregation or differential privacy. Across FNSPID, Qlib CSI300/CSI800, and Open FinLLM forecasting benchmarks, our method achieves the best overall performance, improving CSI300 RankIC from $0.082$ to $0.087$ over the strongest federated PEFT baseline and reducing FNSPID MAE from $0.00537$ to $0.00482$. These results suggest that compositional shared adaptation with local residual personalization is a practical direction for financial LLM deployment under data-local, communication-constrained, and heterogeneous federated settings. Full article

Vision-based traffic accident anticipation is critical for active vehicle safety systems, yet existing architectures frequently conflate performance gains with heavy parameter scaling optimized from scratch on compact domains. Consequently, during real-time inference, these frameworks suffer from severe prediction volatility and early triggering biases that induce dangerous control instability. To address these limitations, this paper shifts the research focus away from network modifications toward a highly controlled, strategy-driven training pipeline executed under a completely invariant spatial–temporal neural backbone. Our proposed paradigm establishes a robust framework through three decoupled milestones. First, an out-of-domain initialization strategy transferred generalized driving kinetics from a large-scale sequence domain (Mapillary) to serve as a stable temporal anchor. Second, a target-domain generative enrichment step injected synthetic nighttime scenes to decouple hazard features from low-light ambient noise. Third, progressive temporal supervision paradigm scaling targeted labels monotonically to align with continuous kinetic risk accumulation. Overall evaluations on the Car Crash Dataset (CCD) benchmark demonstrate that the fully integrated configuration (C4) pipeline achieves 69.89% in frame-level Mean Average Precision (mAP), which is an improvement of +22.81 percentage points over the baseline configuration. Continuous temporal measurements prove that our framework can adapt to tracking volatility, compressing Temporal Confidence Variance to 0.00328, and dropping the Prediction Instability Count to 0.66. While hyper-sensitive baselines report early raw latency averages driven by premature trigger noise, our model purposefully filters this early-frame variability to deliver a secure warning profile, achieving an absolute zero false alarm rate (FAR = 0.00%) across evaluated non-hazardous driving sequences, establishing the sequence-level trustworthiness required for practical autonomous deployment. Full article

This study introduces a training-free, self-contained adaptive low-light image enhancement (LLIE) framework driven by a metaheuristic optimization algorithm (MOA) and a context-aware dual image quality assessment (IQA) engine. Although deep-learning-based methods exhibit rapid inference, their static parameters often suffer from severe performance degradation in out-of-distribution (OOD) scenarios—such as those involving unseen sensor noise or environmental shifts. To bridge this generalization gap, the proposed framework operates within a decoupled HSV color space, specifically targeting the luminance ($V$) channel to formulate image enhancement as an instance-specific optimization task. We introduce a novel hybrid Log-Gamma mapping function that mathematically unifies the localized dark-stretching capabilities of logarithmic compression with the global dynamic range regulation of power-law gamma curves, thereby substantially expanding the expressiveness of the transformation space. To govern parameter convergence without reference images, a multi-stage Low-Light Image Discrimination (LLID) engine classifies the input frame by computing context-specific trimmed skewness residuals and global intensity means, effectively mitigating highlight biases. Under normal-light conditions, the swarm intelligence engine optimizes the Log-Gamma coefficients via the Patch-based Contrast Quality Index (PCQI) to maximize structural fidelity; conversely, under severe low-light degradations, the framework leverages the Blind/Referenceless Image Spatial Quality Evaluator (BRISQUE) to minimize spatial artifacts. Using the Marine Predators Algorithm (MPA), the framework iteratively searches the continuous bounding space to fine-tune a parameter matrix tailored exclusively to each image. Empirical evaluations across four benchmark datasets (Bicycle, DF1000, DICM, and VV) validate the effectiveness of the proposed paradigm. The proposed variant, $OLG\left(\mathrm{MPA}\right)$, secured the top average rank in internal algorithm ablation ($\overline{R}=3.10$) and achieved a competitive global average rank ($\overline{R}=2.95$) against four state-of-the-art deep networks, matching the performance of leading data-driven models. Although the evolutionary optimization loop incurs an average per-frame latency of $17.300 \mathrm{s}$, this instance-specific paradigm successfully trades instantaneous processing speed for absolute domain adaptability and predictable, artifact-free image restoration. Full article

