Pub Date : 2026-01-14DOI: 10.1016/j.compag.2026.111437
Shiyuan Liu , Yumeng Zhou , Weiguang Yang , Jiangtao Tan , Xi Li , Zhenhui Xiong , Zewu Fang , Hong Li , Yifei Chen , Yubin Lan , Shubo Wan , Jianguo Wang , Tingting Chen , Lei Zhang
Empirical and physical models are widely used for monitoring equivalent water thickness (EWT) to adjust plant moisture management. However, model transferability to different times and locations, and insufficient training data remain the two key challenges of field spectroscopy analysis. Therefore, this study aims to construct a hybrid model, which combines the physical models optimized by Wasserstein Generative Adversarial Nets (WGAN) and empirical models for performing hyperparameter searches (the process of finding optimal model settings) to monitor the peanut EWT. Specifically, we develop a large spectral dataset consisting of field-measured data which including 246 peanut varieties in five peanut farms across China and synthetic datasets generated from the physical models optimized by WGAN. Furthermore, the PWLEH was constructed by hyperparameter tuning and pre-training which using synthetic datasets, and then fine-tuned by modular training with field data of peanut canopy water content. Comparing the model constructed with field data (R2 = 0.5618, mean squared error (MSE) = 0.0725) and PROSAIL (a widely used canopy radiative transfer model) (R2 = 0.7105, MSE = 0.0473), PWLEH achieved high accuracy in predicting peanut water content (R2 = 0.7650, MSE = 0.0519). Unlike pure data-driven approaches, the new hybrid model incorporated radiative transfer knowledge and obtained higher predictive performance with fewer field data. This study demonstrates the potential of applying an optimized PROSAIL, hyperparameter search and modular training to improve the accuracy and transferability of the EWT prediction model, providing a new approach for sustainable agricultural management.
{"title":"A novel approach to monitor peanut equivalent water thickness through modular training and transfer learning of an improved PROSAIL model using a Wasserstein generative adversarial network","authors":"Shiyuan Liu , Yumeng Zhou , Weiguang Yang , Jiangtao Tan , Xi Li , Zhenhui Xiong , Zewu Fang , Hong Li , Yifei Chen , Yubin Lan , Shubo Wan , Jianguo Wang , Tingting Chen , Lei Zhang","doi":"10.1016/j.compag.2026.111437","DOIUrl":"10.1016/j.compag.2026.111437","url":null,"abstract":"<div><div>Empirical and physical models are widely used for monitoring equivalent water thickness (EWT) to adjust plant moisture management. However, model transferability to different times and locations, and insufficient training data remain the two key challenges of field spectroscopy analysis. Therefore, this study aims to construct a hybrid model, which combines the physical models optimized by Wasserstein Generative Adversarial Nets (WGAN) and empirical models for performing hyperparameter searches (the process of finding optimal model settings) to monitor the peanut EWT. Specifically, we develop a large spectral dataset consisting of field-measured data which including 246 peanut varieties in five peanut farms across China and synthetic datasets generated from the physical models optimized by WGAN. Furthermore, the PWLEH was constructed by hyperparameter tuning and pre-training which using synthetic datasets, and then fine-tuned by modular training with field data of peanut canopy water content. Comparing the model constructed with field data (R<sup>2</sup> = 0.5618, mean squared error (MSE) = 0.0725) and PROSAIL (a widely used canopy radiative transfer model) (R<sup>2</sup> = 0.7105, MSE = 0.0473), PWLEH achieved high accuracy in predicting peanut water content (R<sup>2</sup> = 0.7650, MSE = 0.0519). Unlike pure data-driven approaches, the new hybrid model incorporated radiative transfer knowledge and obtained higher predictive performance with fewer field data. This study demonstrates the potential of applying an optimized PROSAIL, hyperparameter search and modular training to improve the accuracy and transferability of the EWT prediction model, providing a new approach for sustainable agricultural management.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"243 ","pages":"Article 111437"},"PeriodicalIF":8.9,"publicationDate":"2026-01-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145979075","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 : 2026-01-13DOI: 10.1016/j.compag.2025.111302
Uddhav Bhattarai , Rajkishan Arikapudi , Chen Peng , Steven A. Fennimore , Frank N. Martin , Stavros G. Vougioukas
High-resolution yield maps for manually harvested crops are impractical to generate on commercial scales because yield monitors are available only for mechanical harvesters. However, precision crop management relies on accurately determining spatial and temporal yield variability. This study presents the development of an integrated system for precision yield estimation and mapping for manually harvested strawberries. Conventional strawberry picking carts were instrumented with a Global Positioning System (GPS) receiver, an Inertial Measurement Unit (IMU), and load cells to record real-time geo-tagged harvest data and cart motion. Extensive data were collected in two strawberry fields in California, USA, during a harvest season. To address the inconsistencies and errors caused by the sensors and the manual harvesting process, a robust data processing pipeline was developed by integrating supervised deep learning model with unsupervised algorithms. The pipeline was used to estimate the yield distribution and generate yield maps for season-long harvests at the desired grid resolution. The estimated yield distributions were used to calculate two metrics: the total mass harvested over specific row segments and the total mass of trays harvested. The metrics were compared to ground truth and achieved accuracies of 90.48% and 94.05%, respectively. Additionally, the accuracy of the estimated yield based on the number of trays harvested per cart for season-long harvest was better than 94% achieving a strong correlation (Pearson r = 0.99) with the actual number of counted trays in both fields. The proposed system provides a scalable and practical solution for specialty crops, assisting in efficient yield estimation and mapping, field management, and labor management for sustainable crop production. The dataset and code supporting this study are available at: https://doi.org/10.5061/dryad.v6wwpzh7h and https://github.com/uddhavbhattarai/iCarritoYieldEstimationandMapping.git.
