Pub Date : 2026-01-27DOI: 10.1016/j.compag.2026.111471
Liwei Yang , Ping Li , Zhongshuo Ding , Chao Sun , Fei Chen , Yijiang Zheng , Yun Ge
With the development of agricultural intelligence, mechanized harvesting of crops characterized by short picking periods and growth-pattern-dependent optimal harvest times faces dual challenges in path planning and multi-machine collaborative scheduling. To tackle the complex scenario of safflower harvesting in hilly terrains, this paper proposes a Multi-Task Assignment and Scheduling Mechanism (MTA-SM) aimed at achieving both full-coverage picking paths and time-window-constrained scheduling for multiple machines. The system consists of two major modules: for path planning, a terrain adaptation factor is introduced to improve the Coverage Path Planning (CPP) algorithm, and the turning strategy of harvesters is optimized to reduce ineffective movements and enhance operational coverage. For scheduling, a Vehicle Routing Problem with Time Windows (VRPTW) model is formulated, and an improved Ant Colony Optimization (ACO) algorithm with a dynamic pheromone updating mechanism is employed to realize coordinated scheduling among multiple machines, thereby minimizing path conflicts and idle time. Simulation results indicate that the MTA-SM system not only optimizes the operational path of a single harvester but also significantly enhances the efficiency and resource utilization of multi-machine collaboration. This provides a practical and intelligent solution for the mechanized harvesting of crops with short picking windows.
{"title":"MTA-SM: Multi-machine path planning and time-window scheduling joint optimization method in hilly safflower harvesting","authors":"Liwei Yang , Ping Li , Zhongshuo Ding , Chao Sun , Fei Chen , Yijiang Zheng , Yun Ge","doi":"10.1016/j.compag.2026.111471","DOIUrl":"10.1016/j.compag.2026.111471","url":null,"abstract":"<div><div>With the development of agricultural intelligence, mechanized harvesting of crops characterized by short picking periods and growth-pattern-dependent optimal harvest times faces dual challenges in path planning and multi-machine collaborative scheduling. To tackle the complex scenario of safflower harvesting in hilly terrains, this paper proposes a Multi-Task Assignment and Scheduling Mechanism (MTA-SM) aimed at achieving both full-coverage picking paths and time-window-constrained scheduling for multiple machines. The system consists of two major modules: for path planning, a terrain adaptation factor is introduced to improve the Coverage Path Planning (CPP) algorithm, and the turning strategy of harvesters is optimized to reduce ineffective movements and enhance operational coverage. For scheduling, a Vehicle Routing Problem with Time Windows (VRPTW) model is formulated, and an improved Ant Colony Optimization (ACO) algorithm with a dynamic pheromone updating mechanism is employed to realize coordinated scheduling among multiple machines, thereby minimizing path conflicts and idle time. Simulation results indicate that the MTA-SM system not only optimizes the operational path of a single harvester but also significantly enhances the efficiency and resource utilization of multi-machine collaboration. This provides a practical and intelligent solution for the mechanized harvesting of crops with short picking windows.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"244 ","pages":"Article 111471"},"PeriodicalIF":8.9,"publicationDate":"2026-01-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146079826","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"农林科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
This paper develops and applies a human-centric framework to design a Digital Twin (DT) by applying a people-led approach to a vineyard automation scenario. Current DT systems in agriculture often focus on technical performance, which creates usability challenges such as data overload, lack of role-specific interfaces, and reduced trust among non-technical users. The study applies Personas to represent user groups and introduces a human-centric framework for mapping tasks and decision processes. The framework makes an original contribution by demonstrating how established human-centric methods can be systematically integrated into a coherent DT development process, addressing a recognised methodological gap in the literature. The objective of this research is to evaluate how a structured, human-centric approach can improve usability, cognitive alignment, and stakeholder engagement in vineyard automation. These processes are modeled using Personas, Decision Ladders and Control Task Analysis to align system functionality with user roles and cognitive needs. The research methodology integrates Personas, ConTA, and Decision Ladders within a real-world vineyard case study. This study showcases the impact of applying a structured human-centric DT design framework on improving decision-making support, user engagement, and system efficiency in agricultural contexts. Moreover, it provides expert-informed evidence in what way human-centric methods can be operationalised in a consistent and transparent way for DT redesign. Overall, the work demonstrates how a structured, people-led approach can enhance the usability and adoption of both new and existing DT systems, offering a transferable framework with relevance beyond agriculture.
