Pub Date : 2026-03-15Epub Date: 2026-02-05DOI: 10.1016/j.compag.2026.111439
Baoxian Liang , Lihong Xu , Yu Su , Jianwei Du , Zhichao Deng
Efficient and scalable multi-task allocation presents a fundamental challenge in multi-machine cooperative operation for unmanned farming. Conventional approaches often assume static attributes and fixed-scale instances, thereby facing significant challenges in adapting to the dynamic characteristics of agricultural production processes. To address the dynamic multi-task allocation problem with time windows (DMAPTW), we propose a novel RL framework that automatically learns high-quality scheduling policies. A scale-agnostic representation mechanism is designed to accurately reflect the current system status, ensuring that the derived policy network is scale-agnostic. To enhance adaptability across diverse production environments, a combination method integrating problem-specific dispatching rules is implemented. Concurrently, a dense reward mechanism is proposed to directly associate the optimization objective. Numerical experiments conducted on a comprehensive set of synthetic instances demonstrate that the proposed algorithm exhibits robust flexibility in handling varying production configurations. Furthermore, comparative analyses reveal that this algorithm consistently outperforms meta-heuristic baselines by 28%–40%, indicating superior computational efficiency and robustness.
{"title":"Deep reinforcement learning for unmanned farming dynamic multi-task allocation problem","authors":"Baoxian Liang , Lihong Xu , Yu Su , Jianwei Du , Zhichao Deng","doi":"10.1016/j.compag.2026.111439","DOIUrl":"10.1016/j.compag.2026.111439","url":null,"abstract":"<div><div>Efficient and scalable multi-task allocation presents a fundamental challenge in multi-machine cooperative operation for unmanned farming. Conventional approaches often assume static attributes and fixed-scale instances, thereby facing significant challenges in adapting to the dynamic characteristics of agricultural production processes. To address the dynamic multi-task allocation problem with time windows (DMAPTW), we propose a novel RL framework that automatically learns high-quality scheduling policies. A scale-agnostic representation mechanism is designed to accurately reflect the current system status, ensuring that the derived policy network is scale-agnostic. To enhance adaptability across diverse production environments, a combination method integrating problem-specific dispatching rules is implemented. Concurrently, a dense reward mechanism is proposed to directly associate the optimization objective. Numerical experiments conducted on a comprehensive set of synthetic instances demonstrate that the proposed algorithm exhibits robust flexibility in handling varying production configurations. Furthermore, comparative analyses reveal that this algorithm consistently outperforms meta-heuristic baselines by 28%–40%, indicating superior computational efficiency and robustness.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"244 ","pages":"Article 111439"},"PeriodicalIF":8.9,"publicationDate":"2026-03-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146173989","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 : 2026-03-01Epub Date: 2026-01-08DOI: 10.1016/j.compag.2026.111412
Dror Ettlinger-Levy , Shai Kendler , Iris Meiri Ashkenazi , Shay Tal , Barak Fishbain
Accurate feed management in marine aquaculture is critical for maximizing fish growth and minimizing environmental impacts. While previous approaches have leveraged acoustic monitoring and neural network-based classification to assess feeding, these methods often lack scalability and precision for industrial deployment. This study introduces a matched-filter audio signal processing technique, informed by domain knowledge, to continuously quantify fish feeding intensity in gilthead seabream (Sparus aurata) aquaculture. By extracting a species-specific bite acoustic template and applying matched filtering with sliding window aggregation, we generate a robust, continuous feeding intensity label from passive acoustic recordings. Machine learning regression models (XGBoost and Random Forest) validate the approach, demonstrating that environmental and biological variables explain 89% of the variation in feeding intensity. The proposed methodology offers a simple, scalable, and cost-effective solution for real-time feed optimization and welfare monitoring in aquaculture systems. By reducing data dimensionality and enhancing sensitivity to subtle behavioral changes, this framework supports the deployment of advanced data-driven monitoring tools and paves the way for practical integration in commercial aquaculture operations.
