Agricultural equipment is a crucial support for ensuring food security and strengthening the agricultural sector of a country. Focusing on the strategic position, industrial strength, innovation capability, and support capability of agriculture, this study summarizes the main achievements of the agricultural machinery industry of China since 2015. It also reviews the development experiences from four perspectives: scientific and technological innovation, industrial ecology, industrial policies, and international development. In response to major demands for ensuring national food security and building a strong agricultural nation in the new era, and in line with the development trends of agricultural machinery toward high efficiency, intelligence, networking, greenness, and robotization, the study proposes that the agricultural machinery industry of China should adhere to a Chinese-style path of mechanized and intelligent development over the next decade. Centered on the general direction of "strengthening foundations, increasing varieties, enhancing levels, improving functions, expanding fields, and extending chains," the study emphasizes focusing on the development of agricultural machinery that is high-end, intelligent, low-carbon, and applicable for hilly and mountainous areas, thereby promoting the intelligent manufacturing, quality improvement, application promotion, and thus full-process upgrading of agricultural machinery.

UAV remote sensing technology has been widely studied in agricultural information acquisition, which converts crop condition into digital signals through mounted sensors for precision farming. In the context of farmland mapping, quadruped robots, equipped with LiDAR technology, demonstrate their capability to not only gather phenotypic data of plants but also play a pivotal role in farmland navigation. It is a convenient and efficient intelligent solution to combine the use of drones and quadruped robots for farmland inspection. This paper discusses the application of UAV remote sensing in agriculture, especially growth monitoring. Information processing technologies and mapping for robots are also discussed. Finally, some demonstrations using a Unitree robot and a DJI drone are carried out in pineapple field to illustrate their capabilities in growth monitoring, farmland inspection and mapping.

Sound-based monitoring provides a non-contact and scalable solution for poultry health management. However, existing research has primarily concentrated on respiratory diseases that exhibit pronounced acoustic signatures. In contrast, non-respiratory diseases such as pullorum disease induce only subtle vocal changes, making soundbased detection particularly challenging and insufficiently explored. This study investigates the feasibility of identifying pullorum disease through the analysis of chicken vocalizations, thereby extending the application of acoustic monitoring beyond conventional respiratory conditions. Acoustic data were collected from a sample of 50 chickens, comprising 25 infected and 25 healthy individuals. A hybrid denoising strategy combining least mean square error estimation and spectral subtraction was employed to suppress environmental noise. Mel spectrograms were extracted to characterize disease-related spectral-temporal patterns. A hierarchical pyramid fusion residual network (CsResNet-PF) was proposed to model these subtle pathological variations. In addition, feature-level fusion of Mel spectrograms features and Mel-frequency cepstral coefficients (MFCC), a decisionlevel fusion, integrating traditional classifier with deep feature representations, were employed to enhance robustness and refine classification boundaries. Comparative experiments demonstrate that CsResNet-PF outperforms six representative deep learning models, achieving an accuracy of 94.75% and an F1-score of 94.76%. Notably, the proposed framework exhibits strong early-stage detection capability, with accuracies of 88.00%, 93.67%, and 94.00% on post-infection days 2, 4, and 8, respectively. These results confirm the feasibility of sound-based detection for pullorum disease and suggest that joint feature-level and decision-level fusion can provide complementary benefits for early poultry disease monitoring.

To address the challenges of limited detection range, unstable tracking, and object flickering caused by ground interference from uneven farmland, inaccurate visual masks, and sparse LiDAR point clouds at mid-to-long distances, this study constructs a multi-sensor fusion perception system based on 3D LiDAR and cameras. Furthermore, it proposes an obstacle detection method incorporating cascaded ground filtering, multimodal feature fusion, and temporal tracking. The method first utilizes a cascaded algorithm combining Random Sample Consensus (RANSAC) and Cloth Simulation Filter (CSF) to address the challenge of ground clutter filtering in uneven terrain. To account for the varying reliability of sensor modalities at different distances, a matching strategy based on hybrid cost is developed. This strategy enables a robust association between LiDAR clusters and visual targets by adaptively weighting the precise boundary information from visual masks and the stable positional features from detection bounding boxes. A Kalman filter is subsequently integrated to impose temporal smoothness on the fusion results and eliminate long-range object flickering. Experimental results demonstrate that the proposed method reduces the ground false positive rate to 0.3% and achieves an overall average precision of 91.4%. Regarding mid-to-long distance detection, in contrast to the baseline method where the recall drops sharply beyond 35 m, the proposed method achieves an average recall of 98.2% within the entire 0-45 m range. Furthermore, accuracy (MOTA) and tracking continuity (IDF1) are improved to 92.93% and 98.03%, respectively. Therefore, the proposed method satisfies the requirements for high precision, high recall, and stable environmental perception in complex farmland scenarios.