人工收割作物的高分辨率产量图在商业规模上是不切实际的，因为产量监测器只能用于机械收割。然而，精确的作物管理依赖于准确地确定产量的时空变化。本研究提出了一种用于人工收获草莓的精确产量估算和制图的集成系统的开发。传统的草莓采摘车配备了全球定位系统（GPS）接收器、惯性测量单元（IMU）和称重传感器，以记录实时地理标记的收获数据和推车运动。在一个收获季节，在美国加利福尼亚州的两个草莓田收集了大量数据。为了解决传感器和人工采集过程中产生的不一致和错误，将有监督深度学习模型与无监督算法相结合，开发了鲁棒的数据处理管道。该管道用于估计产量分布，并在所需的网格分辨率下生成季节性收获的产量图。估计的产量分布用于计算两个指标：特定行段收获的总质量和收获的托盘总质量。将这些指标与地面真实度进行比较，准确率分别为90.48%和94.05%。此外，基于每辆车收获的托盘数量的估计产量的准确性优于94%，与两个领域的实际计算托盘数量具有很强的相关性（Pearson r = 0.99）。该系统为特种作物提供了可扩展和实用的解决方案，有助于有效的产量估算和制图、田间管理和可持续作物生产的劳动力管理。支持本研究的数据集和代码可在https://doi.org/10.5061/dryad.v6wwpzh7h和https://github.com/uddhavbhattarai/iCarritoYieldEstimationandMapping.git上获得。
{"title":"Precision yield estimation and mapping in manual strawberry harvesting with instrumented picking carts and a robust data processing pipeline","authors":"Uddhav Bhattarai , Rajkishan Arikapudi , Chen Peng , Steven A. Fennimore , Frank N. Martin , Stavros G. Vougioukas","doi":"10.1016/j.compag.2025.111302","DOIUrl":"10.1016/j.compag.2025.111302","url":null,"abstract":"<div><div>High-resolution yield maps for manually harvested crops are impractical to generate on commercial scales because yield monitors are available only for mechanical harvesters. However, precision crop management relies on accurately determining spatial and temporal yield variability. This study presents the development of an integrated system for precision yield estimation and mapping for manually harvested strawberries. Conventional strawberry picking carts were instrumented with a Global Positioning System (GPS) receiver, an Inertial Measurement Unit (IMU), and load cells to record real-time geo-tagged harvest data and cart motion. Extensive data were collected in two strawberry fields in California, USA, during a harvest season. To address the inconsistencies and errors caused by the sensors and the manual harvesting process, a robust data processing pipeline was developed by integrating supervised deep learning model with unsupervised algorithms. The pipeline was used to estimate the yield distribution and generate yield maps for season-long harvests at the desired grid resolution. The estimated yield distributions were used to calculate two metrics: the total mass harvested over specific row segments and the total mass of trays harvested. The metrics were compared to ground truth and achieved accuracies of 90.48% and 94.05%, respectively. Additionally, the accuracy of the estimated yield based on the number of trays harvested per cart for season-long harvest was better than 94% achieving a strong correlation (Pearson r = 0.99) with the actual number of counted trays in both fields. The proposed system provides a scalable and practical solution for specialty crops, assisting in efficient yield estimation and mapping, field management, and labor management for sustainable crop production. The dataset and code supporting this study are available at: <span><span>https://doi.org/10.5061/dryad.v6wwpzh7h</span><svg><path></path></svg></span> and <span><span>https://github.com/uddhavbhattarai/iCarritoYieldEstimationandMapping.git</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"243 ","pages":"Article 111302"},"PeriodicalIF":8.9,"publicationDate":"2026-01-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145978374","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 : 2026-01-13DOI: 10.1016/j.compag.2025.111392
Jayme Garcia Arnal Barbedo , Marcelo Santos da Silva , Mirela Teixeira Cazzolato , Lucas Pascotti Valem , Renato Tinós , Roseli Aparecida Francelin Romero , Luiz Otavio Murta Junior , Adriano de Jesus Holanda , Joaquim Cezar Felipe , José Baldin Pinheiro , José Tiago Barroso Chagas , Roberto Fray da Silva , Everton Castelão Tetila , Lucio Andre de Castro Jorge , Huaqiang Yuan , Weiling Li , Ketan Kotecha , Liang Zhao
AI is playing an increasingly central role in crop monitoring, driven by rapid advances in deep learning that now tackle recognition and prediction tasks once out of reach. However, translating these gains into soybean production is increasingly constrained by two intertwined requirements, explainability (to support expert scrutiny and responsible use of black-box models) and privacy (to protect sensitive farm data and enable collaboration across stakeholders). This review synthesizes recent advances in interpretable and privacy preserving machine learning, emphasizing soybean related applications where empirical evidence is solid, and covering both post hoc and inherently interpretable approaches alongside privacy mechanisms such as federated learning with secure aggregation and differential privacy. Across the literature, recurring deployment barriers are identified, most notably variability across farms and seasons, the need for explanations that remain meaningful both locally and globally, infrastructure limitations in rural settings, risks of information leakage through explanations, and the scarcity of multi-season validation under real-world conditions. These findings suggest that field-ready soybean monitoring systems should be designed with explainability and privacy as major goals, rather than add-ons, and evaluated under realistic variability and governance requirements. The ultimate goal is to help bridge the gap between academic innovation and practical, deployable solutions that protect farmer data while supporting decision-making where it matters most.
