Accurate fruit detection provides an important basis for subsequent fruit counting, yield estimation, and orchard monitoring. Extending unmanned aerial vehicle (UAV)-based monitoring from daytime to nighttime can reduce the influence of natural illumination variations and improve operational continuity. However, nighttime fruit detection remains challenging due to low-light image degradation, severe fruit occlusion, and limited edge computing resources. In addition, active illumination methods often introduce overexposure and shadow artifacts. To address these challenges, this study utilizes a full-color night vision imaging module to acquire citrus images without active supplementary illumination under extremely low-light conditions and develops a lightweight citrus detection network, termed NightNet, based on the acquired nighttime images. To alleviate the shortage of high-quality nighttime annotations and mitigate the long-tail distribution problem, a fruit spatial distribution augmentation strategy is further introduced to synthesize dense and heavily occluded fruit scenarios. Moreover, NightNet incorporates a newly designed dual-frequency feature reconstruction and edge suppression module, a lightweight detection head, and the existing Focaler-CIoU regression loss. These components jointly improve citrus detection performance in unstructured nighttime orchard environments. Experimental results demonstrated that NightNet achieved a precision of 89.62 %, a recall of 85.94 %, and an mAP@50 of 94.48 % on the self-constructed nighttime aerial dataset, while requiring only 2.36 M parameters and 4.8 GFLOPs. The favorable balance between detection accuracy and computational efficiency indicates that the proposed model has edge-deployment potential for resource-constrained UAV-based orchard perception. Overall, this study provides a useful detection basis for nighttime UAV-based orchard perception and future orchard monitoring applications.
All-terrain vehicles (ATVs) are widely used in agricultural operations, yet rollover incidents remain a major cause of severe injury and fatality among farm workers. The addition of attachments alters the vehicle’s center of gravity (CG) and may significantly affect stability and injury risk. This study developed and validated a multi-body finite element (FE) model of an ATV equipped with front and rear rack attachments, a tandem-axle trailer, and a 50th-percentile rider dummy to evaluate static rollover stability under varying operational conditions, enabling the assessment of configurations such as ATV-trailer systems that are costly and difficult to test experimentally. The model incorporated CG measurements obtained from tilt-table and corner-weight testing. Static stability was quantified using the tilt table ratio (TTR) under forward pitch, rearward pitch, and lateral roll conditions. Head injury risk was assessed using the 15 ms head injury criterion (HIC15) during simulated rollover events. The FE model predictions of TTR showed good agreement with experimental results (average percentage error generally < 10 %). Attachment configuration significantly affected stability across all rollover modes, with rear and combined rack loads producing the largest reductions in TTR. Lower tire inflation pressure reduced forward pitch stability and interacted with attachment configuration in lateral and rearward rollovers. Forward pitch rollovers generally produced the highest HIC15 values, reaching 1095 for the ATV equipped with both front and rear racks. The relative severity of rearward pitch and lateral rollovers varied with attachment configuration. Nevertheless, nearly all predicted HIC15 values remained below the accepted injury threshold of 700.
Aquaculture production systems are becoming increasingly data-driven, with intelligent monitoring, control, and decision-support tools now playing a central role in managing the interaction between cultured organisms and their production environments. This transition is enabled by advances in low-cost sensing, embedded computing, and artificial intelligence, which are changing how farms are instrumented, monitored, and operated under practical constraints such as biofouling, sensor drift, and limited connectivity. This review takes a design-oriented, systems-engineering perspective on artificial intelligence integration across five aquacultural-engineering domains: water quality management, feed management, disease management, genetic selection, and phenotyping. Each domain is framed around a four-stage pipeline (input acquisition, preprocessing, modelling, application). The review (i) abstracts the dominant sensing-to-actuation patterns reported in the artificial-intelligence-based aquacultural engineering literature across these domains; (ii) proposes reference system architectures to support deployment in real-world farm settings within each domain; and (iii) synthesises cross-cutting challenges and future research directions drawn from the surveyed literature. By translating fragmented artificial intelligence advances into practical system-level blueprints, this review aims to support more reliable, scalable, and maintainable adoption of artificial intelligence in aquaculture. These architectures can help the industry move beyond isolated proof-of-concept models towards integrated farm-management systems that improve productivity, animal welfare, resource efficiency, and operational decision-making.
Severe wheat lodging requires rapid quantitative assessment, but satellite-based estimation is often limited by scarce spatially compatible reference labels. This study developed an unmanned aerial vehicle (UAV)-derived reference-label framework for estimating wheat lodging fraction (WLF) from Sentinel-1/2 imagery in Xinxiang City, China. UAV multispectral imagery was used to classify lodged and non-lodged wheat; a multi-layer perceptron (MLP) was selected, and its binary outputs were aggregated within Sentinel-aligned 10 × 10 m grids to generate continuous WLF labels. After matching, 5,380 grids from 46 UAV source orthomosaics were partitioned by complete source groups into model-development (4,337 grids; 36 groups) and sealed independent test (1,043 grids; 10 groups) subsets. Six regressors were compared using four Sentinel configurations representing post-event optical data, post-event optical plus synthetic aperture radar (SAR) data, optical pre–post changes, and combined optical/SAR pre–post changes. Adding post-event SAR features produced limited and model-dependent benefits, whereas temporal-change features substantially improved several nonlinear models. The Feature Tokenizer Transformer (FT-Transformer) with combined Sentinel-1/2 pre–post changes achieved the best grouped cross-validation performance (coefficient of determination (R2) = 0.825; normalized root mean square error (NRMSE) = 12.60%). On the spatially independent test subset, it achieved R2 = 0.770, root mean square error (RMSE) = 0.117, and NRMSE = 12.11%. The model enabled 10 m WLF mapping across winter wheat-growing areas of Xinxiang. Together, the UAV-derived reference-label construction, source-grouped spatial validation, and controlled comparison of Sentinel configurations provide a reproducible framework for extending spatially distributed UAV observations to event-scale satellite WLF assessment.
Site-Specific Management Zones constitute a key component of precision agriculture, enabling the division of agricultural fields into homogeneous subregions based on soil and yield-limiting factors. However, most existing approaches primarily rely on statistical clustering techniques and often neglect the geometric shape of the zones and their interpretability when multiple soil properties are considered simultaneously. In this work, we propose an approach to the Site-Specific Management Zones problem that explicitly incorporates zone-shape constraints, facilitating practical implementation using both conventional and advanced agricultural machinery. Unlike traditional methods, the proposed approach simultaneously considers multiple soil properties while preserving interpretability, avoiding the need for dimensionality reduction techniques. The method is based on a Univariate Marginal Distribution Algorithm and aims to minimize the number of management zones while ensuring homogeneity through a relative variance criterion. The approach is evaluated using a benchmark dataset widely used in the literature. Computational results show that the method produces high-quality solutions across multiple soil properties with low computational effort. The proposed approach is also compared with state-of-the-art methods, demonstrating the advantages of simultaneously considering multiple soil properties while preserving the interpretability of the resulting management zones. In addition, an illustrative example shows how the delineated management zones can be translated into prescription maps to support site-specific agricultural decision-making. The simplicity, interpretability, and practical applicability of the proposed approach make it suitable for a wide range of socio-economic contexts.