Rice is a staple grain crop central to China’s food security. As the core production region of single-cropping rice, the middle and lower reaches of the Yangtze River face escalating high daytime and nighttime temperatures and compound drought–heat stress amid global warming. The accurate assessment of heat hazards is therefore pivotal for regional yield stability and disaster mitigation. Based on meteorological, remote-sensing, and soil data, together with county-level rice yield statistics from 150 major producing counties spanning 1991 to 2024 (5009 county-year calibration units), we first constructed a composite heat damage index (CHI) by integrating daytime harmful accumulated temperature (Ha), nighttime harmful accumulated temperature (HNa), and the Vegetation Health Index (VHI). We then implemented a gradient boosting decision tree (GBDT) machine learning framework in which yield loss was imposed as a physical constraint. This framework was benchmarked against convolutional neural network (CNN), random forest (RF), and support vector machine (SVM) models, with the Shapley additive explanations (SHAP) method used for attribution analysis and an independent temporal partitioning strategy applied for model validation. The results indicate the following: (1) compared to the single daytime heat damage index, the CHI elevated the yield correlation coefficient from 0.52 to 0.63; (2) with yield constraint calibration, the model attained a balanced accuracy of 92.6% and 94.0% consistency with historical disaster records; (3) regional heat hazard presents a spatial pattern of “high in inland areas and low in coastal areas,” with the heading–flowering stage as the critical sensitive period; and (4) high nighttime temperature accounts for approximately 20% of the model’s relative importance, with higher discriminative sensitivity for high-grade hazards, while the amplifying effect of water deficit on heat stress maintains a stable relative importance of around 16%. In this study, the coupled optimization of traditional assessment paradigms and data-driven approaches is achieved, providing a methodological reference for refined growth stage–specific heat hazard assessment. Its cross-regional portability and independent predictive validity require further validation. Full article

Andean livestock agroecosystems depend on grassland and soil functionality to sustain productivity, threatened by climate-driven glacial retreat, pasture degradation, and increasingly frequent frosts, hailstorms, and droughts. Grounded in the Resilience Thinking framework for socio-ecological systems, this study proposes and validates an explanatory model examining how perceived climate change impacts (IMP) drive the adoption of regenerative livestock farming (ADOP) and impact the deterioration of livestock performance (DES), with adaptive capacity and resilience (RES) as a mediating construct. Regenerative livestock farming is operationalized through soil, forage, water, and biodiversity management, animal welfare, and climate adaptation strategies. Survey data from 91 South American camelid producers in the Puno region, Peru, were analyzed using Partial Least Squares Structural Equation Modeling (PLS-SEM) with second-order constructs. Climate impacts emerge as the strongest antecedent of adaptive resilience (β = 0.718) and regenerative adoption (β = 0.604) and have a significant direct effect on performance deterioration (β = 0.292; p = 0.041). However, resilience does not significantly predict adoption (β = 0.075; p = 0.619), revealing institutional, technological, and knowledge-related barriers that disconnect adaptive capacity from ecological transformation. Both adoption (β = 0.327; p = 0.002) and resilience (β = 0.335; p = 0.003) are positively associated with performance deterioration, suggesting reactive containment rather than consolidated restoration. Findings indicate adaptive rather than transformative resilience in high-Andean livestock systems, indicating that an effective regenerative transition requires participatory, context-sensitive approaches supported by policies and technical assistance that strengthen long-term productive sustainability and food security in vulnerable communities. Full article