{"title":"Explainability and privacy in AI-enabled crop monitoring: Trends and future directions in soybean research","authors":"Jayme Garcia Arnal Barbedo , Marcelo Santos da Silva , Mirela Teixeira Cazzolato , Lucas Pascotti Valem , Renato Tinós , Roseli Aparecida Francelin Romero , Luiz Otavio Murta Junior , Adriano de Jesus Holanda , Joaquim Cezar Felipe , José Baldin Pinheiro , José Tiago Barroso Chagas , Roberto Fray da Silva , Everton Castelão Tetila , Lucio Andre de Castro Jorge , Huaqiang Yuan , Weiling Li , Ketan Kotecha , Liang Zhao","doi":"10.1016/j.compag.2025.111392","DOIUrl":"10.1016/j.compag.2025.111392","url":null,"abstract":"<div><div>AI is playing an increasingly central role in crop monitoring, driven by rapid advances in deep learning that now tackle recognition and prediction tasks once out of reach. However, translating these gains into soybean production is increasingly constrained by two intertwined requirements, explainability (to support expert scrutiny and responsible use of black-box models) and privacy (to protect sensitive farm data and enable collaboration across stakeholders). This review synthesizes recent advances in interpretable and privacy preserving machine learning, emphasizing soybean related applications where empirical evidence is solid, and covering both post hoc and inherently interpretable approaches alongside privacy mechanisms such as federated learning with secure aggregation and differential privacy. Across the literature, recurring deployment barriers are identified, most notably variability across farms and seasons, the need for explanations that remain meaningful both locally and globally, infrastructure limitations in rural settings, risks of information leakage through explanations, and the scarcity of multi-season validation under real-world conditions. These findings suggest that field-ready soybean monitoring systems should be designed with explainability and privacy as major goals, rather than add-ons, and evaluated under realistic variability and governance requirements. The ultimate goal is to help bridge the gap between academic innovation and practical, deployable solutions that protect farmer data while supporting decision-making where it matters most.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"243 ","pages":"Article 111392"},"PeriodicalIF":8.9,"publicationDate":"2026-01-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145979073","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 : 2026-01-13DOI: 10.1016/j.compag.2025.111299
Hongli Zhang , Jingjian Li , Fenghua Huang , Shulin Liu , Sha Wei , Haihua Xiao , Feng Ding
This paper proposes a path planning method based on the A-star(A*) algorithm to address the coverage path planning problem for bagging young peaches in orchards. First, based on young peach spatial coordinates and the robot’s workspace, a method is proposed to determine and concisely represent the bagging area. Then, a work position determination method called Radius-Weighted K-Means(RWK-means) is designed, incorporating radius weights of the bagging areas. Additionally, a coverage path planning algorithm named Centroid Set A*(CSA*) is introduced, which includes a turning penalty mechanism. Experimental results demonstrate that, given the same number of bagging positions, the RWK-means algorithm results in fewer unbagged peaches and a lower unbagged peach ratio. Moreover, the CSA* algorithm outperforms the traditional path planning algorithm in both point-to-point and coverage path planning tasks. In particular, for the coverage task of bagging young peaches, the CSA* algorithm produces paths that are 33.68% shorter than those generated by the Boustrophedon method. These results confirm the effectiveness of the proposed approach in coverage path planning for peach bagging.