{"title":"A human-centric framework for enhancing usability in a vineyard digital twin system","authors":"Meysam Zareiee , Baixiang Zhao , Claire Palmer , Mahsa Mehrad , Yee Mey Goh , Rebecca Grant , Ella-Mae Hubbard , Jörn Mehnen , Anja Maier","doi":"10.1016/j.compag.2026.111490","DOIUrl":"10.1016/j.compag.2026.111490","url":null,"abstract":"<div><div>This paper develops and applies a human-centric framework to design a Digital Twin (DT) by applying a people-led approach to a vineyard automation scenario. Current DT systems in agriculture often focus on technical performance, which creates usability challenges such as data overload, lack of role-specific interfaces, and reduced trust among non-technical users. The study applies Personas to represent user groups and introduces a human-centric framework for mapping tasks and decision processes. The framework makes an original contribution by demonstrating how established human-centric methods can be systematically integrated into a coherent DT development process, addressing a recognised methodological gap in the literature. The objective of this research is to evaluate how a structured, human-centric approach can improve usability, cognitive alignment, and stakeholder engagement in vineyard automation. These processes are modeled using Personas, Decision Ladders and Control Task Analysis to align system functionality with user roles and cognitive needs. The research methodology integrates Personas, ConTA, and Decision Ladders within a real-world vineyard case study. This study showcases the impact of applying a structured human-centric DT design framework on improving decision-making support, user engagement, and system efficiency in agricultural contexts. Moreover, it provides expert-informed evidence in what way human-centric methods can be operationalised in a consistent and transparent way for DT redesign. Overall, the work demonstrates how a structured, people-led approach can enhance the usability and adoption of both new and existing DT systems, offering a transferable framework with relevance beyond agriculture.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"244 ","pages":"Article 111490"},"PeriodicalIF":8.9,"publicationDate":"2026-01-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146079827","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"农林科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-23DOI: 10.1016/j.compag.2026.111436
Qirui Wang , Yang Liu , Shenyao Hu , Yuting Yan , Bing Li , Hanping Mao
The accuracy of intelligent corn detasselling systems is severely compromised by low-light conditions, which degrade image quality and impede tassel recognition. To address the limitations of existing methods, such as noise amplification, detail distortion, and inadequate global illumination modeling, a Low-Light Corn Plant Image Enhancement Model (L2CP-IEM) is proposed. The core of L2CP-IEM is an innovative closed-loop dynamic gamma correction mechanism. This mechanism, guided by the discriminator’s confidence, is embedded within a residual encoder-decoder architecture, enabling adaptive illumination adjustment and stable training. By using a green cardboard calibration method, a high-quality dataset consisting of 950 paired low-light and normal-light corn images was created. Experiments on the LOL-v1 benchmark dataset demonstrate that L2CP-IEM outperforms state-of-the-art methods such as GSAD and CIDNet in terms of the SSIM (0.908) and LPIPS (0.059). Ablation studies further validate the critical roles of residual connections and the dynamic gamma correction mechanism. In practical corn tassel image tests, L2CP-IEM achieves balanced performance in terms of brightness and colour restoration, significantly enhances the reconstruction of natural textures and hierarchical details, and fully restores the confidence of the Mask R-CNN in image segmentation. By synergizing physical principles with data-driven approaches, this method significantly improves the quality of low-light images and the robustness of recognition, thus offering a reliable and efficient solution for agricultural visual automation.
{"title":"Dynamic gamma correction-guided CNN for low-light corn tassel enhancement in intelligent detasselling systems","authors":"Qirui Wang , Yang Liu , Shenyao Hu , Yuting Yan , Bing Li , Hanping Mao","doi":"10.1016/j.compag.2026.111436","DOIUrl":"10.1016/j.compag.2026.111436","url":null,"abstract":"<div><div>The accuracy of intelligent corn detasselling systems is severely compromised by low-light conditions, which degrade image quality and impede tassel recognition. To address the limitations of existing methods, such as noise amplification, detail distortion, and inadequate global illumination modeling, a Low-Light Corn Plant Image Enhancement Model (L2CP-IEM) is proposed. The core of L2CP-IEM is an innovative closed-loop dynamic gamma correction mechanism. This mechanism, guided by the discriminator’s confidence, is embedded within a residual encoder-decoder architecture, enabling adaptive illumination adjustment and stable training. By using a green cardboard calibration method, a high-quality dataset consisting of 950 paired low-light and normal-light corn images was created. Experiments on the LOL-v1 benchmark dataset demonstrate that L2CP-IEM outperforms state-of-the-art methods such as GSAD and CIDNet in terms of the SSIM (0.908) and LPIPS (0.059). Ablation studies further validate the critical roles of residual connections and the dynamic gamma correction mechanism. In practical corn tassel image tests, L2CP-IEM achieves balanced performance in terms of brightness and colour restoration, significantly enhances the reconstruction of natural textures and hierarchical details, and fully restores the confidence of the Mask R-CNN in image segmentation. By synergizing physical principles with data-driven approaches, this method significantly improves the quality of low-light images and the robustness of recognition, thus offering a reliable and efficient solution for agricultural visual automation.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"244 ","pages":"Article 111436"},"PeriodicalIF":8.9,"publicationDate":"2026-01-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146025259","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"农林科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-23DOI: 10.1016/j.compag.2026.111465
Hao Wang , Yixue Liu , Bin Sun , Juncheng Ma , Chao Liang , Xiao Yang , Renli Qi , Chaoyuan Wang
Floor fouling monitoring in pig facilities is essential for early disease detection and environmental hygiene management, as diarrheal feces indicates digestive disorders while manure accumulation directly impacts animal health and welfare. Current manual inspection methods are labor-intensive and subjective, while existing computer vision approaches suffer from unstable color features under varying lighting conditions and misclassification of background textures as fouling patterns. To address these challenges, we propose FreCANet, a frequency-aware deep learning framework that achieves multi-level fouling classification through hierarchical visual interference suppression. The method integrates three key innovations: Mask R-CNN preprocessing that eliminates pig body occlusion (improving detection recall by up to 17.21%), Frequency Dynamic Convolution that separates manure contamination features from environmental noise across different frequency bands, and Efficient Channel Attention embedded within residual connections for selective feature enhancement. Using a comprehensive dataset of 25,228 images covering seven fouling categories across the complete growth cycle, FreCANet achieved 88.31% accuracy and 0.8679 F1-Score, outperforming ResNet-152 by 2.44% and 2.93% respectively. Diarrheal feces detection reached 95.9% precision on slatted floors and 89.3% recall on solid floors, enabling reliable early warning for digestive health issues. The four-level manure contamination classification achieved 77.2–87.4% precision across fouling gradients from clean to severely soiled conditions. These results demonstrate FreCANet’s effectiveness in transforming subjective manual inspection into quantitative pen hygiene assessment for precision livestock farming applications.