To ensure the quality of fresh corn and improve production line efficiency, automating the identification and detection of deficiencies and diseases in corn grains is essential. This study addresses the limitations and accuracy issues among contemporary techniques in real-world production settings by developing a corn detection model known as Corn-Net, which utilizes semantic segmentation. By imitating the UNet architecture, Corn-Net incorporates a multi-scale strip convolutional attention block (MSCAB) into its encoder–decoder structure, significantly enhancing feature extraction for various sizes of corn kernels, deficiencies, and diseased areas. To mitigate the potential computational costs and efficiency reductions associated with large-scale convolution kernels, MSCAB adopts a strip convolution kernel design of K×1+1×K, effectively achieving a receptive field effect similar to K×K. Experimental results demonstrate that Corn-Net, utilizing the K×1+1×K design, achieves an average segmentation rate of 83.03%, an average Dice coefficient of 80.22%, and an average accuracy of 90.82%, surpassing vanilla K×K designs. Furthermore, this approach reduces the model latency by 41. 09% and increases the frame rate, thus improving the operational efficiency compared to K×K designs. Corn-Net exhibits superior segmentation performance on a self-compiled fresh corn dataset compared to existing models. Incorporating Corn-Net into a unified hardware and software system offers a practical automated solution to sorting corn grades, marking a significant advance in automating and enhancing agricultural production processes.

Autonomous farming systems depend on close alignment between crop traits, cultivation practices and field machinery. Pineapple, a major tropical fruit grown in large-scale fields in southern China, offers a useful case for assessing autonomous fruit production. Drawing on more than 130 references overall, with a primary focus on publications from 2020 to 2025, this review evaluates full-process mechanization for pineapple production across land preparation, planting, field management and harvesting. Mechanization has progressed in land preparation, planting and field management through integrated tillage, mechanized planting and tractor-based or drone-based spraying, but harvesting remains the least mature stage. Semi-automatic harvesting devices improve efficiency by 16%–26% compared with manual operations, whereas fully automatic mechanisms report 82%–85% harvesting rates. In current pineapple recognition and detection studies, most deep-learning-based detection methods report average precision of 80.30%–97.9%, although occlusion, variable lighting and dense planting still limit robustness. Varietal assessment indicates that Tainung No. 17, Tainung No. 21 and Tainung No. 23 are suitable machinable candidates under plant heights of 75-85 cm and single-fruit weights of 1.1-1.5 kg. Future work should strengthen machinery-agronomy integration, intelligent sensing and control, modular equipment design and generalized deployment.

ABSTRACT Mobile wheel‐legged robots exhibiting mobility, stability and reliability have garnered heightened research attention in demanding real‐world scenarios, especially in material transport, emergency response and space exploration. The kinematics model merely delineates the geometric relationship of the controlled objective, disregarding force feedback. This study investigates model predictive trajectory tracking control utilising the robot dynamic model (DRMPC) in the context of unpredictable interactions. The predictive tracking controller for the wheel‐legged robot is introduced in the context of position tracking. A dynamic approximator is employed to address the uncertain interactions in the tracking process. Ultimately, co‐simulation and empirical tests are conducted to demonstrate the efficacy of the devised control methodology, which achieves high precision and dependable robustness. This work can elucidate the technical and practical oversight of autonomous movement in complicated environments and enhance the manoeuverability and flexibility.

Anomaly detection in agricultural environments is crucial for autonomous systems to manage dynamic challenges such as occlusion, lighting variations, and soil texture inconsistencies, which introduce significant uncertainty. To address these challenges, this study investigates ARNet, an agriculture visual reasoning network incorporating diffusion models, which enables the extraction and restoration of missing or damaged image regions. The network performs sequence modeling over images via a contextual spatial module and, within the deep supervision mechanism, leverages diffusion modeling to ensure smooth propagation of critical information, proactively identify and mitigate environmental disturbances, thereby isolating corrupted visual cues prior to path reasoning to ensure robust navigation decisions while accelerating model convergence. The GeoAxis module converts segmentation masks into accurate navigation paths, facilitating real-world deployment. Experimental results show that ARNet achieves 88.24% accuracy, with a 2.35 degrees navigation angle error and a compact 0.5M parameter size. Despite its small footprint, it outperforms leading segmentation models. Under low-light conditions, the line error is 4.53 degrees, and in densely planted regions, it is 3.17 degrees, demonstrating its robustness and adaptability across different environments.