{"title":"Coverage path planning for a bagging robot via spatial coordinate projection clustering of young peaches","authors":"Hongli Zhang , Jingjian Li , Fenghua Huang , Shulin Liu , Sha Wei , Haihua Xiao , Feng Ding","doi":"10.1016/j.compag.2025.111299","DOIUrl":"10.1016/j.compag.2025.111299","url":null,"abstract":"<div><div>This paper proposes a path planning method based on the A-star(A*) algorithm to address the coverage path planning problem for bagging young peaches in orchards. First, based on young peach spatial coordinates and the robot’s workspace, a method is proposed to determine and concisely represent the bagging area. Then, a work position determination method called Radius-Weighted K-Means(RWK-means) is designed, incorporating radius weights of the bagging areas. Additionally, a coverage path planning algorithm named Centroid Set A*(CSA*) is introduced, which includes a turning penalty mechanism. Experimental results demonstrate that, given the same number of bagging positions, the RWK-means algorithm results in fewer unbagged peaches and a lower unbagged peach ratio. Moreover, the CSA* algorithm outperforms the traditional path planning algorithm in both point-to-point and coverage path planning tasks. In particular, for the coverage task of bagging young peaches, the CSA* algorithm produces paths that are 33.68% shorter than those generated by the Boustrophedon method. These results confirm the effectiveness of the proposed approach in coverage path planning for peach bagging.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"243 ","pages":"Article 111299"},"PeriodicalIF":8.9,"publicationDate":"2026-01-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145979079","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 : 2026-01-13DOI: 10.1016/j.compag.2026.111428
Saeed Karimzadeh , Robert D. McAllister , Md Shamim Ahamed
Achieving closed-loop hydroponics necessitates precise adjustment of individual macro- and micronutrients within the nutrient solution. However, nutrient management in hydroponics remains constrained to electrical conductivity (EC) and pH-based approaches, due to the complexity of steering individual ions and the coupling inherent in multi-element fertilizer formulations. In this study, an optimal control framework, termed OptiDose, is implemented to optimize daily fertigation strategies for hydroponically grown lettuce. The system integrates six fertilizer sources—calcium nitrate, magnesium sulfate, monopotassium phosphate, potassium nitrate, magnesium nitrate, and potassium sulfate—to maintain the concentrations of the macronutrients nitrogen (N), phosphorus (P), potassium (K), calcium (Ca), magnesium (Mg), and sulfur (S) within crop-specific adequacy ranges. Five scenarios are tested in the simulator to evaluate system performance under varying operational constraints. Results indicate that OptiDose maintained suitable nutrient concentrations for plants throughout the growth cycle—without nutrient deficiencies or toxicities—while markedly improving resource-use efficiency. Relative to a single-shot nutrient preparation (baseline), the strategy using properly sized solution tanks with daily recipe adjustment (Scenario 1) increased water-use efficiency sixfold and doubled fertilizer-use efficiency, achieving 32.3 ± 1.4 g/L and 12.3 ± 0.3 g/g, respectively. Additionally, water and fertilizer costs decreased significantly (p < 0.05), by approximately 76% and 51%, respectively. The results underscore the promise of element-specific fertigation and optimization for precision nutrient management in controlled environment agriculture.
{"title":"OptiDose: An optimal control for macronutrient dosing in hydroponics","authors":"Saeed Karimzadeh , Robert D. McAllister , Md Shamim Ahamed","doi":"10.1016/j.compag.2026.111428","DOIUrl":"10.1016/j.compag.2026.111428","url":null,"abstract":"<div><div>Achieving closed-loop hydroponics necessitates precise adjustment of individual macro- and micronutrients within the nutrient solution. However, nutrient management in hydroponics remains constrained to electrical conductivity (EC) and pH-based approaches, due to the complexity of steering individual ions and the coupling inherent in multi-element fertilizer formulations. In this study, an optimal control framework, termed <em>OptiDose</em>, is implemented to optimize daily fertigation strategies for hydroponically grown lettuce. The system integrates six fertilizer sources—calcium nitrate, magnesium sulfate, monopotassium phosphate, potassium nitrate, magnesium nitrate, and potassium sulfate—to maintain the concentrations of the macronutrients nitrogen (N), phosphorus (P), potassium (K), calcium (Ca), magnesium (Mg), and sulfur (S) within crop-specific adequacy ranges. Five scenarios are tested in the simulator to evaluate system performance under varying operational constraints. Results indicate that <em>OptiDose</em> maintained suitable nutrient concentrations for plants throughout the growth cycle—without nutrient deficiencies or toxicities—while markedly improving resource-use efficiency. Relative to a single-shot nutrient preparation (baseline), the strategy using properly sized solution tanks with daily recipe adjustment (Scenario 1) increased water-use efficiency sixfold and doubled fertilizer-use efficiency, achieving 32.3 ± 1.4 g/L and 12.3 ± 0.3 g/g, respectively. Additionally, water and fertilizer costs decreased significantly (<em>p</em> < 0.05), by approximately 76% and 51%, respectively. The results underscore the promise of element-specific fertigation and optimization for precision nutrient management in controlled environment agriculture.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"243 ","pages":"Article 111428"},"PeriodicalIF":8.9,"publicationDate":"2026-01-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145978386","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 : 2026-01-13DOI: 10.1016/j.compag.2025.111394
Yao Gu, Baozhi Li, Wei Lu
In the cultivation of mushroom (Agaricus bisporus), the factory-style environment with constant temperature and humidity, along with mechanized production equipment, is adopted to enhance the efficient production and supply of mushrooms throughout the four seasons. However, the harvesting process still relies heavily on manual labor, which severely restricts industrial development and the improvement of enterprise efficiency. Therefore, there is an urgent need to develop automatic mushroom harvesting machines. This paper conducts systematic research and develops a bioinspired mushroom harvesting robot inspired by the manual mushroom-picking movements. First, the configuration optimization of the mushroom harvesting robot was carried out, based on the requirements of being able to walk between mushroom shelves, reach mushroom beds at different heights, and achieve low weight, low energy consumption, and high efficiency. Secondly, a measurement and control system for the robot was developed to achieve the detection of position and target information, as well as dynamic and precise control of different actuators. Then, a fast algorithm for mushroom recognition, size measurement, and positioning based on YOLOv8n, as well as a compliant control algorithm for mushroom-picking grippers, were successively researched and implemented. Finally, field harvesting experiments were conducted in a mushroom factory. The results showed that the robot’s recognition and positioning accuracy was less than 1 cm, the harvesting accuracy reached 99.3%, the mushroom damage rate was less than 3.2%, and the average harvesting efficiency was 20.8 kg per hour—equivalent to the efficiency of one worker. The experimental results verify the rationality of the robot design, the accuracy of the visual algorithm and the robustness of the control, and can effectively complete the automated harvesting of multi-layer mushrooms.