Background: Camel husbandry underpins food security in drylands, and dromedary (Camelus dromedarius) milk is valued as a functional food whose composition is thought to be shaped by the desert forage that camels browse. Dietary plant microRNAs (miRNAs) have been proposed as one molecular route linking diet to mammalian physiology, and two interactions are widely cited from experimental reports: rice miR168a repressing LDLRAP1 and plant miR159 repressing TCF7. The hypothesis remains contested, however, and it has not been tested for camel forage against camel transcripts. Methods: We generated de novo, hairpin-based miRNA repertoires for three arid-zone forage plants relevant to camel feeding (Moringa oleifera, Ziziphus jujuba, Medicago sativa) and screened the mature miRNAs against 48,746 reconstructed C. dromedarius 3′-UTRs under stringent thresholds, retaining one transcript per gene, weighting interactions by read abundance, normalising scores for 3′-UTR length, and testing over-representation against a matched background. The two previously reported cross-kingdom pairs served as internal positive controls and seed-level grouping as a sensitivity control. Results: The three forages yielded 170 hairpin-validated miRNA loci (81 M. sativa, 49 Z. jujuba, 40 M. oleifera) peaking at 21 nt, collapsing to 116 mature sequences and 104 seed groups. The pooled read set was strongly concentrated: five mature sequences shared by all three species carried 51% of reads, miR159 alone 31%, and 19 sequences assignable to conserved miRBase families carried 64%. The pipeline recovered the reported miR168a–LDLRAP1 pairing in the camel; the miR159–TCF7 pairing, by contrast, was not recovered, although TCF7 was among the genes targeted by other plant miRNAs. Genome-wide, predicted targeting was sparse (median 2 miRNAs per gene) and no GO, KEGG or Hallmark category was enriched at either threshold (best FDR 0.56); the nominal p-value distribution was approximately uniform, giving no evidence of systematic enrichment under the tested framework. Targeting multiplicity scaled with 3′-UTR length (Pearson r = 0.63; Spearman ρ = 0.58), so apparent “hub” genes are largely long-3′-UTR genes. Of 19, 12 curated milk-fat and lactation genes were among predicted targets, without over-representation (Fisher p = 0.36). Conclusions: A sensitive, species-matched analysis recovered a previously reported cross-kingdom pairing yet found no coordinated enrichment of dietary plant miRNAs on the dromedary transcriptome, and showed that an individual cross-kingdom pairing cannot be assumed to transfer between mammalian species. The work provides a first forage miRNA resource in the context of camel nutrition, sets out a reusable, species-matched analytical framework for cross-kingdom claims, and narrows future experimental work to five abundant forage-derived sequences and a short list of candidate genes (LDLRAP1, TCF7/TCF7L2, PRLR, INSR). Full article

Security operations can detect that an intrusion is under way, yet they cannot say how long an ongoing attack still needs to reach a critical objective such as data exfiltration. Prior work on multi-stage attacks identifies the active stage or predicts the next step, but does not estimate the residual time to compromise from real traffic using survival models. This paper addresses that gap through a temporal framework built on empirically measured stage durations, with three contributions. First, the MITRE ATT&CK taxonomy is given a temporal layer, in which each stage carries a duration distribution estimated empirically from the observed episodes of that stage. Second, a probability-weighted multi-path formulation combines these durations with stage-transition probabilities to estimate the time remaining before the objective. Third, the framework is validated on a real multi-stage campaign rather than on synthetic traffic, and three survival models are compared under a matched protocol as a benchmark of how learnable the durations are. Random Survival Forest, DeepSurv, and DeepHit are compared on DAPT 2020, a public advanced-persistent-threat dataset of 82,577 real network flows collected across five days. Random Survival Forest reaches a stable concordance index of 0.92, with a standard deviation of 0.006 across twenty repeated stratified splits on leakage-free features, and it retains a concordance of 0.79 when benign traffic is excluded entirely. When the three models are placed on a single concordance scale and trained on an identical subsample of 40,000 flows, DeepSurv reaches 0.955 and DeepHit 0.879, so the neural models are competitive at that scale. DeepSurv nevertheless fails to converge on the full flow set, returning no survival estimates in any of five seeds, whereas the forest fits successfully at every training size examined. A stage-transition graph recovered from the data, built from 25 observed transitions across ten multi-stage sessions, shows branching progression, and the residual time, reported at the entry to each stage, falls along the campaign, from about 139 min at reconnaissance to about 31 min at lateral movement, conditional on reaching the objective. All stage-level estimates rest on 74 episodes from a single campaign, of which 43 carry a positive duration, so cross-environment generalisation remains to be confirmed. The framework gives a security operations centre a data-driven estimate of the active attack effort that remains before compromise, supporting informed containment decisions. Full article

