To achieve precise identification, length estimation, and pruning point localization for robotic tomato pruning, this study presents a method for branch length recognition and pruning point localization based on an improved YOLOv8 model. The proposed YOLOv8n-CE integrates the CBAM attention mechanism to enhance the model's focus on critical branch features, and replaces the original loss with the EIOU loss to improve bounding box regression accuracy and convergence speed. Instance segmentation is subsequently performed within the detection boxes to obtain the main stem mask and extract the upper and lower endpoints of the stem. The pixel coordinates of these endpoints are transformed into three-dimensional camera coordinates to compute the stem length, and the actual branch length is derived by combining the pixel height ratio of the bounding box with the segmentation mask. The pruning point is localized along the central line of the main stem, 1.5 cm above the lower endpoint, ensuring reduced pathogen intrusion and faster wound healing. Experimental results demonstrate that the YOLOv8n-CE model improves mAP50 by 1.7 percentage points compared with YOLOv8n. For branch length measurement, the method achieves an R2 of 0.929, an MAE of 0.411 cm, and an RMSE of 0.494 cm. The pruning point localization success rate reaches 92%, with a mean absolute error of 0.247 cm. These results verify that the proposed approach meets the accuracy requirements for tomato branch measurement and pruning point localization, providing a reliable theoretical and technical foundation for robotic pruning applications.

In recent years, the mechanization of open-field white radish production in China has advanced significantly. Nevertheless, fully mechanical operations still encounter critical challenges in the planting-to-harvesting stage, particularly insufficient cross-stage adaptability and poor coordination of operational parameters. To overcome these limitations, taking open-field white radish as an example, a hybrid planting-harvesting consistency tuning framework integrated with a Large Language Model (LLM, and Qwen3-8B as the base model in this work) and several Quantitative Models (QMs) is proposed. Retrieval Augmented Generation (RAG) mechanism is used for LLM tuning, and a Consistency-Guided RAG (CG-RAG) architecture is proposed to reduce the hallucinations generated by the LLM. To provide the support for the feature extraction through satellite maps, an improved Unet model EA-MSAtt-UNet is proposed for high-precision image segmentation and feature extraction. All the information provided by the fine tuned CG-RAG LLM are delivered to four QMs through our carefully designed interface, and finally the consistency decisions are provided by the designed four QMs-a ridge-line planning model, a machinery-agronomy matching model, a planting-harvesting consistency model, and a maximum profit model-strictly constrained by the knowledge by LLM outputs. This hybrid mechanism and framework substantially enhanced decision reliability and mitigates hallucination risks inherent to standalone LLMs. Experimental results demonstrate that the CG-RAG LLM delivers superior performance on Qwen3-8B, attaining an LLM-Metric of 95.12 +/- 0.88 and an Expert Acceptability Rate (EAR) of 91/100. Following integration of the QMs, the system achieves scores of 38, 35, and 49 (on a 50-question test set) for machinery-agronomy executability, planting-harvesting consistency, and profit calculation accuracy, respectively. A seeding-stage field-based calculation example further demonstrates the preliminary applicability of the framework for field identification, ridge-line planning, machinery recommendation, and projected profit estimation. Historical comparison with previous-year field data under similar conditions indicates an approximate 3% error level in the calculation chain, providing a reference for its practical reliability. It indicates that the proposed LLM-QM fusion approach provides robust technical support for collaborative planting-harvesting decision-making, paving the way for more reliable and efficient mechanical production systems in open-field white radish cultivation.