在蘑菇（Agaricus bisporus）的栽培中，采用恒温恒湿的工厂化环境，配合机械化生产设备，提高一年四季蘑菇的高效生产和供应。然而，采收过程仍然严重依赖人工，严重制约了产业的发展和企业效益的提高。因此，迫切需要研制全自动蘑菇采收机。本文通过系统研究，开发了一种以人工采菇动作为灵感的仿生采菇机器人。首先，根据能够在蘑菇货架之间行走，到达不同高度的蘑菇床，实现低重量、低能耗、高效率的要求，对蘑菇收获机器人进行了配置优化。其次，开发了机器人的测控系统，实现了机器人位置和目标信息的检测，以及不同执行器的动态精确控制。在此基础上，研究并实现了基于YOLOv8n的蘑菇快速识别、尺寸测量和定位算法，以及蘑菇采摘机械手的柔性控制算法。最后，在香菇厂进行了田间收获试验。结果表明，该机器人的识别定位精度小于1 cm，收获精度达到99.3%，蘑菇破损率小于3.2%，平均收获效率为20.8 kg / h，相当于一名工人的效率。实验结果验证了机器人设计的合理性、视觉算法的准确性和控制的鲁棒性，能够有效地完成多层蘑菇的自动化采收。
{"title":"Mushroom (Agaricus bisporus) picking robot based on mimicry of manual harvesting","authors":"Yao Gu, Baozhi Li, Wei Lu","doi":"10.1016/j.compag.2025.111394","DOIUrl":"10.1016/j.compag.2025.111394","url":null,"abstract":"<div><div>In the cultivation of<!--> <!-->mushroom (<em>Agaricus bisporus</em>), the factory-style environment with constant temperature and humidity, along with mechanized production equipment, is adopted to enhance the efficient production and supply of mushrooms throughout the four seasons. However, the harvesting process still relies heavily on manual labor, which severely restricts industrial development and the improvement of enterprise efficiency. Therefore, there is an urgent need to develop automatic mushroom harvesting machines. This paper conducts systematic research and develops a bioinspired mushroom harvesting robot inspired by the manual mushroom-picking movements. First, the configuration optimization of the mushroom harvesting robot was carried out, based on the requirements of being able to walk between mushroom shelves, reach mushroom beds at different heights, and achieve low weight, low energy consumption, and high efficiency. Secondly, a measurement and control system for the robot was developed to achieve the detection of position and target information, as well as dynamic and precise control of different actuators. Then, a fast algorithm for mushroom recognition, size measurement, and positioning based on YOLOv8n, as well as a compliant control algorithm for mushroom-picking grippers, were successively researched and implemented. Finally, field harvesting experiments were conducted in a mushroom factory. The results showed that the robot’s recognition and positioning accuracy was less than 1 cm, the harvesting accuracy reached 99.3%, the mushroom damage rate was less than 3.2%, and the average harvesting efficiency was 20.8 kg per hour—equivalent to the efficiency of one worker. The experimental results verify the rationality of the robot design, the accuracy of the visual algorithm and the robustness of the control, and can effectively complete the automated harvesting of multi-layer mushrooms.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"243 ","pages":"Article 111394"},"PeriodicalIF":8.9,"publicationDate":"2026-01-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145979078","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 : 2026-01-12DOI: 10.1016/j.compag.2026.111423
Changqing An , Penghui Liu , Xiaopeng Lv , Zihao Wu , Maozhen Qu , Zhizhong Sun , Xiuqin Rao , Huirong Xu
A universal model for predicting soluble solids content (SSC) of multi-cultivar watermelons is of great significance to enhance online detection efficiency and minimize redundant modeling. However, the accuracy of universal model is challenged by the spectral difference of multi-cultivar watermelons. This study mapped SSC distribution within the seedless and seeded watermelons, and compared their optical properties between different watermelon tissues (green rind, white rind and flesh). A one-dimensional convolutional neural network (1D-CNN) with gradient-weighted class activation mapping-based (Grad-CAM) was trained for classification of 163 seedless and 160 seeded watermelons, and to localize cultivar-sensitive wavelengths. These spectral bands were then down-weighted to generate spectra less affected by cultivar differences, which were then used to predict SSC using the partial least squares regression (PLSR). For large-sized watermelons, central flesh exhibited the highest SSC and the strongest correlation with whole SSC, supporting a practical sampling strategy. Seedless and seeded watermelons showed marked transmittance differences attributable to their optical properties, and Monte Carlo simulations reproduced the differences of light attenuation, consistent with measurements. By combining standard normalized variate (SNV) with Grad-CAM for spectral preprocessing, the coefficient of determination (Rp2), root mean square error (RMSEP), and residual prediction deviation (RPD) in the prediction set reached 0.79, 0.55 °Brix, and 2.2, respectively. Compared to the model developed with original spectra, the RMSEP decreased by 0.14 °Brix, while Rp2 and RPD increased by 0.11 and 0.4, respectively. Grad-CAM located and down-weighted composition-sensitive wavelength regions (chlorophyll at ∼ 680 nm and water/organic bands at ∼ 920 nm), thereby mitigating the cultivar effect on spectral modelling. The established method also casts innovative light on universal model for fruit from different origins, seasons etc.