{"title":"Frequency-aware deep learning for diarrheal feces and floor fouling monitoring in pig pens","authors":"Hao Wang , Yixue Liu , Bin Sun , Juncheng Ma , Chao Liang , Xiao Yang , Renli Qi , Chaoyuan Wang","doi":"10.1016/j.compag.2026.111465","DOIUrl":"10.1016/j.compag.2026.111465","url":null,"abstract":"<div><div>Floor fouling monitoring in pig facilities is essential for early disease detection and environmental hygiene management, as diarrheal feces indicates digestive disorders while manure accumulation directly impacts animal health and welfare. Current manual inspection methods are labor-intensive and subjective, while existing computer vision approaches suffer from unstable color features under varying lighting conditions and misclassification of background textures as fouling patterns. To address these challenges, we propose FreCANet, a frequency-aware deep learning framework that achieves multi-level fouling classification through hierarchical visual interference suppression. The method integrates three key innovations: Mask R-CNN preprocessing that eliminates pig body occlusion (improving detection recall by up to 17.21%), Frequency Dynamic Convolution that separates manure contamination features from environmental noise across different frequency bands, and Efficient Channel Attention embedded within residual connections for selective feature enhancement. Using a comprehensive dataset of 25,228 images covering seven fouling categories across the complete growth cycle, FreCANet achieved 88.31% accuracy and 0.8679 F1-Score, outperforming ResNet-152 by 2.44% and 2.93% respectively. Diarrheal feces detection reached 95.9% precision on slatted floors and 89.3% recall on solid floors, enabling reliable early warning for digestive health issues. The four-level manure contamination classification achieved 77.2–87.4% precision across fouling gradients from clean to severely soiled conditions. These results demonstrate FreCANet’s effectiveness in transforming subjective manual inspection into quantitative pen hygiene assessment for precision livestock farming applications.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"244 ","pages":"Article 111465"},"PeriodicalIF":8.9,"publicationDate":"2026-01-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146025153","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"农林科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-23DOI: 10.1016/j.compag.2026.111453
Li-Quan Lu , Ze-Peng Zhang , Guang-Lin Zhang , Hao-Ran Yang , Zhong-Xiang Zhu , Zheng-He Song , Zhi-Qiang Zhai , Jian-Hua Wang , Chuan-Chuan Zhang
To address the significant degradation in start-up comfort, smoothness, and safety of high-power power-shift tractors caused by variations in multiple start-up parameters, rapid vehicle state transitions, and environmental excitations during practical operations, this paper proposes a coordinated control approach consisting of a start-up condition recognition method based on spatial rule mapping and subspace partitioning, and a multi-mode model predictive control (MPC)-based start-up control strategy. First, a start-up dynamic model of the power-shift transmission system is established, and single-factor comparative simulations are conducted in the Matlab/Simulink environment to analyze the influence mechanisms of throttle opening, main/sub gearbox gear selection, road slope, and vehicle initial state on start-up time, clutch friction work, and jerk. Based on these analyses, the multi-source parameters are reduced and normalized into three dimensionless indicators, namely driver start-up intention, tractor initial state, and load state, and the three-dimensional feature space is partitioned into four subspaces according to their impact on start-up performance, enabling real-time start-up condition recognition. A multi-mode MPC controller is then constructed, and a multi-objective genetic algorithm is employed to determine the optimal control parameters for each subspace, achieving an adaptive balance among start-up rapidity, smoothness, and component wear under different operating conditions. The hardware-in-the-loop (HIL) test results indicate that, compared with conventional control methods, the proposed multi-mode MPC exhibits more stable and well-balanced overall performance under different start-up conditions. For example, in a typical flat-road start-up scenario, the maximum jerk is reduced to 46.2 m/s3, while in a slope start-up condition, the reverse travel distance is shortened to 0.178 m. These results demonstrate the effectiveness of the proposed method in improving start-up smoothness and operational safety of tractors under complex start-up conditions, and provide a basis for subsequent real-vehicle experiments and engineering application studies.