To address issues such as significant communication delay fluctuations and low task scheduling efficiency in the coordinated operation of multiple intelligent agricultural machines on large-scale farms, this study proposes a distributed control node-driven multi-machine coordination control method for intelligent agricultural machinery. Taking rice harvesting and grain transportation coordination as the research scenario, the method employs a hybrid low-latency communication network model based on edge computing to establish a distributed control node-driven agricultural machinery coordination system. By integrating game theory and partially observable Markov decision models, it achieves global profit optimization. This approach constructs a hybrid low-latency communication network model based on edge computing to establish a distributed control node-driven agricultural machinery cooperative operation system. By integrating game theory and partially observable Markov decision models, it achieves global benefit optimization. A dynamic priority scheduling algorithm is designed using a proximity policy optimization algorithm, enabling global optimization of operation time and enhanced stability of the cluster control system in large-scale agricultural environments. To validate the proposed method's effectiveness, simulations and field trials were conducted. Simulation results demonstrate that by dynamically adjusting machinery priorities, the system's optimal gap is reduced from 4.8% to 1.1%. Field tests further confirmed that the introduction of the dynamic priority mechanism increased the system benefit ratio from 87.9% to 92.1%, while maintaining stable operational performance even during equipment failure and recovery scenarios. These research findings provide technical support for addressing the challenge of efficient cooperative control in intelligent agricultural machinery clusters within complex agricultural environments, effectively enhancing stability.

The hard bottom layer in paddy fields significantly impacts the driving stability, operational quality, and efficiency of agricultural machinery. Continuously improving the precision and efficiency of unmanned, precision operations for paddy field machinery is essential for realizing unmanned smart rice farms. Addressing the unclear influence patterns of hard bottom contours on typical scenarios of agricultural machinery motion and posture changes, this paper employs a rice transplanter chassis equipped with GNSS and AHRS. It proposes methods for acquiring motion state information and hard bottom contour data during agricultural operations, establishing motion state expression models for key points on the machinery antenna, bottom of the wheel, and rear axle center. A correlation analysis method between motion state and hard bottom contour parameters was established, revealing the influence mechanisms of typical hard bottom contours on machinery trajectory deviation, attitude response, and wheel trapping. Results indicate that hard bottom contour height and local roughness exert extremely significant effects on agricultural machinery heading deviation and lateral movement. Heading variation positively correlates with ridge height and negatively with wheel diameter. The constructed mathematical model for heading variation based on hard bottom contour height difference and wheel diameter achieves a coefficient of determination R2 of 0.92. The roll attitude variation in agricultural machinery is primarily influenced by the terrain characteristics encountered by rear wheels. A theoretical model was developed for the offset displacement of the antenna position relative to the horizontal plane during roll motion. The accuracy of lateral deviation detection using the posture-corrected rear axle center and bottom of the wheel center improved by 40.7% and 39.0%, respectively, compared to direct measurement using the positioning antenna. During typical vehicle-trapping events, a segmented discrimination function for trapping states is developed when the terrain profile steeply declines within 5 s and roughness increases from 0.008 to 0.012. This method for analyzing how hard bottom terrain contours affect the position and attitude changes in agricultural machinery provides theoretical foundations and technical support for designing wheeled agricultural robots, path-tracking control for unmanned precision operations, and vehicle-trapping early warning systems. It holds significant importance for enhancing the intelligence and operational efficiency of paddy field machinery.

To address the problems of low picking efficiency, high pot-breaking rate, and poor stability in the pick-up and drop mechanisms of existing automatic vegetable transplanters, this paper describes the design of a whole-row mechanical two-claw seedling-picking mechanism. A “reciprocating common-rail”-type seedling picking trajectory is introduced. A whole-row mechanical two-claw seedling-picking mechanism composed of a planetary gear-linkage mechanism and a gripping mechanism is designed. The coordination method between seedling feeding and seedling picking is determined, and the parameters of the supporting seedling-feeding mechanism are established, completing the design and modelling of key parameters. Through single-factor simulation experiments, the optimal combination of key parameter levels for the seedling-picking claws was selected. A three-factor three-level orthogonal experiment was conducted to determine the optimal combination of working parameters. Seedling-picking performance tests were carried out on peppers, tomatoes, and broccoli. The test results meet relevant national standards, indicating that the pick-up and drop mechanism exhibits strong stability and good versatility, fulfilling the design requirements. This study lays a solid foundation for the design of seedling-picking mechanisms in automatic transplanters.