{"title":"An explainable deep learning–guided universal model for improving soluble solids content detection accuracy in seeded/seedless watermelons","authors":"Changqing An , Penghui Liu , Xiaopeng Lv , Zihao Wu , Maozhen Qu , Zhizhong Sun , Xiuqin Rao , Huirong Xu","doi":"10.1016/j.compag.2026.111423","DOIUrl":"10.1016/j.compag.2026.111423","url":null,"abstract":"<div><div>A universal model for predicting soluble solids content (SSC) of multi-cultivar watermelons is of great significance to enhance online detection efficiency and minimize redundant modeling. However, the accuracy of universal model is challenged by the spectral difference of multi-cultivar watermelons. This study mapped SSC distribution within the seedless and seeded watermelons, and compared their optical properties between different watermelon tissues (green rind, white rind and flesh). A one-dimensional convolutional neural network (1D-CNN) with gradient-weighted class activation mapping-based (Grad-CAM) was trained for classification of 163 seedless and 160 seeded watermelons, and to localize cultivar-sensitive wavelengths. These spectral bands were then down-weighted to generate spectra less affected by cultivar differences, which were then used to predict SSC using the partial least squares regression (PLSR). For large-sized watermelons, central flesh exhibited the highest SSC and the strongest correlation with whole SSC, supporting a practical sampling strategy. Seedless and seeded watermelons showed marked transmittance differences attributable to their optical properties, and Monte Carlo simulations reproduced the differences of light attenuation, consistent with measurements. By combining standard normalized variate (SNV) with Grad-CAM for spectral preprocessing, the coefficient of determination (R<sub>p</sub><sup>2</sup>), root mean square error (RMSEP), and residual prediction deviation (RPD) in the prediction set reached 0.79, 0.55 °Brix, and 2.2, respectively. Compared to the model developed with original spectra, the RMSEP decreased by 0.14 °Brix, while R<sub>p</sub><sup>2</sup> and RPD increased by 0.11 and 0.4, respectively. Grad-CAM located and down-weighted composition-sensitive wavelength regions (chlorophyll at ∼ 680 nm and water/organic bands at ∼ 920 nm), thereby mitigating the cultivar effect on spectral modelling. The established method also casts innovative light on universal model for fruit from different origins, seasons etc.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"243 ","pages":"Article 111423"},"PeriodicalIF":8.9,"publicationDate":"2026-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145979157","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 : 2026-01-12DOI: 10.1016/j.compag.2026.111429
Yuxing Fu , Xinjie Yin , Hongcheng Zheng , Wei Fu , Bin Zhang
Coconut canopy structure is complex with dense leaves, which limits the working space for picking robots. Therefore, the recognition of canopy targets (coconut clusters and their stalks, leaves) and the location of picking points are critical to achieve intelligent picking of coconut clusters. To address this problem, we created a dataset for canopy multi-target segmentation and introduced four advanced modules to enhance YOLOv8-seg accuracy in complex canopies. C2F with omni-dimensional dynamic convolution (C2F_OD) and spatial-channel decoupled downsampling (SCDown) form a new backbone network, enhancing context information extraction and target localization through omni-dimensional attention, and improving the network’s ability to recognize multi-scale and multi-shape targets. The augmented neck network incorporating dynamic upsampler (Dysample) and lightweight bi-directional feature pyramid network (BiFPN-n) respectively enhances the semantic utilization of low-resolution feature map through content-aware upsampling and enhance feature fusion through the weight distribution of contributions, ultimately improving the accuracy of multi-target morphological segmentation. The proposed picking point location algorithm optimizes the contour structure of the stalk mask by convex hull processing, and uses the rotating calipers algorithm for optimal morphological fitting to accurately locate the picking point. The experimental results show that precision (P), recall (R) and mean average precision (mAP) of CLS-seg achieves 88.2 %, 86.4 % and 91.3 % respectively, which is superior to the existing models. Parameters, FLOPs and FPS are 3.0 M, 11.2G and 63.7 respectively, making target segmentation more efficient. The morphological fitting degree of the picking point location algorithm is 74.7 %, and the success rate of picking point location is 91.6 %. This study provides a foundation for the development of the perception system and automatic picking of intelligent coconut picking devices, and also offers a reference for predicting picking points of other fruits and vegetables with similar growth characteristics.