Universal Plug and Play (UPnP) is widely used for resource discovery in internet of things and smart-edge environments because of its lightweight and decentralized operation. However, conventional UPnP and Simple Service Discovery Protocol (SSDP) mechanisms expose static identifiers and service metadata, making them vulnerable to spoofing, replay, unauthorized resource enumeration, device fingerprinting, and long-term traffic correlation. Existing software-based authentication methods rely on stored credentials and do not protect discovery privacy, while conventional Arbiter Physical Unclonable Functions (PUFs) remain vulnerable to machine-learning modeling attacks and are typically used only for device authentication. This paper presents a novel Recursive Hybrid Entropy PUF (RHE-PUF) and a privacy-preserving secure UPnP discovery architecture. The proposed RHE-PUF introduces recursive adaptive delay propagation, entropy injection, feed-forward coupling, and multi-path timing diversification to increase challenge–response nonlinearity and modeling resistance. Its responses are used to generate dynamic ephemeral identities, authenticate devices anonymously, and encrypt SSDP service advertisements without exposing permanent device identifiers. The complete framework was implemented on a Xilinx Spartan-7 FPGA and evaluated under realistic UPnP discovery and attack scenarios. The RHE-PUF achieved $49.31\%$ uniqueness, $98.14\%$ reliability, $50.22\%$ uniformity, and $98.91\%$ entropy. The implementation operated at up to 192 MHz with 0.84 W dynamic power, 0.88 µs authentication latency, and 13.4 ms secure discovery delay. The strongest deep-neural-network modeling attack achieved only $58.27\%$ prediction accuracy. Replay and spoofing attack success rates were reduced to at or below $1.1\%$ and $0.8\%$, respectively, while long-term tracking probability remained below $13\%$. These results demonstrate that the proposed joint hardware-security and privacy-preserving discovery framework provides resource-efficient authentication, anonymous UPnP resource discovery, and resistance to network and machine-learning attacks. Full article

Internet of Vehicles (IoV) security mechanisms often classify isolated messages or assign node-level trust scores, yet these decisions do not explain whether a malicious but authenticated event has distorted the intended evolution of traffic. This paper proposes IntentProv-IoV, a causally grounded provenance framework for traffic-intent preservation in V2X environments. Traffic intent is modeled as the short-horizon collective state expected under non-adversarial conditions, and deviation is measured between predicted and observed traffic states. The framework constructs temporal provenance graphs linking vehicles, roadside units (RSUs), cooperative perception outputs, prediction nodes, and traffic-control decisions. To remove the ambiguity of marginal contribution, node contribution is formalized as an interventional effect in a structural causal model and estimated through Monte Carlo counterfactual edge-weight attenuation, with a linear sensitivity fallback for real-time edge deployment. A calibrated composite score integrates anomaly evidence, traffic-intent deviation, trust risk, and provenance contribution. The evaluation design compares IntentProv-IoV with detection, trust, blockchain trust, graph anomaly, Granger causal, structural causal, and counterfactual GNN baselines and includes predictor sensitivity, adaptive adversaries, prediction noise, packet loss, trajectory-only real-data validation, and edge overhead. Simulation-scale results indicate improved attribution precision, stronger traffic-intent deviation reduction, and edge-suitable latency. By shifting V2X security from message-level detection to causally explainable traffic-intent assurance, IntentProv-IoV provides a more accountable security objective for cooperative vehicular systems. Full article

Short-term electric load forecasting is essential to the secure and stable operation and economic dispatch of power systems, and its accuracy directly affects grid dispatch decisions and operational efficiency. To address the inadequate modeling of heterogeneity among historical exogenous variables, the underutilization of known future calendar information, and the difficulty of correcting local biases in multi-step forecasts, this paper proposes KCD-GEIRTimeXer, a short-term electric load forecasting method built upon TimeXer. Before exogenous variable embedding, a Grouped Exogenous Importance Recalibration (GEIR) module is introduced. Historical exogenous variables are first grouped a priori according to their sources and physical meanings. Importance scores are then computed at the feature, variable-group, and time-step levels, and the corresponding exogenous variable weights are obtained through a bounded residual gating mechanism. At the prediction stage, a Known Calendar–Causal Dual Correction (KCD) module is further incorporated. The module refines the initial forecasts using known future calendar variables and causal statistical features derived exclusively from historical load observations, thereby mitigating local biases in multi-step forecasting. Experiments on the publicly available Panama electricity load dataset demonstrate that the proposed model outperforms all baseline models. Averaged over three random seeds, KCD-GEIRTimeXer reduces MSE, RMSE, MAE, and MAPE by 26.25%, 14.12%, 13.10%, and 12.91%, respectively, compared with TimeXer. Full article

