Songchao Zhang

Detecting fraud on financial transaction networks is difficult for three reasons that tend to appear together: fraudulent users deliberately disguise their behavior, their activity is linked to other entities through several distinct relation types, such as a shared merchant, a shared rating or a short time window, and they make up only a small fraction of the accounts in an otherwise benign population. Standard graph neural networks (GNNs) fold these relational cues into a single homogeneous neighbor aggregation and supervise every node with a label-balanced loss, which caps both their expressiveness and their recall on the rare fraud class. We introduce RGT-Fraud, a relational graph transformer that couples relation-aware message passing with global self-attention through a per-node topology gate, and we train it with an imbalance-aware contrastive (IAC) objective. The model operates on a multi-relation graph whose labeled nodes are the review or account entities to be classified, connected by three complementary relation types that encode a shared merchant, a shared rating or amount, and a short co-occurrence time window; merchants, ratings and timestamps are transaction attributes that define these relations rather than independently featured nodes. This is the standard construction of the YelpChi, Amazon and T-Finance benchmarks, and the same formulation extends to additional node types when a dataset exposes them. A relational position encoding writes edge-type information into the attention scores, and a topology-gated block lets each node decide how much local aggregation and how much long-range attention to use. The IAC objective adds three terms: a focal cross-entropy term for class imbalance, a supervised contrastive term that tightens the fraud cluster in embedding space, and a neighbor-consistency term that damps adversarial structural noise. On three public benchmarks (YelpChi, Amazon and T-Finance) RGT-Fraud improves F1-macro by 1.4 to 1.6 points and AUC by 0.9 to 1.5 points over the strongest competitor among twelve baselines, which include the recent heterogeneous multi-relational detector TMR-GGNN, and it keeps more than 70% F1-macro under 30% label noise. Ablations attribute a clear, positive contribution to each module and each loss term. Taken together, the framework offers an interpretable semantic layer that links raw transaction streams, a graph inference engine and human risk analysts inside financial expert systems.

How much a college education pays off varies widely from one student to the next, and that variation matters for admissions, financial aid, and mobility policy. Most estimates, however, report a single average return and treat each institution as an isolated row in a table, ignoring how schools relate to one another and how outcomes are distributed across demographic groups. We present FairEdu-GCT (Graph-Enhanced Causal Transformer), a framework that couples a heterogeneous graph encoder with a Transformer sequence model and an explicit fairness penalty. Institutions, geographic regions, and academic disciplines form a typed graph whose edges record graduate flows, spatial proximity, and disciplinary overlap. A Relational Graph Attention Network (R-GAT) turns this structure into institutional ecosystem embeddings that carry peer effects, regional labor-market signals, and the spread of institutional prestige, none of which survives in tabular representations. A Transformer then encodes each student’s educational history and merges it with the institutional embedding through a cross-modal attention bridge, and a counterfactual decoding head returns the full conditional earnings distribution under alternative institutional choices rather than a single point estimate. Because students are not randomly assigned to schools, we make no claim of strict causal identification; we treat CATE and PEHE strictly as estimation-quality diagnostics for a confounding-adjusted contrast, not as evidence of a proven causal effect. We instead adjust for observed confounders through a doubly robust objective and add an equalized opportunity regularizer so that accuracy does not come at the expense of protected subgroups. On linked U.S. College Scorecard, IPEDS, and NLSY97 data (6256 institutions drawn from 7312 Title IV schools; 161,043 person-institution-year records from 8984 respondents followed for ten years), FairEdu-GCT lowers RMSE by 16.2% and Precision in Estimation of Heterogeneous Effects (PEHE) by 25.1% against the strongest baseline (Causal Forest, DragonNet, TARNet, and TabTransformer), and shrinks the demographic parity gap by 46.5%. All gains are reported with 95% confidence intervals and effect sizes over ten seeds, and an extended fairness audit (calibration, predictive parity, and subgroup robustness) confirms the improvement is not confined to the two metrics we optimize. Ablations attribute a 9.0% RMSE reduction to the R-GAT encoder alone. A group-conditional SHAP analysis further shows that graph-derived proximity to regional technology clusters is an unusually strong predictor for first-generation minority students, a signal that tabular features cannot recover and one we read as a within-model association rather than a policy lever.