{"title":"Optimized aquaculture feeding through matched-filter audio signal processing and machine learning","authors":"Dror Ettlinger-Levy , Shai Kendler , Iris Meiri Ashkenazi , Shay Tal , Barak Fishbain","doi":"10.1016/j.compag.2026.111412","DOIUrl":"10.1016/j.compag.2026.111412","url":null,"abstract":"<div><div>Accurate feed management in marine aquaculture is critical for maximizing fish growth and minimizing environmental impacts. While previous approaches have leveraged acoustic monitoring and neural network-based classification to assess feeding, these methods often lack scalability and precision for industrial deployment. This study introduces a matched-filter audio signal processing technique, informed by domain knowledge, to continuously quantify fish feeding intensity in gilthead seabream (Sparus aurata) aquaculture. By extracting a species-specific bite acoustic template and applying matched filtering with sliding window aggregation, we generate a robust, continuous feeding intensity label from passive acoustic recordings. Machine learning regression models (XGBoost and Random Forest) validate the approach, demonstrating that environmental and biological variables explain 89% of the variation in feeding intensity. The proposed methodology offers a simple, scalable, and cost-effective solution for real-time feed optimization and welfare monitoring in aquaculture systems. By reducing data dimensionality and enhancing sensitivity to subtle behavioral changes, this framework supports the deployment of advanced data-driven monitoring tools and paves the way for practical integration in commercial aquaculture operations.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"243 ","pages":"Article 111412"},"PeriodicalIF":8.9,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145927534","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 : 2026-03-01Epub Date: 2026-01-05DOI: 10.1016/j.compag.2025.111398
Jing Xu , Xuemin Zhang , Xiaoyan Wang , Hao Song , Yajuan Wang
Accurate and efficient detection of the pollination status of strawberry flowers is essential for intelligent pollination robots, as it directly affects the determination of optimal pollination timing and improves fruit set rates. However, the small size of strawberry anthers, their visual similarity, varied opening states, and complex field environments make pollination status detection highly formidable. To overcome these constraints, this paper presents a streamlined and resource-efficient detection approach (ELSF-DETR), built upon the Real-Time DEtection Transformer (RT-DETR) and specially refined for detecting densely packed and visually similar small objects in agricultural scenes. A lightweight LS-ResNet backbone is constructed to better capture small and densely clustered anther structures in strawberry flowers while reducing model complexity for improved deployment efficiency. In addition, the integration of a P2 detection head with full-kernel convolution enhance the network’s capacity to focus on delicate anther contours and cracking characteristics. Furthermore, the Hierarchical Attention Fusion Block (HAFB) is employed to balance local detail extraction with global context understanding, reducing misjudgments caused by misleading fine-grained features. Lastly, by employing the Wise-IoU (WIoU) loss mechanism, the model achieves improved sensitivity to minor positional discrepancies in visually similar anther objects. Experiments conducted on a self-built strawberry flower dataset demonstrate that ELSF-DETR achieves superior performance, it achieves 88.2 % accuracy, 85.8 % recall, 87.1 % mAP@50, and F1 score of 86.98 %. Relative to the baseline architecture, mAP@50 and F1 improved by 7.1 % and 4.33 %, respectively, while the model parameters and GFLOPs were reduced by 6.86 MB and 13.7 G, meeting the requirements of high precision and low complexity. This work provides practical support for intelligent pollination systems in precision agriculture.
{"title":"ELSF-DETR: an efficient lightweight network for detecting strawberry flowers pollination status in non-structured greenhouse environments","authors":"Jing Xu , Xuemin Zhang , Xiaoyan Wang , Hao Song , Yajuan Wang","doi":"10.1016/j.compag.2025.111398","DOIUrl":"10.1016/j.compag.2025.111398","url":null,"abstract":"<div><div>Accurate and efficient detection of the pollination status of strawberry flowers is essential for intelligent pollination robots, as it directly affects the determination of optimal pollination timing and improves fruit set rates. However, the small size of strawberry anthers, their visual similarity, varied opening states, and complex field environments make pollination status detection highly formidable. To overcome these constraints, this paper presents a streamlined and resource-efficient detection approach (ELSF-DETR), built upon the Real-Time DEtection Transformer (RT-DETR) and specially refined for detecting densely packed and visually similar small objects in agricultural scenes. A lightweight LS-ResNet backbone is constructed to better capture small and densely clustered anther structures in strawberry flowers while reducing model complexity for improved deployment