The original ALPHA benchmark introduced a taxonomy-aware penalty for evaluating CWE-level vulnerability prediction in Python and proposed that the penalty could theoretically also serve as a training signal. This paper tests that proposal empirically, comparing three delivery mechanisms–supervised fine-tuning, a dual-head classification loss, and reinforcement learning with a dense reward derived from the normalised penalty. We find that supervised approaches consistently regress below the zero-shot baseline under distribution shift, while GRPO succeeds and generalises across independent training seeds. Using a default, non-test-informed configuration, GRPO reduces the mean cumulative ALPHA penalty of Qwen2.5-Coder-7B-Instruct on Security Hardening and Adversarial Testing (SVEN) by 13.5%. A further test-selected configuration reduces the mean penalty by 22.9% (27.9% in the best-observed run), reaching statistical parity with its 4.5× larger zero-shot teacher. Our experiments show that reinforcement learning outperforms the alternative delivery mechanisms we test. We conclude that delivery mechanism, rather than the specific hierarchical structure of the reward, is the primary variable—though whether taxonomy-aware structure contributes further remains an open question. Full article

Accommodation establishments constitute a core component of tourism infrastructure, and their location choices directly affect water resource utilization, land pressure, and the spatial equilibrium of tourism development—issues that are particularly acute in vast arid regions. Yet the spatial organization of accommodation supply across extensive drylands characterized by fragmented oasis distribution, and the reasons why standard and non-standard accommodation follow divergent location logics, remain poorly understood. This study addresses three questions: (1) How are nine accommodation categories, differentiated by type and quality, distributed across Xinjiang? (2) Do directional spatial associations exist among categories that are consistent with hierarchical, path-dependent development? (3) Which factors drive these patterns, and do their effects exhibit the nonlinearity and threshold behavior predicted by location theory? Drawing on 12,073 accommodation establishments from the Ctrip platform, we construct a staged analytical framework in which each technique answers a specific question: the nearest-neighbor index and standard deviational ellipse characterize global patterns; kernel density estimation and OPTICS clustering identify local agglomerations; directional local co-location quotients measure asymmetric spatial associations; and XGBoost–SHAP isolates nonlinear drivers and threshold effects. Results reveal a highly concentrated “single-core, multi-center” structure anchored by Urumqi, Yining, and Kashgar, with rapid expansion toward the Ili Valley, Kashgar, and Altay since 2019. Standard accommodation tracks urban centrality and transport nodes, while non-standard accommodation tracks tourism resource endowments, consistent with location-theoretic expectations. Directional co-location analysis reveals hierarchical spatial associations among categories, and driving factors exhibit pronounced nonlinear threshold effects. From a sustainability perspective, the identified thresholds—elevation (1360 m), water-body proximity, and distance to rural tourism demonstration sites (3 km)—constitute quantifiable, spatially explicit sustainability indicators that can be incorporated into planning tools to monitor and steer accommodation development away from ecologically sensitive zones. Global Moran’s I diagnostics of model residuals (reduction of 83–99.7%) suggest that these findings are unlikely to be artifacts of spatial autocorrelation; this diagnostic, however, complements rather than replaces spatially blocked validation. The study contributes category-differentiated, spatially directed evidence for policies balancing tourism expansion against water security and ecosystem integrity, serving sustainable tourism development in arid-region destinations. Full article

Intelligent agents are emerging as an important system-level paradigm for smart agriculture. This review focuses on modern agricultural intelligent agents driven by large language models and related multimodal foundation models and examines how this emerging field is reshaping the organization of intelligent agricultural systems. It first clarifies the conceptual boundaries of agricultural intelligent agents and distinguishes them from traditional multi-agent systems, agent-based modeling, agricultural foundation models, and static retrieval-augmented question-answering systems. It then synthesizes their architectural foundations, key capabilities, application scenarios, deployment challenges, and future research directions. The reviewed literature indicates that agricultural intelligent agents are moving beyond isolated perception, prediction, and response generation toward the goal-oriented coordination of agricultural knowledge, dynamic data, external tools, and decision-making processes across agricultural task chains. They are beginning to support more integrated forms of knowledge services, crop monitoring and diagnosis, decision support, and farm-level collaborative management. Nevertheless, their transition from prototype systems to dependable and deployable agricultural systems remains constrained by context-aware knowledge grounding, heterogeneous data and tool integration, long-horizon reliability, the stability of multi-agent collaboration, and system security. This review further introduces an assessment perspective based on evidence reported in the original studies, comparing representative agricultural intelligent agents in terms of task decomposition, agronomic evidence applicability, tool-use validity, workflow reliability, multi-agent coordination, and deployment-related evidence. By distinguishing demonstrated capabilities from unevaluated dimensions, this review provides a structured framework for understanding the current status of agricultural intelligent agents and for guiding their future development toward reliable, deployable, and domain-oriented intelligent systems for smart agriculture. Full article