In order to address the problems of manual dependence, low replenishment efficiency, and insufficient operational continuity in unmanned field operations of large sprayers, an autonomous docking device for an unmanned replenishment vehicle was designed. The device is composed of three principal components: a robotic-arm docking system, a pesticide delivery system, and a docking control system. RTK positioning information is utilised to determine the relative position between the unmanned replenishment vehicle and the large sprayer. The robotic arm approaches the high-position filling port, and the end effector completes guidance, flexible compensation, and electromagnetic coupling. A geometric model of the robotic arm was established, and its workspace was analysed using the Monte Carlo method. Single-factor tests and response surface optimization tests were conducted to optimize the key parameters of the end effector, and robotic arm control accuracy tests and field collaborative docking tests were carried out to evaluate the performance of the device. The results showed that the workspace of the designed robotic arm covered the elevated filling-port area of the large sprayer and met the docking requirements within a vehicle spacing of 0.25–1.25 m. After parameter optimization, the predicted cumulative docking time of the end effector was 2.051 s. The field collaborative docking test showed that, within a vehicle spacing range of 25–125 cm, 56 of 60 docking trials were successful, giving an overall success rate of 93.33%. Within the medium-spacing range, stable docking was achieved with an average docking time of 44.10–47.89 s. The results indicate that the proposed autonomous docking device can support unmanned vehicle approach, robotic arm positioning, end-effector guidance, and stable pesticide replenishment of large sprayers.

Real-time and accurate crop row extraction is a fundamental requirement for vision-based perception in autonomous agricultural machinery. In maize fields, however, row detection is easily affected by variable illumination, leaf occlusion, weed interference, and uneven soil backgrounds, which can reduce the reliability of both GNSS- and image-based navigation methods. To address these challenges, this study proposes a plant-oriented crop row perception framework that reconstructs row structures from individual maize plant detections. A lightweight detection model, named Maize–YOLO, was developed based on YOLOv11n for maize seedling detection. Three key improvements were introduced to enhance the balance between accuracy and efficiency. First, the C3k2_Faster_CGLU module replaces the original C3k2 block to reduce redundant convolutional computation while improving selective feature representation through convolutional gated linear units, thereby enhancing robustness under complex field backgrounds. Second, a lightweight shared detection head, Detect_LSH, was designed to share convolutional parameters across multi-scale feature maps and adaptively adjust feature amplitudes, reducing detection-head redundancy while maintaining multi-scale prediction capability. Third, a Layer-Adaptive Magnitude-Based Pruning strategy was applied to remove low-contribution channels and further improve computational efficiency for CPU-based deployment. Experimental results on field-collected maize seedling images showed that Maize–YOLO achieved an mAP@0.5 of 97.6%, reduced GFLOPs by 61.9%, and maintained a CPU inference speed of 84.4 FPS. After plant detection, row centerlines were estimated using an IPM–DBSCAN–LSM pipeline, which transformed detected plant centers into a quasi-top-view space, clustered them into crop rows, and fitted continuous centerlines. The extracted crop rows reached a positional accuracy of 98.6%, with a mean angular deviation of 0.44°. These results demonstrate that the proposed method can provide accurate, lightweight, and real-time crop row perception for autonomous agricultural navigation and precision field operations.