efficiency. In addition, the integration of a P2 detection head with full-kernel convolution enhance the network’s capacity to focus on delicate anther contours and cracking characteristics. Furthermore, the Hierarchical Attention Fusion Block (HAFB) is employed to balance local detail extraction with global context understanding, reducing misjudgments caused by misleading fine-grained features. Lastly, by employing the Wise-IoU (WIoU) loss mechanism, the model achieves improved sensitivity to minor positional discrepancies in visually similar anther objects. Experiments conducted on a self-built strawberry flower dataset demonstrate that ELSF-DETR achieves superior performance, it achieves 88.2 % accuracy, 85.8 % recall, 87.1 % mAP@50, and F1 score of 86.98 %. Relative to the baseline architecture, mAP@50 and F1 improved by 7.1 % and 4.33 %, respectively, while the model parameters and GFLOPs were reduced by 6.86 MB and 13.7 G, meeting the requirements of high precision and low complexity. This work provides practical support for intelligent pollination systems in precision agriculture.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"243 ","pages":"Article 111398"},"PeriodicalIF":8.9,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145927632","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 : 2026-03-01Epub Date: 2025-12-27DOI: 10.1016/j.compag.2025.111347
Huibin Li , Jianyu Zhu , Xing Mao , Xueli Hao , Shiyao Li , Qiangyi Yu , Yun Shi , Jianping Qian
The efficient extraction of cropland parcels from satellite imagery is of crucial importance for modern agricultural management. The advent of the Segment Anything Model (SAM) presents the potential to reduce the need for annotations and complex training in the context of cropland extraction. However, SAM faces challenges in handling diverse and heterogeneous cropland types. To address these issues, this study proposes a novel, unsupervised methodology that integrates SAM with an adaptive mask refinement strategy, enabling accurate cropland extraction under minimal supervision. The refinement strategy comprises three key components: (1) an adaptive prompt point module that leverages superpixels to dynamically generate optimised prompt points, (2) an overlap filtering module to eliminate redundant cropland parcels and (3) a boundary-matching stitching module to maintain spatial continuity across image tiles. The efficacy of the method was evaluated using diverse satellite images (∼160 km2) from seven representative regions in China, the United States, and South Africa. Ablation experiment results showed that the proposed approach achieved notable improvements over the baseline SAM, with increases in recall (R), Intersection over Union (IoU) and global total classification errors (GTC) of 0.971, 0.908 and 0.124, respectively. Furthermore, it outperformed five contemporary state-of-the-art methods, achieving a precision (P) of 0.960. The method also generalised well across different cropland configurations, ranging from large, regular parcels (e.g. Xinjiang, Illinois) to fragmented landscapes (e.g. Guangdong, Western Cape). Seasonal analysis confirmed that images captured during the sowing period yielded the highest extraction accuracy. These findings highlight the potential of SAM-based approaches for scalable and accurate cropland parcel mapping in complex agricultural landscapes under low-supervision settings.
{"title":"Achieving precise cropland parcel extraction from remote sensing images through integration of segment anything model and adaptive mask refinement","authors":"Huibin Li , Jianyu Zhu , Xing Mao , Xueli Hao , Shiyao Li , Qiangyi Yu , Yun Shi , Jianping Qian","doi":"10.1016/j.compag.2025.111347","DOIUrl":"10.1016/j.compag.2025.111347","url":null,"abstract":"<div><div>The efficient extraction of cropland parcels from satellite imagery is of crucial importance for modern agricultural management. The advent of the Segment Anything Model (SAM) presents the potential to reduce the need for annotations and complex training in the context of cropland extraction. However, SAM faces challenges in handling diverse and heterogeneous cropland types. To address these issues, this study proposes a novel, unsupervised methodology that integrates SAM with an adaptive mask refinement strategy, enabling accurate cropland extraction under minimal supervision. The refinement strategy comprises three key components: (1) an adaptive prompt point module that leverages superpixels to dynamically generate optimised prompt points, (2) an overlap filtering module to eliminate redundant cropland parcels and (3) a boundary-matching stitching module to maintain spatial continuity across image tiles. The efficacy of the method was evaluated using diverse satellite images (∼160 <!--> <!