{"title":"Research on a complex start-up control strategy for power-shift tractors based on rule mapping and multi-mode model predictive control","authors":"Li-Quan Lu , Ze-Peng Zhang , Guang-Lin Zhang , Hao-Ran Yang , Zhong-Xiang Zhu , Zheng-He Song , Zhi-Qiang Zhai , Jian-Hua Wang , Chuan-Chuan Zhang","doi":"10.1016/j.compag.2026.111453","DOIUrl":"10.1016/j.compag.2026.111453","url":null,"abstract":"<div><div>To address the significant degradation in start-up comfort, smoothness, and safety of high-power power-shift tractors caused by variations in multiple start-up parameters, rapid vehicle state transitions, and environmental excitations during practical operations, this paper proposes a coordinated control approach consisting of a start-up condition recognition method based on spatial rule mapping and subspace partitioning, and a multi-mode model predictive control (MPC)-based start-up control strategy. First, a start-up dynamic model of the power-shift transmission system is established, and single-factor comparative simulations are conducted in the Matlab/Simulink environment to analyze the influence mechanisms of throttle opening, main/sub gearbox gear selection, road slope, and vehicle initial state on start-up time, clutch friction work, and jerk. Based on these analyses, the multi-source parameters are reduced and normalized into three dimensionless indicators, namely driver start-up intention, tractor initial state, and load state, and the three-dimensional feature space is partitioned into four subspaces according to their impact on start-up performance, enabling real-time start-up condition recognition. A multi-mode MPC controller is then constructed, and a multi-objective genetic algorithm is employed to determine the optimal control parameters for each subspace, achieving an adaptive balance among start-up rapidity, smoothness, and component wear under different operating conditions. The hardware-in-the-loop (HIL) test results indicate that, compared with conventional control methods, the proposed multi-mode MPC exhibits more stable and well-balanced overall performance under different start-up conditions. For example, in a typical flat-road start-up scenario, the maximum jerk is reduced to 46.2 m/s<sup>3</sup>, while in a slope start-up condition, the reverse travel distance is shortened to 0.178 m. These results demonstrate the effectiveness of the proposed method in improving start-up smoothness and operational safety of tractors under complex start-up conditions, and provide a basis for subsequent real-vehicle experiments and engineering application studies.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"244 ","pages":"Article 111453"},"PeriodicalIF":8.9,"publicationDate":"2026-01-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146025343","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"农林科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-23DOI: 10.1016/j.compag.2026.111466
Zhenyu Chen , Hanjie Dou , Changyuan Zhai , Zhichong Wang , Yuanyuan Gao , Xiu Wang
Autonomous navigation technology is playing an increasingly vital role in intelligent orchard production and management. However, the sparsely structured and feature-degraded environments of standardized orchards pose significant challenges to existing localization and navigation methods. To address these issues, this study proposes a multi-sensor fusion localization framework that integrates LiDAR, RTK-GNSS, and IMU data to overcome localization failures in degraded environments. Building on this framework, we develop an autonomous navigation system capable of robust tree-row centerline tracking and introduce a point-to-point navigation control method to achieve precise orchard operations. A 3D point cloud map and a 2D occupancy grid map are constructed using the LIO-SAM algorithm combined with a trunk-height-based point cloud extraction method. LiDAR point clouds provide measurement updates for 3D map matching, while tightly coupled RTK and IMU data supply motion estimates. A particle filter fuses these measurements to ensure reliable localization. Evaluation experiments—including map construction accuracy, localization error, navigation precision, and row-center tracking—show that the proposed multi-sensor fusion method reduces localization error by 66.27 % compared with LiDAR-only NDT matching. The row-center tracking error is 4.37 cm and the headland turning error is 20.18 cm, representing reductions of 69.86 % and 48.74 %, respectively, meeting the centerline navigation requirements for spraying. In point-to-point navigation tests, the average longitudinal and lateral errors are 0.225 m and 0.088 m, satisfying the accuracy requirements of harvesting, fertilization, and transport operations. This study provides a comprehensive solution for orchard autonomous navigation and practical techniques for intelligent orchard production in complex field environments.