In order to elucidate the trait structure of yield formation and optimize the selection strategy for breeding high-yield spring rapeseed, this study systematically evaluated the genetic variation, interrelationship, and contribution to yield of 10 key agronomic traits. A comprehensive assessment of 26 varieties across five test environments was conducted using the coefficient of variation, phenotypic correlation, path analysis, principal component analysis, and grey relational analysis. The results showed that the variations in plant height, branch position, and the number of primary effective branches were the most abundant (CV > 0.20), indicating high genetic improvement potential. Among the yield components, a significant positive correlation was observed between the number of effective pods per plant and the number of seeds per pod. The direct positive effect of pod length on yield per plant was the strongest (path coefficient = 0.467), indicating that yield formation was more dependent on pod structure and grain filling ability. Principal component analysis showed that PC1 had a contribution rate of 94.2%, driven mainly by the effective pod number of the whole plant. This could be used as a comprehensive index to distinguish between different ecological groups and evaluate the overall growth potential. Grey correlation analysis further clarified that the effective length of the main inflorescence was most closely related to yield per plant (correlation degree = 0.847). In summary, this study proposes a high-yield breeding strategy of ‘quality first, collaborative improvement’, whereby pod length, 1000-grain weight, and effective length of the main inflorescence are used as core selection traits. This novel study involves coordinating and optimizing the number of effective branches and inflorescence structure, as well as screening stable genotypes through multi-environment identification, in order to achieve the efficient integration of yield components.

This study employed a combined approach of computational fluid dynamics (CFD), numerical simulations, and wind tunnel tests to investigate droplet drift characteristics and develop prediction models in order to address the issues of low pesticide utilization rates and high drift risk, associated with droplet drift during agricultural unmanned aerial vehicle (UAV) spraying, as well as the unreliable results of field experiments. Firstly, a numerical model of the rotor wind field was established using the multiple reference frame (MRF) method, while the realizable k-ε turbulence model was employed to analyze the flow field. The model’s reliability was verified through wind field tests. Next, the Euler–Lagrange method was used to couple the wind field with droplet movement. The drift characteristics of two flat-fan nozzles (FP90-02 and F80-02) were then compared and analyzed. The results showed that the relative error between the simulated and wind tunnel test values was within 20%. Centrifugal nozzle experiments were carried out using single-factor and orthogonal designs to analyze the effects of flight height, rotor wind speed, flight speed, and droplet size on drift. The priority order of influence was found to be “rotor wind speed > flight height > flight speed”, while droplet size (D_V50 = 100–300 µm) was found to have no significant effect. Based on the simulation data, a multiple linear regression drift prediction model was constructed with a goodness of fit R² value of 0.9704. Under the verification condition, the relative error between the predicted and simulated values was approximately 10%. These results can provide a theoretical basis and practical guidance for assessing drift risk and optimizing operational parameters for agricultural UAVs.

The operational efficiency and precision of boom sprayers, as critical equipment for protecting field crops, are vital to global food security and agricultural sustainability. In precision agriculture systems, achieving uniform pesticide application fundamentally depends on maintaining stable boom posture during operation. However, severe boom vibration not only directly causes issues like missed spraying, double spraying, and pesticide drift but also represents a critical bottleneck constraining its functional realization in cutting-edge applications. Despite its importance, achieving absolute boom stability is a complex task. Its suspension system design faces a fundamental technical contradiction: effectively isolating high-frequency vehicle vibrations caused by ground surfaces while precisely following large-scale, low-frequency slope variations in the field. This paper systematically traces the evolutionary path of self-balancing boom technology in addressing this core contradiction. First, the paper conducts a dynamic analysis of the root causes of boom instability and the mechanism of its detrimental physical effects on spray quality. This serves as a foundation for the subsequent discussion on technical approaches for boom support and balancing systems. The paper also delves into the evolution of sensing technology, from “single-point height measurement” to “point cloud morphology perception,” and provides a detailed analysis of control strategies from classical PID to modern robust control and artificial intelligence methods. Furthermore, this paper explores the deep integration of this technology with precision agriculture applications, such as variable rate application and autonomous navigation. In conclusion, the paper summarizes the main challenges facing current technology and outlines future development trends, aiming to provide a comprehensive reference for research and development in this field.