{"title":"Multi-target recognition and picking point location for intelligent coconut picking","authors":"Yuxing Fu , Xinjie Yin , Hongcheng Zheng , Wei Fu , Bin Zhang","doi":"10.1016/j.compag.2026.111429","DOIUrl":"10.1016/j.compag.2026.111429","url":null,"abstract":"<div><div>Coconut canopy structure is complex with dense leaves, which limits the working space for picking robots. Therefore, the recognition of canopy targets (coconut clusters and their stalks, leaves) and the location of picking points are critical to achieve intelligent picking of coconut clusters. To address this problem, we created a dataset for canopy multi-target segmentation and introduced four advanced modules to enhance YOLOv8-seg accuracy in complex canopies. C2F with omni-dimensional dynamic convolution (C2F_OD) and spatial-channel decoupled downsampling (SCDown) form a new backbone network, enhancing context information extraction and target localization through omni-dimensional attention, and improving the network’s ability to recognize multi-scale and multi-shape targets. The augmented neck network incorporating dynamic upsampler (Dysample) and lightweight bi-directional feature pyramid network (BiFPN-n) respectively enhances the semantic utilization of low-resolution feature map through content-aware upsampling and enhance feature fusion through the weight distribution of contributions, ultimately improving the accuracy of multi-target morphological segmentation. The proposed picking point location algorithm optimizes the contour structure of the stalk mask by convex hull processing, and uses the rotating calipers algorithm for optimal morphological fitting to accurately locate the picking point. The experimental results show that precision (P), recall (R) and mean average precision (mAP) of CLS-seg achieves 88.2 %, 86.4 % and 91.3 % respectively, which is superior to the existing models. Parameters, FLOPs and FPS are 3.0 M, 11.2G and 63.7 respectively, making target segmentation more efficient. The morphological fitting degree of the picking point location algorithm is 74.7 %, and the success rate of picking point location is 91.6 %. This study provides a foundation for the development of the perception system and automatic picking of intelligent coconut picking devices, and also offers a reference for predicting picking points of other fruits and vegetables with similar growth characteristics.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"243 ","pages":"Article 111429"},"PeriodicalIF":8.9,"publicationDate":"2026-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145979158","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 : 2026-01-11DOI: 10.1016/j.compag.2026.111415
Hassan A.A. Sayed , Qinghui Lai , Jielei Tu , Mahmoud A. Abdelhamid , Tarek Kh. Abdelkader , T.M. Tawfik , Rasheed Olalekan Olajide , Misbah Uddin , Mohamed Refai
The integration of solar energy with agricultural robots presents a revolutionary approach to sustainable and autonomous farming by providing a clean, renewable energy source for field activities. Despite the considerable promise, a thorough understanding of the feasibility, scalability, and synergistic integration of these systems remains insufficiently investigated. This comprehensive study consolidates the existing knowledge on solar-powered agricultural robots, assessing their technological effectiveness, economic viability, and environmental consequences. In accordance with the PRISMA-ScR criteria, we reviewed literature from primary databases up to 2026. Our findings are organized according to a functional taxonomy of robotic platforms, photovoltaic (PV) integration methods, and particular agricultural applications. The assessment highlights significant advancements in energy independence, showcasing successful prototypes that demonstrate operational durability and reduced carbon emissions. Nevertheless, significant obstacles remain, including energy storage constraints, substantial initial capital requirements, performance variability under diverse agronomic conditions, and inadequate recycling mechanisms. There are also still limitations regarding the impact of the weight of solar cells and batteries on the performance of agricultural robots. The future progress of this sector depends on the development of more efficient and adaptable PV technologies, resilient economic models, sophisticated energy management systems, and robots capable of performing intricate tasks in unstructured environments. This assessment positions solar-powered agribots not only as automation tools but as essential facilitators of a resilient, data-driven, and low-carbon agricultural framework.
{"title":"Solar-powered agricultural robots: a systematic review of technological synergies, sustainability impacts, and future pathways for autonomous farming","authors":"Hassan A.A. Sayed , Qinghui Lai , Jielei Tu , Mahmoud A. Abdelhamid , Tarek Kh. Abdelkader , T.M. Tawfik , Rasheed Olalekan Olajide , Misbah Uddin , Mohamed Refai","doi":"10.1016/j.compag.2026.111415","DOIUrl":"10.1016/j.compag.2026.111415","url":null,"abstract":"<div><div>The integration of solar energy with agricultural robots presents a revolutionary approach to sustainable and autonomous farming by providing a clean, renewable energy source for field activities. Despite the considerable promise, a thorough understanding of the feasibility, scalability, and synergistic integration of these systems remains insufficiently investigated. This comprehensive study consolidates the existing knowledge on solar-powered agricultural robots, assessing their technological effectiveness, economic viability, and environmental consequences. In accordance with the PRISMA-ScR criteria, we reviewed literature from primary databases up to 2026. Our findings are organized according to a functional taxonomy of robotic platforms, photovoltaic (PV) integration methods, and particular agricultural applications. The assessment highlights significant advancements in energy independence, showcasing successful prototypes that demonstrate operational durability and reduced carbon emissions. Nevertheless, significant obstacles remain, including energy storage constraints, substantial initial capital requirements, performance variability under diverse agronomic conditions, and inadequate recycling mechanisms. There are also still limitations regarding the impact of the weight of solar cells and batteries on the performance of agricultural robots. The future progress of this sector depends on the development of more efficient and adaptable PV technologies, resilient economic models, sophisticated energy management systems, and robots capable of performing intricate tasks in unstructured environments. This assessment positions solar-powered agribots not only as automation tools but as essential facilitators of a resilient, data-driven, and low-carbon agricultural framework.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"243 ","pages":"Article 111415"},"PeriodicalIF":8.9,"publicationDate":"2026-01-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145979072","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 : 2026-01-10DOI: 10.1016/j.compag.2026.111414