{"title":"Autonomous centerline and point-to-point navigation control method based on multi-sensor fusion in degraded orchards environments","authors":"Zhenyu Chen , Hanjie Dou , Changyuan Zhai , Zhichong Wang , Yuanyuan Gao , Xiu Wang","doi":"10.1016/j.compag.2026.111466","DOIUrl":"10.1016/j.compag.2026.111466","url":null,"abstract":"<div><div>Autonomous navigation technology is playing an increasingly vital role in intelligent orchard production and management. However, the sparsely structured and feature-degraded environments of standardized orchards pose significant challenges to existing localization and navigation methods. To address these issues, this study proposes a multi-sensor fusion localization framework that integrates LiDAR, RTK-GNSS, and IMU data to overcome localization failures in degraded environments. Building on this framework, we develop an autonomous navigation system capable of robust tree-row centerline tracking and introduce a point-to-point navigation control method to achieve precise orchard operations. A 3D point cloud map and a 2D occupancy grid map are constructed using the LIO-SAM algorithm combined with a trunk-height-based point cloud extraction method. LiDAR point clouds provide measurement updates for 3D map matching, while tightly coupled RTK and IMU data supply motion estimates. A particle filter fuses these measurements to ensure reliable localization. Evaluation experiments—including map construction accuracy, localization error, navigation precision, and row-center tracking—show that the proposed multi-sensor fusion method reduces localization error by 66.27 % compared with LiDAR-only NDT matching. The row-center tracking error is 4.37 cm and the headland turning error is 20.18 cm, representing reductions of 69.86 % and 48.74 %, respectively, meeting the centerline navigation requirements for spraying. In point-to-point navigation tests, the average longitudinal and lateral errors are 0.225 m and 0.088 m, satisfying the accuracy requirements of harvesting, fertilization, and transport operations. This study provides a comprehensive solution for orchard autonomous navigation and practical techniques for intelligent orchard production in complex field environments.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"244 ","pages":"Article 111466"},"PeriodicalIF":8.9,"publicationDate":"2026-01-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146025075","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"农林科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-23DOI: 10.1016/j.compag.2026.111464
Shuai Li , Yajun An , Yuxiao Han , Yingyan Yang , Yuanda Yang , Han Li , Man Zhang
The poultry breeding industry plays a vital role in global agriculture. This study presents a compact and mobile system for synchronized image acquisition in poultry aimed at improving traditional, labor-intensive, and subjective methods for measuring individual chicken temperature through worker-carried and robot-autonomous inspection. The system uses object detection to locate the chicken and combines binocular and thermal infrared cameras to solve the problem of time and position inconsistency to accurately obtain the temperature information of the chicken. It is designed with modules for central control, image acquisition, network transmission, power management, and real-time monitoring. The central control module processes video data and connects with external systems via the network transmission module, while the power management module ensures that all components receive adequate power. The real-time monitoring module supports the display and storage of image data. In recent years, the Robot Operation System (ROS) has greatly improved software development efficiency. By employing ROS for multieye timing synchronization, the system achieves consistent data recording. In a two-day trial at the Deqingyuan Chicken Coop in Beijing, the system collected 2657 pairs of binocular and thermal infrared images, demonstrating real-time capabilities with a maximum time difference tmax of 4.97 ms, a minimum tmin of 0.66 ms, an average tmean of 1.25 ms, and a root mean square error tRMSE of 1.31 ms. Target detection tests indicated that You Only Look Once version 5 (YOLOv5) performed best, with a Precision (P) of 92.71 %, a Recall (R) of 93.91 %, mean Average Precision (mAP) of 96.32 %, and an inference speed of 49.6 ms, showing high-precision and real-time detection. Image registration tests revealed a maximum matching error Hmax of 0.86 pixels, a minimum Hmin of 0.35 pixels, and an average Hmean of 0.61 pixels. The maximum structural similarity index (SSIMmax) was 0.86, the minimum SSIMmin was 0.61, and the average SSIMmean was 0.78. High-precision target detection methods, image time synchronization methods, and spatial registration methods are used to confirm high-precision image registration and accurate target temperature measurement. This system provides a technical solution and equipment support for the quick and accurate acquisition of chicken temperature information in the poultry breeding industry.