As a pioneering solution for precision agriculture, unmanned, robotics-centred greenhouse farms have become a key technological pathway for intelligent upgrades. The robotic arm is the core unit responsible for achieving full automation, and the level of technological development of this unit directly affects the productivity and intelligence of these farms. This review aims to systematically analyze the current applications, challenges, and future trends of robotic arms and their key technologies within unmanned greenhouse. The paper systematically classifies and compares the common types of robotic arms and their mobile platforms used in greenhouses. It provides an in-depth exploration of the core technologies that support efficient manipulator operation, focusing on the design evolution of end-effectors and the perception algorithms for plants and fruit. Furthermore, it elaborates on the framework for integrating individual robots into collaborative systems analyzing typical application cases in areas such as plant protection and fruit and vegetable harvesting. The review concludes that greenhouse robotic arm technology is undergoing a profound transformation evolving from single-function automation towards system-level intelligent integration. Finally, it discusses the future development directions highlighting the importance of multi-robot systems, swarm intelligence, and air-ground collaborative frameworks incorporating unmanned aerial vehicles (UAVs) in overcoming current limitations and achieving fully autonomous greenhouses.

The traditional continuous, quantitative spraying technology ignores the severity of pests, diseases and grasses, spatial distribution and other differences, resulting in low effective utilization of pesticides, environmental pollution and other problems. Variable-rate spray technology has become an important development direction in the field of precision agriculture by dynamically sensing crop canopy morphology, pest and disease distribution, and environmental parameters, adjusting the application amount in real time, and significantly improving pesticide utilization. In this study, we systematically review the core progress of variable-rate spray technology; focus on the technical system of information detection, spray volume model, and control system; analyze the current bottlenecks; and propose an optimization path to adapt to the complex agricultural conditions. At the level of information perception, LiDAR, machine vision, and multi-source sensor fusion technology constitute the main perception architecture, and infrared and ultrasonic sensors assist target recognition in complex scenes. In the construction of the spray volume model, models based on canopy volume, leaf area density, etc., are used to realize dynamic application decision by fusing equipment operating parameters, pest and disease levels, meteorological conditions, and so on. The control system takes the solenoid valve + PID control as the core program, and improves the response speed through PWM regulation and closed-loop feedback. The current technical bottlenecks are mainly concentrated in the sensor dynamic detection accuracy, model environmental adaptability, and the reliability of the execution parts. In the future, it is necessary to further promote anti-jamming multi-source heterogeneous sensor data fusion, multi-factor adaptive spray model development, lightweight edge computing deployment, and solenoid valve structural parameter optimization and other technical research, with a view to promoting the application of variable-rate spray technology to the field on a large scale and providing a theoretical reference and technological support for the green transformation of agriculture.

Introduction: Conventional air-assisted sprayers used in orchards often suffer from excessive pesticide waste, high residue levels, and uneven droplet distribution on fruit tree canopies. Precision spraying technologies have emerged to address these limitations by enabling dynamic regulation of spray parameters according to canopy characteristics. Among these, leaf area density is a key indicator for describing canopy sparseness. However, accurate and automated measurement of canopy leaf area density remains challenging due to leaf shading effects. As a result, few fully functional variable-rate spraying systems have been developed based on this parameter.Methods: This study presents a variable-rate spraying method that integrates global navigation satellite system (GNSS) positioning with wind-excited audio-conducted estimation of canopy leaf area density. A self-propelled orchard spraying platform was developed to acquire real-time GNSS positioning and audio-conducted canopy leaf area density data. Based on this, a method was established for generating prescription maps that integrate spatial positioning and canopy density information. A variable-rate spray control model and algorithm were then constructed to regulate spray flow according to the spatial distribution of leaf area density across the orchard.Results: Field experiments demonstrated that the system achieved a mean relative error of only 5.52% in spray flow rate regulation. Compared with conventional constant-rate spraying, the variable-rate mode reduced the longitudinal coefficient of variation (CV) of droplet deposition by 55.75% on adaxial leaf surfaces and by 33.22% on abaxial surfaces, with a maximum reduction of 62.32% in transverse CV. Ground runoff of spray solution was reduced by 62.29%, and droplet deposition density on leaf surfaces exceeded 25 droplets/cm², meeting the standard for low-volume insecticide application.Discussion: The integration of GNSS and wind-excited audio sensing for real-time canopy density assessment enables more precise and efficient pesticide application in orchards. This system significantly improves droplet deposition uniformity while reducing environmental losses, offering a promising technical solution for the development of intelligent and sustainable plant protection equipment.