Accurate identification of pineapple maturity is of paramount significance for enhancing its market value and consumer satisfaction, and it also aids fruit growers in pinpointing the optimal harvesting time. However, in real-world cultivation settings, fluctuations in environmental factors such as light, temperature, and humidity give rise to high variability in the maturity characteristics of pineapple fruits. This results in an extremely limited amount of data for certain maturity stages of pineapple and, consequently, a highly uneven distribution of data across different maturity stages. These factors collectively impede the generalization capability of traditional convolutional neural network-based object detection. To enhance the accuracy and generalization performance of pineapple maturity classification algorithms, this paper introduces an improved Generative Adversarial Network (GAN) model, namely the Pineapple Maturity Transformation GAN (PMT-GAN). By optimizing the generator architecture and incorporating a multiscale feature extraction module, the model’s performance in handling substantial geometric transformations is significantly bolstered. Additionally, the integration of the Swin Transformer module further augments the model’s explicit semantic modeling capabilities. Moreover, the adoption of the SSIM as the cycle consistency loss effectively preserves the structural coherence of images. Experimental results demonstrate that the optimized cycle consistency loss is markedly lower than the conventional loss, with reduced oscillation amplitude. Through UMAP and Grad-CAM visualization techniques, a high degree of similarity between original and generated pineapple images is observed. Compared to the original CycleGAN network, PMT-GAN achieves a maximum reduction of 30.43 percentage points in FID and 38.29 percentage points in KID. Furthermore, the network exhibits commendable generalization performance in maturity transformation experiments with other fruit species. Ultimately, the dataset size for the four maturity stages of pineapple is successfully expanded, with sample numbers becoming more balanced across stages (Stage 1: 1787 images, Stage 2: 1847 images, Stage 3: 1848 images, Stage 4: 1825 images). Detection network experiments based on the expanded dataset indicate that the expanded dataset outperforms the original dataset in key performance metrics such as precision, recall, and mAP. Research findings confirm that PMT-GAN effectively enhances the generation of pineapple maturity data, possesses robust generalization capabilities, and generates data that better serve detection networks. This provides an important methodological reference for the fine-grained classification of fruit maturity and the generation and recognition of relevant data.

Vegetable mechanized transplanting is a key link bridging industrial seedling raising and field cultivation, whose technical level directly determines operating efficiency and planting standardization. Despite its importance, current transplanting systems still struggle with instability and limited coordination between modules. This review adopts a systematic literature analysis methodology, covering core databases including Web of Science, Scopus, CNKI, and CAB Abstracts. In response to prominent issues in current transplanting equipment, such as continuous seedling supply, low-damage seedling picking, synchronization of conveying and planting actions, and adaptability to high-speed operation, this paper systematically reviews and evaluates the latest research progress in related key technologies worldwide. From the perspective of kinematic chain coupling, the transplanting process is deconstructed into four core stages: “seedling supply—seedling picking—seedling delivery—seedling planting,” with a focus on analyzing the temporal coordination, spatial constraints, state transitions, and their dynamic coupling relationships within the machine-seedling-soil system. Research indicates that vegetable transplanting technology is evolving from localized mechanism optimization toward whole-process collaborative design and system stability control, with typical high-speed operation efficiency reaching 60–140 plants per minute per row. However, significant challenges remain in low-damage seedling picking and planting at high speeds, adaptability to diverse varieties and seedling states, online perception and real-time error correction, as well as engineering reliability. The seedling damage rate under high-speed operation exceeds 8% in most existing equipment, and the planting upright rate drops by more than 5% when the operating speed increases from 60 plants/min to 120 plants/min. Future research should prioritize multi-stage collaborative optimization design, in-depth investigation of machine-seedling-soil interaction mechanisms, innovation in intelligent perception and precise control strategies, and the development of modular, low-cost, and high-performance transplanting equipment. These efforts will drive vegetable mechanized transplanting technology toward greater intelligence, efficiency, and versatility.