家禽养殖业在全球农业中发挥着至关重要的作用。本研究提出了一种紧凑的、可移动的家禽同步图像采集系统，旨在改进传统的、劳动密集型的、主观的方法，通过工人携带和机器人自主检测来测量单个鸡的温度。该系统采用目标检测对鸡进行定位，并结合双目和热红外摄像机解决时间和位置不一致的问题，准确获取鸡的温度信息。它设计了中央控制、图像采集、网络传输、电源管理和实时监控等模块。中控模块负责处理视频数据，并通过网络传输模块与外部系统连接，电源管理模块负责保证各部件都能获得充足的电源。实时监控模块支持图像数据的显示和存储。近年来，机器人操作系统（ROS）极大地提高了软件开发效率。系统采用ROS进行多眼定时同步，实现了数据的一致性记录。在北京德清源鸡舍进行的为期两天的试验中，该系统采集了2657对双目和热红外图像，最大时差tmax为4.97 ms，最小tmin为0.66 ms，平均tms为1.25 ms，均方根误差tRMSE为1.31 ms。目标检测测试结果表明，YOLOv5 （You Only Look Once version 5）表现最佳，精密度（Precision）为92.71%，召回率（Recall）为93.91%，平均平均精密度（mAP）为96.32%，推理速度为49.6 ms，具有较高的检测精度和实时性。图像配准测试显示，最大匹配误差Hmax为0.86像素，最小匹配误差Hmin为0.35像素，平均匹配误差Hmean为0.61像素。最大结构相似指数（SSIMmax）为0.86，最小结构相似指数（SSIMmin）为0.61，平均SSIMmean为0.78。采用高精度目标检测方法、图像时间同步方法和空间配准方法，实现高精度图像配准和精确目标温度测量。该系统为家禽养殖业快速准确地获取鸡体温度信息提供了技术解决方案和设备支持。
{"title":"Poultry image synchronization acquisition system based on binocular and thermal infrared cameras","authors":"Shuai Li , Yajun An , Yuxiao Han , Yingyan Yang , Yuanda Yang , Han Li , Man Zhang","doi":"10.1016/j.compag.2026.111464","DOIUrl":"10.1016/j.compag.2026.111464","url":null,"abstract":"<div><div>The poultry breeding industry plays a vital role in global agriculture. This study presents a compact and mobile system for synchronized image acquisition in poultry aimed at improving traditional, labor-intensive, and subjective methods for measuring individual chicken temperature through worker-carried and robot-autonomous inspection. The system uses object detection to locate the chicken and combines binocular and thermal infrared cameras to solve the problem of time and position inconsistency to accurately obtain the temperature information of the chicken. It is designed with modules for central control, image acquisition, network transmission, power management, and real-time monitoring. The central control module processes video data and connects with external systems via the network transmission module, while the power management module ensures that all components receive adequate power. The real-time monitoring module supports the display and storage of image data. In recent years, the Robot Operation System (ROS) has greatly improved software development efficiency. By employing ROS for multieye timing synchronization, the system achieves consistent data recording. In a two-day trial at the Deqingyuan Chicken Coop in Beijing, the system collected 2657 pairs of binocular and thermal infrared images, demonstrating real-time capabilities with a maximum time difference <em>t<sub>max</sub></em> of 4.97 ms, a minimum <em>t<sub>min</sub></em> of 0.66 ms, an average <em>t<sub>mean</sub></em> of 1.25 ms, and a root mean square error <em>t<sub>RMSE</sub></em> of 1.31 ms. Target detection tests indicated that You Only Look Once version 5 (YOLOv5) performed best, with a Precision (<em>P</em>) of 92.71 %, a Recall (<em>R</em>) of 93.91 %, mean Average Precision (<em>mAP</em>) of 96.32 %, and an inference speed of 49.6 ms, showing high-precision and real-time detection. Image registration tests revealed a maximum matching error <em>H<sub>max</sub></em> of 0.86 pixels, a minimum <em>H<sub>min</sub></em> of 0.35 pixels, and an average <em>H<sub>mean</sub></em> of 0.61 pixels. The maximum structural similarity index (<em>SSIM<sub>max</sub></em>) was 0.86, the minimum <em>SSIM<sub>min</sub></em> was 0.61, and the average <em>SSIM<sub>mean</sub></em> was 0.78. High-precision target detection methods, image time synchronization methods, and spatial registration methods are used to confirm high-precision image registration and accurate target temperature measurement. This system provides a technical solution and equipment support for the quick and accurate acquisition of chicken temperature information in the poultry breeding industry.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"244 ","pages":"Article 111464"},"PeriodicalIF":8.9,"publicationDate":"2026-01-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146025256","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"农林科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Accurate and temporally consistent multispectral observations are essential for monitoring alfalfa yield and quality, given its frequent harvest cycles and rapid regrowth. However, optical satellite imagery is often constrained by cloud cover, revisit intervals, and sensor availability. To overcome these limitations, we propose a novel Alfalfa Multimodal Generative Adversarial Network (AMGAN) designed for near-daily multispectral image reconstruction. Unlike conventional image-to-image or spatiotemporal fusion methods that overlook crop-specific characteristics, are restricted to observed timestamps, or depend heavily on dense temporal series, AMGAN leverages multisource (Landsat-8/9, Sentinel-1, PlanetScope) and multimodal (climate, geographic, temporal) information within an adversarial learning paradigm. This enables high-quality image generation from minimal inputs. Extensive experiments across five major alfalfa-producing states in the United States (2022–2024) show that AMGAN consistently surpasses four state-of-the-art (SOTA) deep learning baselines. It achieves higher reconstruction accuracy across all spectral bands, with pronounced gains in red-edge and near-infrared (NIR) regions critical for vegetation assessment. Multisource integration and multimodal cues enhance robustness, ensuring reliable performance under diverse observation scenarios. The reconstructed imagery was subsequently evaluated in alfalfa yield and quality prediction tasks. Results demonstrated high predictive accuracy for dry matter yield (DM) in the cross validation (CV) experiment with a coefficient of determination (R2) of 0.80, and moderate correlations for selected quality traits such as crude protein (CP), non-fiber carbohydrates (NFC), and minerals, while nutritive value traits tied to complex biochemical processes remained more challenging. Overall, this study underscores the potential of multimodal adversarial learning to bridge observational gaps in alfalfa monitoring. The proposed framework provides a scalable, crop-specific approach for generating temporally dense imagery, supporting precision management for biomass-related and proximate quality traits, while performance for digestibility traits remains limited.