Abstract Traditional greenhouse cleaning methods are labor-intensive, prone to human error, and inefficient, often compromising light transmittance and productivity. To address these challenges, this study proposes an autonomous robot designed to clean greenhouse roofs efficiently and reliably. The robot features an integrated cleaning system with adjustable brushes, wipers, and water sprinklers, ensuring optimal performance and significantly improving light transmittance. Powered by a 500 W PV system, it utilizes electric wheels for smooth, stable movement and incorporates a replaceable brush-wiper mechanism for enhancing durability and maintenance efficiency. The design process involved SolidWorks modeling for mass properties, CFD simulations with the k-ε turbulence model to evaluate wind load conditions, and ANSYS structural analysis to confirm durability under extreme wind speeds of up to 126 km/h (ten times greater than normal conditions). Structural tested at different robot’s rotational speeds 25 rpm and 50 rpm confirmed optimal performance at 25 rpm, balancing cleaning efficiency and long-term durability. Additionally, the robot incorporates advanced control unit with sensors for autonomous operation, real-time light transmission monitoring, and navigation capabilities, distinguishing it from traditional manual or semi-automated methods. The results demonstrated robust performance in extreme conditions, surpassing existing systems limited to standard weather. The robot’s performance is limited by speed (0.35 m/s), battery life, roof complexity, maintenance, adaptability, and cost, indicating areas for improvement. Future developments will integrate AI for autonomous decision-making, GPS for precise navigation, and a smart cleaning system to optimize performance based on real-time data, further reducing maintenance costs and ensuring optimal greenhouse lighting.

Abstract The conflict between operational efficiency and mechanical damage to tubers that arises during mechanical potato harvesting has become increasingly pronounced. This study investigated collision damage to potatoes using numerical simulation, and a damage prediction model was developed. We first conducted sampling and parameter determination through laboratory experiments, and a high-fidelity 3D geometric model was then reconstructed via reverse engineering. Following this, a multi-factor explicit dynamic simulation scenario was established in which the effects of the mass of the tuber, drop height, and impact angle on the damage characteristics were systematically analysed. Finally, a second-order response surface regression model incorporating three-factor interaction effects was developed using response surface methodology, based on the simulation data. Experimental results demonstrated that the quadratic response surface model achieved a high predictive accuracy for both the collision force and area of damage (R2 > 0.92), with prediction errors ranging from 1.54% to 25.22%. The optimal combination of parameters (a tuber mass of 320 g, an impact angle of 30°, and a drop height of 36.8 cm) was obtained through multi-objective optimisation, and reduced the collision force and area of bruising by 26.14% and 50.83%, respectively. The proposed damage prediction model was validated as reasonable and effective, and provides a theoretical basis for low-damage structural design and optimisation of separation systems in potato harvesters.

Accurate weed stem localization with machine vision and Neural Network (NN) is a key technology of modern spot weed control, such as precise herbicide spraying and laser weeding. However, weed species are diverse, making it impractical to train a particular neural network for each type. To address the challenge, by taking weed stem localization in carrot field as example, we proposed a weed stem dual-stage localization method by crop exclusion, avoiding training particular NN model for each weed species. On the first stage, a baseline model YOLOv8 is enhanced by integrating ShuffleNet Version 2(ShuffleNetV2) and Cross-Stage Feature Fusion with Enhanced Module(C2f_EM) into the backbone, and a Coordinate Attention module (CA) and a bounding box regression loss function, Wise-IoU (WIoU), are also introduced to achieve model light-weight and accuracy improvement. The improved model is employed to segmentate crop seedlings from images, with setting the pixel values within identification box to zero. On the second stage, A green contour extraction algorithm, combining Hue-Saturation-Value (HSV) color space filtering and morphological operations, is proposed to eliminate soil background and indirectly segment the weed regions. Regional thresholding is applied to further reduce noise. Principal Component Analysis (PCA) is then employed to extract the main axis of each weed. The stem localization is estimated by projecting the image moment-based centroid onto the principal axis. With the proposed Dual-Stage Localization, experimental results demonstrate that the optimized YOLOv8 model achieves a mean Average Precision at IoU=0.50 (mAP@50) of 92.9%, with a 66.6% reduction in parameters, a 57.3% decrease in computational complexity, and a 3.2% increase in Precision. The green contour extraction algorithm yields an MPA (Mean Pixel Accuracy) of 98.4% for soil background segmentation. The average weed stem detection rate reaches 92.44% within a tolerance of 15 pixels (5.04 mm), and the mean Euclidean distance between predicted and ground truth centers is 10.4 pixels (3.52 mm). The results verified the proposed method for stem localization in spot weed control is theoretically grounded, and the inherent limitations of conventional multi-class weed recognition approaches, including their annotation expenses and generalization constraints, are overcome.