Ziheng Feng , Ke Wu , Bo Xu , Haiyan Cen , Yuan Zhang , Xiangtai Jiang , Hanyu Xue , Heguang Sun , Hao Yang , Haikuan Feng , Huiling Long , Xingang Xu , Yuanyuan Fu , Changbin Liu , Xiangyuan Wan , Guijun Yang
Leaf area index (LAI) is an important structural parameter of crops and it is usually estimated non-destructively using reflectance spectra from various reflectometers. Prevailing models, often trained on single-crop and single-year data, lack generalizability. As rotation crops with similar morphology, rice and wheat present an opportunity to develop generalized models; however, their spectral response patterns are not well compared, and adaptable multi-year, multi-crop LAI models remain scarce. To bridge this gap, we developed a generalized LAI estimation model for both crops by integrating physically-based simulation with data-driven deep learning. Key steps included canopy spectral simulation, data augmentation, and model construction with a 1D-CNN and transfer learning. The PROSAIL model was employed to simulate canopy reflectance spectra, with crop growth stages stratified into two phenological phases: sowing-heading stage (LAI: 0.01–5, increment: 0.2) and heading-grouting stage (LAI: 3–8, increment: 0.2). To enhance ecological fidelity, the LSMM was integrated to simulate mixed spectral scenarios involving soil background, water interactions, and spike contributions, while 5% Gaussian noise was systematically introduced to approximate real-world environmental variability. The results showed that the R2 values of the SMOTE-1D-CNN model for the different datasets (four rice and two wheat) ranged from 0.62 to 0.87, and the RMSE values ranged from 0.55 to 1.22. The model achieved a relatively high R2 (0.79 ± 0.09) for rice LAI estimation but exhibited a larger RMSE (0.8 ± 0.29). For wheat, the R2 was slightly lower (0.74 ± 0.17), while the RMSE was smaller and more stable (0.56 ± 0.01). These discrepancies reflect how crop characteristics or data distribution may influence estimation accuracy. SMOTE is used as a data enhancement to reduce the “high underestimation” phenomenon of the model, and the model performance is stabilized when the multiplicity of the sample size (n) is greater than or equal to 5. And the model input feature importance is only related to the original sample (the original unenhanced dataset) and does not change with “n”. This study demonstrates that a hybrid methodology, fusing physically-based simulation with deep learning, offers significant potential for robust, multi-crop LAI inversion, providing novel insights and technical support for crop monitoring and management.
{"title":"Utilization of synthetic minority oversampling technique and transfer learning for improving rice and wheat LAI estimation","authors":"Ziheng Feng , Ke Wu , Bo Xu , Haiyan Cen , Yuan Zhang , Xiangtai Jiang , Hanyu Xue , Heguang Sun , Hao Yang , Haikuan Feng , Huiling Long , Xingang Xu , Yuanyuan Fu , Changbin Liu , Xiangyuan Wan , Guijun Yang","doi":"10.1016/j.compag.2026.111414","DOIUrl":"10.1016/j.compag.2026.111414","url":null,"abstract":"<div><div>Leaf area index (LAI) is an important structural parameter of crops and it is usually estimated non-destructively using reflectance spectra from various reflectometers. Prevailing models, often trained on single-crop and single-year data, lack generalizability. As rotation crops with similar morphology, rice and wheat present an opportunity to develop generalized models;<!--> <!-->however, their spectral response patterns are not well compared, and adaptable multi-year, multi-crop LAI models remain scarce. To bridge this gap, we developed a generalized LAI estimation model for both crops by integrating physically-based simulation with data-driven deep learning. Key steps included canopy spectral simulation, data augmentation, and model construction with a 1D-CNN and transfer learning. The PROSAIL model was employed to simulate canopy reflectance spectra, with crop growth stages stratified into two phenological phases: sowing-heading stage (LAI: 0.01–5, increment: 0.2) and heading-grouting stage (LAI: 3–8, increment: 0.2). To enhance ecological fidelity, the LSMM was integrated to simulate mixed spectral scenarios involving soil background, water interactions, and spike contributions, while 5% Gaussian noise was systematically introduced to approximate real-world environmental variability. The results showed that the R<sup>2</sup> values of the SMOTE-1D-CNN model for the different datasets (four rice and two wheat) ranged from 0.62 to 0.87, and the RMSE values ranged from 0.55 to 1.22. The model achieved a relatively high R<sup>2</sup> (0.79 ± 0.09) for rice LAI estimation but exhibited a larger RMSE (0.8 ± 0.29). For wheat, the R<sup>2</sup> was slightly lower (0.74 ± 0.17), while the RMSE was smaller and more stable (0.56 ± 0.01). These discrepancies reflect how crop characteristics or data distribution may influence estimation accuracy. SMOTE is used as a data enhancement to reduce the “high underestimation” phenomenon of the model, and the model performance is stabilized when the multiplicity of the sample size (n) is greater than or equal to 5. And the model input feature importance is only related to the original sample (the original unenhanced dataset) and does not change with “n”. This study demonstrates that a hybrid methodology, fusing physically-based simulation with deep learning,<!--> <!-->offers significant potential for robust, multi-crop LAI inversion, providing novel insights and technical support for crop monitoring and management.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"243 ","pages":"Article 111414"},"PeriodicalIF":8.9,"publicationDate":"2026-01-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145927540","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