{"title":"AMGAN: A multimodal generative adversarial network for near-daily alfalfa multispectral image reconstruction","authors":"Tong Yu , Jiang Chen , Jerome H. Cherney , Zhou Zhang","doi":"10.1016/j.compag.2026.111468","DOIUrl":"10.1016/j.compag.2026.111468","url":null,"abstract":"<div><div>Accurate and temporally consistent multispectral observations are essential for monitoring alfalfa yield and quality, given its frequent harvest cycles and rapid regrowth. However, optical satellite imagery is often constrained by cloud cover, revisit intervals, and sensor availability. To overcome these limitations, we propose a novel Alfalfa Multimodal Generative Adversarial Network (AMGAN) designed for near-daily multispectral image reconstruction. Unlike conventional image-to-image or spatiotemporal fusion methods that overlook crop-specific characteristics, are restricted to observed timestamps, or depend heavily on dense temporal series, AMGAN leverages multisource (Landsat-8/9, Sentinel-1, PlanetScope) and multimodal (climate, geographic, temporal) information within an adversarial learning paradigm. This enables high-quality image generation from minimal inputs. Extensive experiments across five major alfalfa-producing states in the United States (2022–2024) show that AMGAN consistently surpasses four state-of-the-art (SOTA) deep learning baselines. It achieves higher reconstruction accuracy across all spectral bands, with pronounced gains in red-edge and near-infrared (NIR) regions critical for vegetation assessment. Multisource integration and multimodal cues enhance robustness, ensuring reliable performance under diverse observation scenarios. The reconstructed imagery was subsequently evaluated in alfalfa yield and quality prediction tasks. Results demonstrated high predictive accuracy for dry matter yield (DM) in the cross validation (CV) experiment with a coefficient of determination (R<sup>2</sup>) of 0.80, and moderate correlations for selected quality traits such as crude protein (CP), non-fiber carbohydrates (NFC), and minerals, while nutritive value traits tied to complex biochemical processes remained more challenging. Overall, this study underscores the potential of multimodal adversarial learning to bridge observational gaps in alfalfa monitoring. The proposed framework provides a scalable, crop-specific approach for generating temporally dense imagery, supporting precision management for biomass-related and proximate quality traits, while performance for digestibility traits remains limited.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"244 ","pages":"Article 111468"},"PeriodicalIF":8.9,"publicationDate":"2026-01-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146025260","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"农林科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Precise decision-making in agricultural production has always been a key bottleneck restricting the development of modern agriculture. Traditional decision-making models that rely on manual experience are no longer able to meet the needs of modern intelligent agriculture. This study, by integrating the Agricultural Internet of Things (IoT) with generative big models, has realized a fully autonomous agricultural intelligent system called the ”Fuxi Brain.” This system consists of two parts. The first part constructs a comprehensive ”sky-air-ground-human-machine” data collection system, enabling digital perception of all factors in agricultural production. The second part is the intelligent decision-making system. First, within the brain’s dynamic decision-making layer, a multi-agent collaborative architecture based on a hybrid multi-model (a general big model + a specialized agricultural model) is proposed. Furthermore, a dynamic optimal matrix algorithm (DOMA) is designed to improve the system’s decision-making efficiency significantly. Finally, a full-modality alignment training method is developed to effectively address the challenge of integrating multi-source heterogeneous data. Experimental results show that, in the AlpacaEva and MT-Bench benchmarks, the system’s decision accuracy improved by 36.7 percentage points compared to mainstream models such as ChatGLM. The full-modal alignment training method significantly outperformed traditional methods in cross-modal understanding tasks. Test results on a one-stop agricultural service decision-making platform demonstrated an accuracy rate of 92.3% compared to those of human experts. In an actual application on a 1367-acre corn planting site at Dahewan Farm in Inner Mongolia, the system autonomously generated 127 decisions throughout the production cycle, achieving an accuracy rate of 89.7%. This successfully enabled autonomous and precise decision-making throughout the entire process, from planting to harvesting. This research provides innovative technical paths and practical examples for the development of intelligent agriculture.