In order to solve the problem of low orientation adjustment success rate and unsatisfactory orientation effect in garlic mechanical seeding, a garlic seed posture detection and adjustment device for garlic was designed. The device was developed to recognize garlic posture and control garlic posture through video acquisition, target detection and garlic directional motion control, which consisted of a posture adjustment system and an automatic control system. The posture adjustment device was designed to finish garlic bud orientation change, which consisted of a rotating plate, a shell, a conveyor pipe, a garlic funnel, a seed tube, and a fixed plate. The diameter of the conveyor pipe was determined based on the morphological parameters of the garlic seeds. Based on the mechanical analysis of garlic seeds in the conveyor pipe, the length and inclination of the conveyor pipe were determined. The automatic control system was designed to execute and control camera shooting, image processing and target detection, input action signal to the controller and receive feedback of action signal. The improved SSD algorithm was adopted to improve the identification accuracy of garlic posture detection. After the key components of the device were designed, the garlic seed posture detection and adjustment device test bench was built. The total time for garlic seed scale bud orientation detection, conveyance, and adjustment is 0.81 s. After adjustment, the upward rate of garlic buds was greater than or equal to 98.6%, and the orientation time of garlic buds adjustment was 0.5 s. The design of the device significantly improves the efficiency and accuracy of garlic seed orientation adjustment, which meets the requirements of garlic mechanical seeding.

Accurate prediction of greenhouse temperature and humidity variations was essential for optimizing crop growth conditions. To improve prediction accuracy under complex nonlinear environments, this study proposed an Improved Sparrow Search Algorithm-optimized Bidirectional Long Short-Term Memory model (ISSA-BILSTM). By introducing Tent chaotic mapping, an adaptive discoverer ratio mechanism, and a L & eacute;vy flight disturbance strategy, the proposed approach enhanced population diversity, balanced global exploration and local exploitation, and improved the convergence stability of hyperparameter optimization. Validation results showed that the proposed model achieved an R2 of 0.9777 (MAE = 0.0159) for temperature prediction and an R2 of 0.9762 (MAE = 0.0213) for humidity prediction, outperforming standard BILSTM and SSA-LSTM models. The proposed model enabled accurate short-term prediction of temperature and humidity, providing effective support for intelligent environmental regulation and contributing to reduced energy consumption and production costs.

To improve the light environment of asymmetric large-span externally insulated plastic greenhouses, a solar radiation model that considers the projection path equations of the insulation quilts and validated the model was established. The model was employed to investigate the impact of different heights, spans, and north lighting projection lengths on the greenhouses’ light environment. The results revealed that ground radiation interception, a key component of winter lighting, was most influenced by height, followed by span, and least influenced by the projection length of the north lighting roof. Additionally, ground radiation spatial uniformity was most affected by height, followed by the projection length of the north lighting roof, and least influenced by span. The optimization objectives for solar radiation were set to maximize solar radiation interception and minimize the coefficient of variation. The optimal structural parameters for the asymmetric large-span externally insulated plastic greenhouse were determined using the NSGA-II method and the entropy weight-TOPSIS method: the height of 6.97 m, and the projection length of north lighting roof is 7.44 m for a greenhouse with a span of 20.00 m. Compared to the initial greenhouse, the optimized design enhances both radiation interception performance and ensures uniform light distribution. These results offer valuable theoretical guidance for greenhouse construction.

Accurate prediction of environmental changes in Agaricus bisporus cultivation is essential for better managing climatic conditions within mushroom houses, ultimately enhancing the yield and quality of Agaricus bisporus. However, traditional control systems for Agaricus bisporus production environments can only monitor the current conditions and lack the ability to predict environmental changes, leading to issues such as delayed feedback on environmental data and the effectiveness of control measures. In response to these challenges, this study establishes a temperature and humidity prediction model based on the DOA-BP algorithm. Experimental results demonstrate that the DOA optimization algorithm exhibits strong global search capabilities. By rapidly searching for optimal weights and biases, it overcomes the drawback of the BP neural network getting stuck in local minima, accelerates network convergence, and improves the performance of the BP neural network. The MAE values for temperature and humidity prediction inside the mushroom house are 0.021 and 0.013, respectively. The RMSE values are 0.044 and 0.038, respectively, and the R2 values are 0.976 and 0.968, respectively. Through validation, the DOA-BP temperature and humidity prediction model proposed in this study accurately predicts the temperature and humidity inside mushroom houses. This model can enhance environmental control for cultivation, optimize resource utilization, and reduce production costs effectively.