In multi-machine operations on modern unmanned farms, local progress deviations can readily propagate into fleet-wide coordination imbalance, while conventional control strategies struggle to reconcile operational efficiency with physical hardware constraints. To address this problem, an event-periodic hybrid-driven cooperative control method for agricultural machinery fleets is proposed. A global performance index integrating the accumulated deviation cost and control-input cost is formulated, with operating speed, operation quality, and engine power imposed as physical bounds of the optimal control problem. For recoverable routine disturbances, a receding horizon control (RHC) strategy adjusts the control input online. For unrecoverable severe failures, a mixed-integer linear programming (MILP)-based dynamic scheduling model is embedded in the RHC framework to couple macroscopic task scheduling with low-level control. Simulation and field experiments were conducted using Yanmar YR60D fully electronically controlled rice transplanters. The results showed that, under multiple random disturbances, the proposed method reduced the simulated completion time by 4.8% compared with conventional PI control while maintaining the peak engine power within the rated safe range. In the field test, the proposed method completed the same predefined task 234 s earlier than PI control, corresponding to a 9.06% reduction in completion time. In the simulated severe single-machine failure scenario, dynamic scheduling and residual-task reallocation were successfully performed, and the progress deviations of the healthy machines from their updated global trajectories decreased to within 0.1% in 550 and 516 s, respectively. The proposed method improves progress-recovery performance under routine disturbances and enables task reconfiguration and coordinated recovery following a machine failure while satisfying the constraints on operating speed, operation quality, and engine power.

The trajectory of information science for agricultural equipment technology is undergoing a profound paradigm shift, moving beyond the mere substitution of physical labor toward the emulation of human cognitive effort. Although contemporary digital automation, characterized by deterministic control logic and high-definition mapping, has established a robust baseline in engineered and structured settings like the northern plains, it encounters severe bottlenecks in unstructured scenarios. Specifically, in the fragmented and unstructured hilly regions of southern China, rule-based systems are inadequate for understanding complex, time-varying environmental dynamics, leading to fragile decision-making and severe model failures in uncertain environments. Furthermore, engineering-centric interaction designs impose significant cognitive loads for average farmers, hindering their ability to properly employ advanced agricultural technologies and adapt to the changing requirements of contemporary farming operations. To overcome the limitations of current technological paradigms, this review systematically examines the transition to a new technical direction, cognitive intelligence driven by neural dynamics. We investigate how neural dynamics with emerging technologies, specifically Vision-Language Models (VLMs), Large Language Model (LLM) agents, and embodied AI, empower agricultural machinery with open-vocabulary semantic perception, chain-of-thought intent reasoning, and resilient, learning-based neuro-dynamic control. We advocate for a revolutionary transition to an autonomous cognitive ecosystem founded on the principles of brain dynamics to establish a robust and accessible framework. Finally, we delineate a comprehensive roadmap for building an autonomous cognitive ecosystem, emphasizing continual learning for spatiotemporal adaptation, swarm neuro-dynamics for decentralized collaboration, neuromorphic computing for edge-level efficiency, and dynamic intent inference for human-agent symbiosis. Ultimately, this review provides a framework for democratizing next-generation smart agricultural equipment, enabling safe, robust, and intent-driven operations in unstructured environments.

Detecting the daily behavior of broiler chickens allows early detection of irregular activity patterns and, thus, problems in the flock. In an attempt to resolve the problems of the slow detection speed, low accuracy, and poor generalization ability of traditional detection models in the actual breeding environment, we propose a chicken behavior detection method called FCBD-DETR (Faster Chicken Behavior Detection Transformer). The FasterNet network based on partial convolution (PConv) was used to replace the Resnet18 backbone network to reduce the computational complexity of the model and to improve the speed of model detection. In addition, we propose a new cross-scale feature fusion network to optimize the neck network of the original model. These improvements led to a 78% decrease in the number of parameters and a 68% decrease in GFLOPs. The experimental results show that the proposed model is superior to the traditional network in the speed, accuracy and generalization ability of broiler behavior detection. (1) The detection speed is improved from 49.5 frames per second to 68.5 frames per second, which is 22.6 frames and 10.9 frames higher than Yolov7 and Yolov8, respectively. (2) mAP0.5 reaches 99.4%, and MAP0.5:0.95 increases from 84.9 to 88.4%. (3) Combined with the multi-target tracking algorithm, the chicken flock counting, behavior recognition, and individual tracking tasks are successfully realized.