{"title":"Agricultural autonomous decision-making system ”Fuxi Brain” Based on generative large model fusion internet of things","authors":"Haihua Chen , Guangyu Hou , Chen Hua , Shuo Wang , Ziyu Chen , Yanbin Zhang","doi":"10.1016/j.compag.2026.111454","DOIUrl":"10.1016/j.compag.2026.111454","url":null,"abstract":"<div><div>Precise decision-making in agricultural production has always been a key bottleneck restricting the development of modern agriculture. Traditional decision-making models that rely on manual experience are no longer able to meet the needs of modern intelligent agriculture. This study, by integrating the Agricultural Internet of Things (IoT) with generative big models, has realized a fully autonomous agricultural intelligent system called the ”Fuxi Brain.” This system consists of two parts. The first part constructs a comprehensive ”sky-air-ground-human-machine” data collection system, enabling digital perception of all factors in agricultural production. The second part is the intelligent decision-making system. First, within the brain’s dynamic decision-making layer, a multi-agent collaborative architecture based on a hybrid multi-model (a general big model + a specialized agricultural model) is proposed. Furthermore, a dynamic optimal matrix algorithm (DOMA) is designed to improve the system’s decision-making efficiency significantly. Finally, a full-modality alignment training method is developed to effectively address the challenge of integrating multi-source heterogeneous data. Experimental results show that, in the AlpacaEva and MT-Bench benchmarks, the system’s decision accuracy improved by 36.7 percentage points compared to mainstream models such as ChatGLM. The full-modal alignment training method significantly outperformed traditional methods in cross-modal understanding tasks. Test results on a one-stop agricultural service decision-making platform demonstrated an accuracy rate of 92.3% compared to those of human experts. In an actual application on a 1367-acre corn planting site at Dahewan Farm in Inner Mongolia, the system autonomously generated 127 decisions throughout the production cycle, achieving an accuracy rate of 89.7%. This successfully enabled autonomous and precise decision-making throughout the entire process, from planting to harvesting. This research provides innovative technical paths and practical examples for the development of intelligent agriculture.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"244 ","pages":"Article 111454"},"PeriodicalIF":8.9,"publicationDate":"2026-01-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146025257","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"农林科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-23DOI: 10.1016/j.compag.2026.111469
K. Colton Flynn , H.K. Chinmayi , Gurjinder S. Baath , Bala Ram Sapkota , Chris Delhom , Douglas R. Smith
Accurate estimation of the Leaf Area Index (LAI) is essential for assessing vegetation health and managing agricultural productivity. This study examines the application of Unmanned Aerial Vehicle (UAV)-based hyperspectral imaging and convolved EnMAP spectral data for estimating corn LAI, utilizing machine learning (ML) models to improve prediction accuracy. Various ML models, including k-nearest Neighbors (KNN), Support Vector Machines (SVM), Partial Least Squares Regression (PLS), and Random Forests (RF), were assessed to predict LAI from hyperspectral, EnMAP, and vegetation index features. Results demonstrate that PLS models consistently outperformed other ML approaches, achieving coefficients of determination (R2) ranging from 0.79 to 0.82. Notably, for the top two performing models (PLS and SVM) spectral indices such as NDRE, GNDVI, and NDVI proved more effective for LAI prediction than individual spectral bands. Interestingly, no matter the incorporation of hyperspectral wavelengths or EnMAP bands, the models predicting LAI were comparable. Feature importance analysis reinforced the dominance of vegetation indices as key predictors. The findings emphasize the benefits of high-resolution UAV hyperspectral imaging, convolved satellite spectral data, and machine learning, particularly PLS, for scalable and accurate LAI estimation in agroecosystems.
{"title":"UAV-based estimates of corn LAI using hyperspectral and EnMAP spectral resolutions","authors":"K. Colton Flynn , H.K. Chinmayi , Gurjinder S. Baath , Bala Ram Sapkota , Chris Delhom , Douglas R. Smith","doi":"10.1016/j.compag.2026.111469","DOIUrl":"10.1016/j.compag.2026.111469","url":null,"abstract":"<div><div>Accurate estimation of the Leaf Area Index (LAI) is essential for assessing vegetation health and managing agricultural productivity. This study examines the application of Unmanned Aerial Vehicle (UAV)-based hyperspectral imaging and convolved EnMAP spectral data for estimating corn LAI, utilizing machine learning (ML) models to improve prediction accuracy. Various ML models, including k-nearest Neighbors (KNN), Support Vector Machines (SVM), Partial Least Squares Regression (PLS), and Random Forests (RF), were assessed to predict LAI from hyperspectral, EnMAP, and vegetation index features. Results demonstrate that PLS models consistently outperformed other ML approaches, achieving coefficients of determination (<em>R<sup>2</sup></em>) ranging from 0.79 to 0.82. Notably, for the top two performing models (PLS and SVM) spectral indices such as NDRE, GNDVI, and NDVI proved more effective for LAI prediction than individual spectral bands. Interestingly, no matter the incorporation of hyperspectral wavelengths or EnMAP bands, the models predicting LAI were comparable. Feature importance analysis reinforced the dominance of vegetation indices as key predictors. The findings emphasize the benefits of high-resolution UAV hyperspectral imaging, convolved satellite spectral data, and machine learning, particularly PLS, for scalable and accurate LAI estimation in agroecosystems.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"244 ","pages":"Article 111469"},"PeriodicalIF":8.9,"publicationDate":"2026-01-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146025258","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"农林科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号