Color change is the most obvious characteristic of the tomato ripening stage and an important indicator of the tomato ripening condition, which directly affects the commodity value of tomato. To visualize the color change of tomato fruit during the mature stage, this paper proposes a gated recurrent unit network with an encoder–decoder structure. This structure dynamically simulates the growth and development of tomatoes using time-dependent lines, incorporating real-time information such as tomato color and shape. Firstly, the .json file was converted into a mask.png file, the tomato mask was extracted, and the tomato was separated from the complex background environment, thus successfully constructing the tomato growth and development dataset. The experimental results showed that for the gated recurrent unit network with the encoder–decoder structure proposed, when the hidden layer number was 1 and hidden layer number was 512, a high consistency and similarity between the model predicted image sequence and the actual growth and development image sequence was realized, and the structural similarity index measure was 0.746. It was proved that when the average temperature was 24.93 °C, the average soil temperature was 24.06 °C, and the average light intensity was 11.26 Klux, the environment was the most suitable for tomato growth. The environmental data-driven tomato growth model was constructed to explore the growth status of tomato under different environmental conditions, and thus, to understand the growth status of tomato in time. This study provides a theoretical foundation for determining the optimal greenhouse environmental conditions to achieve tomato maturity and it offers recommendations for investigating the growth cycle of tomatoes, as well as technical assistance for standardized cultivation in solar greenhouses.

Garlic is a versatile crop of high economic value, with an increase in the growth of the cultivation scale. Planting garlic with bottom-side (root) down and pointy-tip (clove) up is essential in agriculture because the orientation of the clove significantly affects garlic quality. In this study, we propose a clove orientation recognition technique based on capacitance sensing technology, where clove states are determined by utilizing characteristic differences in capacitance variations associated with different orientations of cloves. First, we applied Maxwell simulations to obtain capacitance variation during the falling process. Second, we conducted field experiments using the capacitive sensing device to obtain the capacitance variations during the falling of garlic in an unstructured environment. Third, the continuous capacitance signal collected during the field experiment was segmented into short-term feature signals containing individual garlic fallings. Finally, Long Short-Term Memory (LSTM) and Frequency-Residual-GoogLeNet (F-Res-GoogLeNet) deep models were trained to recognize garlic falling states. Our best model achieved an accuracy of 96.75%, which meets the agricultural requirements for garlic cultivation. This study demonstrates that monitoring capacitance data in field environments can identify garlic cloves’ orientation, eventually enhancing garlic’s final yield, quality, and economic benefits.

Abstract The rapid and efficient removal of weeds is currently a research hotspot. With the integration of robotics and automation technology into agricultural production, intelligent field weeding robots have emerged. The development status of weeding robots is overviewed based on bibliometric and scientific mapping methods. Secondly, the two key technologies of weeding robots are summarized. Then, the research progress of precision spraying weeding robots, mechanical weeding robots, and thermal weeding robots with laser devices, categorized by their weeding methods, is reviewed. Finally, a summary and an outlook on the future development trends of intelligent field weeding robots are provided, aiming to offer a reference for further promoting the development of weeding robots.

传统设施园艺生产过程中水资源的管理方式粗放,灌溉施肥方式落后、水肥资源浪费严重,不合理的施肥方式会造成土壤中残留肥料,对环境造成污染.为提高设施作物的水肥利用效率和作物产量,本研究围绕设施精准灌溉施肥技术研究基于云平台的设施水肥一体化控制系统.采用模块化的思想,设计了以STM32为控制终端的智能水肥一体化控制系统,实时采集水肥参数,利用灌溉施肥控制算法,实行闭环反馈控制,保证水肥溶液配制的精确性;采用触摸屏组态技术,设计了人机交互系统,拥有手动、定时定量、配肥等多种操作模式;利用物联网技术,设计了水肥一体化远程监控EMCP云平台,实现了云平台间的实时通信,满足了设施作业的网络化管理和智能化控制要求.

A service evaluation model and a solution strategy for agricultural machinery maintenance services based on big data areproposed considering the characteristics of a large business volume, a diversity of service types, a wide geographical distribution and on-site service for agricultural machinery maintenance services in China. First, a maintenance order priority evaluation model is developed based onanevaluation model of quality of service (QoS), and order confirmation is followed by developing a staffing evaluation model. A fuzzy analytical hierarchy process is used to determine the weights of the evaluation indicators. The evaluation models are solved by an improved genetic-bee colony algorithm (GAABC). In the selection operation stage of the genetic algorithm, a combination of an elite retention strategy and a roulette strategy is adopted, which not only guarantees the convergence speed of the algorithm and the diversity of individuals, but also prevents the loss of good genes of individual parents after the crossover and mutation operations. The attractor is introduced at the onlooker bee stage, where the onlooker bees shrink in proportion to the attractor at the center, thereby increasing the convergence speed and algorithm development for subsequent stages, as well as developing the area at the current stage. The mutation operation is added to the artificial bee colony algorithm (ABC) according to the degree of honey source aggregation to improve the local search ability. In addition, the best fusion point assessment strategy is used to determine the switching time between the genetic algorithm (GA) and the artificial bee colony algorithm to increase the convergence speed and accuracy of the solution. Finally, the feasibility and effectiveness of the agricultural machinery maintenance service model and the solution algorithm are verified by simulation experiments. This research provides theoretical support for the decision analysis of agricultural machinery maintenance enterprises in China.

The accurate prediction of greenhouse environment variation based on the constructed prediction model is helpful to precisely regulate the crop environment, and promote the growth of fruits and vegetables. Due to the coexistence of multiple parameters, complex coupling with each other, temporality and nonlinearity of greenhouse microclimate environment, the accurate prediction model is difficult to establish. Based on above issues, a greenhouse environment prediction model was proposed based on the sparrow search algorithm（SSA） optimized-long short term memory（LSTM） neural network method, so as to realize the prediction of greenhouse environment data sequence with the Internet of things（IoT） collecting accurate multipoint environment data. The experimental results showed that the automatic parametric optimization process by SSA could deal with the time consuming problem of manual parameter selection for the LSTM model. The proposed SSA-LSTM method could lower the model training time, and the optimal parameters selection could make sure the model worked with the optimum capability. The trained SSA-LSTM model was used to predict six kinds of greenhouse environment data, including the air temperature, air humidity, soil temperature, soil humidity, CO 2 concentration, and the illumination intensity. The proposed SSA-LSTM could realize a 97.6% average prediction fit index, compared with the back-propagation network, the gated recurrent unit neural network and the LSTM, the prediction fit index was elevated by 8.1 percentage points, 4.1 percentage points and 4.3 percentage points. Therefore, the prediction accuracy of SSA-LSTM was obviously improved. The research result could provide reference for the development of greenhouse environment control strategy and deal with the lag problem of environment control.

In order to solve the problem of low recognition accuracy of cucumber canopy vine tops image caused by the high density and intermingling of the cucumber canopy vine, an image recognition model of cucumber canopy tops based on the deep learning model of improved YOLOv5 is proposed in this paper. A combination of online and offline augmentation is used to amplify the original data to ensure a balanced number of images in different types of datasets during training and testing. Considering the small leaves characteristics of the cucumber canopy vine tops, the CA mechanism module (YOLOv5-CA) is introduced into the backbone network of the YOLOv5 model to improve the recognition accuracy of the small leaf target. On this basis, the regression Loss GIOU of the target position was changed to EIOU and the cross-entropy function of confidence loss was changed to Focal Loss function (YOLOv5-CA-LS) to reduce the influence of the imbalance of difficult and easy samples on the detection results. The models before and after optimization were compared. The results show that compared with the YOLOv5 model, the accuracy, recall rate, mAP@0.5 and mAP@0.5:0.95 of the YOLOv5-CA-LS model is improved by 4.4%, 5.0%, 1.3% and 1.8% respectively. And average recognition accuracy value of the instance of the cucumber canopy tops adopted the YOLOv5-CA-LS model is 97.1% under different angle and cloudy conditions, which can meet the recognition requirement of high density and intermingling of the cucumber canopy vine tops.

In order to meet the requirements of target detection and location for intelligent tomato picking, a recognition and location algorithm based on improved YOLOv5s is proposed in this paper. The CARAFE module structure was used to optimize the upsampling algorithm of YOLOv5s, which increased the receptive field of the network while maintaining the lightweight. EIoU and Quality Focal Loss were introduced to optimize the loss function of the network, which solved the problem of reduced accuracy caused by uneven samples, and at the same time accelerated the training convergence speed and improved the regression accuracy. The improved model is denoted as YOLOv5s-CQE. Compared with YOLOv5s, the mAP_0.5 and mAP_0.5:0.95 of YOLOv5s-CQE was increased by 1.67 and 3.43 percentages points, respectively. The recognition accuracy of YOLOv5s-CQE model in the test set was 99.77%, which had increased by 2.40 percentages points compared with before improvement, and was 4.10, 6.47 and 3.03 percentages points higher than that of the lightweight networks YOLOv4-tiny, YOLOv5-Lite-e and YOLOv5-Lite-s, respectively. Compared with YOLOv7 and Faster RCNN, the recognition accuracy rate was increased by 1.69, 3.97 percentages points respectively. In order to improve the accuracy of positioning, distortion removal and ROI clipping were carried out on the obtained images. The accuracy of location was tested by laboratory positioning test and field picking test. The results showed the total average errors decreased by 6.65 mm compared with those before distortion removal, and the field positioning accuracy was improved by 6.67 percentages points. The experimental results showed that the algorithm in this paper had the advantages of high precision, fast detection speed and strong robustness, which provided a theoretical basis for intelligent tomato picking.

This paper proposes a fertilizer mixing device that combines pressure relief return and mechanical agitation to address the issues in the current water-fertilizer integration equipment related to limited fertilizer mixing methods and inconvenient irrigation pressure regulation. The device employs different mixing modes for various fertilizer forms and uses pressure relief return to adjust irrigation pressure, thereby enhancing the efficiency of water-fertilizer mixing and optimizing energy consumption. The experimental results indicate that the reflux mode is suitable for liquid-type fertilizers which are fast dissolving and easy to diffuse, and its EC value is stable at about 6.60 mS/cm, which is close to the calibrated value of 6.80 mS/cm. The stirring paddle mode compensates for the reflux mode's weak mixing effect, making it suitable for solid powder-type fertilizers' mixing operation. The EC value remains stable at approximately 8.60 mS/cm when calibrated at 8.70 mS/cm. The "stirring paddle + two-way reflux" mode demonstrates the most robust mixing effect and is suitable for mixing solid granular fertilizers. When calibrated at 8.20 mS/cm, it stabilizes at approximately the calibration value after 105 s. This research provides technical support and a theoretical basis to accomplish efficient, energy-saving, and rational application of water-fertilizer integration across diverse fertilizer forms.

Robots play a major role in solving problems, especially in the agricultural sector.The uses of agricultural automation and robotics have increased dramatically, which has solved the obstacles and complexities of fields such as agricultural operations from planting to harvesting, giving high effectiveness and efficiency in the fields.The review is divided into four sections: (i) various robotic agricultural processes, in which current robotic technologies in agriculture from planting to harvesting were investigated; (ii) types of robots used in agriculture such as aerial robots, ground robots, special robots, and multi-robot systems; (iii) a study of real applied projects that use agricultural robots, and (iv) Artificial Intelligence (AI) techniques used in agriculture.The study shows that agriculture greatly benefits from robots because they save time and effort, enhance production rates, and reduce costs.The study recommends further improvement and development of agricultural robots, and this should be achieved by creating faster processing algorithms, improving communication between platforms and robotic tools, and advanced sensor systems.