-->km<sup>2</sup>) from seven representative regions in China, the United States, and South Africa. Ablation experiment results showed that the proposed approach achieved notable improvements over the baseline SAM, with increases in recall (R), Intersection over Union (IoU) and global total classification errors (GTC) of 0.971, 0.908 and 0.124, respectively. Furthermore, it outperformed five contemporary state-of-the-art methods, achieving a precision (P) of 0.960. The method also generalised well across different cropland configurations, ranging from large, regular parcels (e.g. Xinjiang, Illinois) to fragmented landscapes (e.g. Guangdong, Western Cape). Seasonal analysis confirmed that images captured during the sowing period yielded the highest extraction accuracy. These findings highlight the potential of SAM-based approaches for scalable and accurate cropland parcel mapping in complex agricultural landscapes under low-supervision settings.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"243 ","pages":"Article 111347"},"PeriodicalIF":8.9,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145842812","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 : 2026-03-01Epub Date: 2025-12-31DOI: 10.1016/j.compag.2025.111385
Xindong Lai , Jianzhi Huang , Yongmei Mo , Hongwei Li , Tianyun Dong , Tao Wu , Deqiang He
Accurate navigation line extraction is fundamental for agricultural robots. However, most existing studies have not explored the synergistic potential of integrating detection and segmentation on navigation line extraction. To address this gap, this study proposes a model named YOLO for Synchronized Stem-Row (YOLO-SSR). The core contributions of our work include a lightweight dual-task network for perception and an adaptive fusion pipeline for navigation. The proposed model simultaneously detects crop stems and segments crop rows. Its architecture incorporates several key innovations. Specifically, tri-path adaptive convolution (TriPAC) modules are integrated into its backbone to facilitate efficient multi-scale feature capture. The detection branch is augmented with space-to-depth convolution (SPDC) to enrich shallow features and dynamic group shuffle transformer (DGST) to refine contextual information. The minimalist segmentation branch reuses optimized features from the detection neck, ensuring high computational efficiency. Furthermore, an adaptive fusion pipeline is developed to precisely extract navigation lines by integrating detection and segmentation outputs. Comparative experiments demonstrate that YOLO-SSR achieves a competitive performance (AP50: 67.2 %, mIoU: 89.71 %), while maintaining a lightweight architecture (2.45 M parameters, 10.5 GFLOPs). Notably, its real-time processing capability is validated on Nvidia Jetson Orin Nano (27.2 FPS), indicating its suitability for resource-constrained edge devices. Moreover, the fused navigation lines yield a mean normalized lateral distance of 0.72 %, outperforming the results obtained from either task individually. This study provides new insights to explore agricultural navigation line extraction, which can further enrich the theoretical and technical foundations for visual navigation of agricultural robots.
{"title":"A lightweight deep learning model for synchronized crop stem detection and row segmentation at the seedling stage: Exploring their contribution to agricultural navigation line extraction","authors":"Xindong Lai , Jianzhi Huang , Yongmei Mo , Hongwei Li , Tianyun Dong , Tao Wu , Deqiang He","doi":"10.1016/j.compag.2025.111385","DOIUrl":"10.1016/j.compag.2025.111385","url":null,"abstract":"<div><div>Accurate navigation line extraction is fundamental for agricultural robots. However, most existing studies have not explored the synergistic potential of integrating detection and segmentation on navigation line extraction. To address this gap, this study proposes a model named YOLO for Synchronized Stem-Row (YOLO-SSR). The core contributions of our work include<!--> <!-->a lightweight dual-task network for perception and an adaptive fusion pipeline for navigation. The proposed model simultaneously detects crop stems and segments crop rows. Its architecture incorporates several key innovations. Specifically, tri-path adaptive convolution (TriPAC) modules are integrated into its backbone to facilitate efficient multi-scale feature capture. The detection branch is augmented with space-to-depth convolution (SPDC) to enrich shallow features and dynamic group shuffle transformer (DGST) to refine contextual information. The minimalist segmentation branch reuses optimized features from the detection neck, ensuring high computational efficiency. Furthermore, an adaptive fusion pipeline is developed to precisely extract navigation lines by integrating detection and segmentation outputs. Comparative experiments demonstrate that YOLO-SSR achieves a competitive performance (AP<sub>50</sub>: 67.2 %, mIoU: 89.71 %), while maintaining a lightweight architecture (2.45 M parameters, 10.5 GFLOPs). Notably, its real-time processing capability is validated on Nvidia Jetson Orin Nano (27.2 FPS), indicating its suitability for resource-constrained edge devices. Moreover, the fused navigation lines yield a mean normalized lateral distance of 0.72 %, outperforming the results obtained from either task individually. This study provides new insights to explore agricultural navigation line extraction, which can further enrich the theoretical and technical foundations for visual navigation of agricultural robots.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"243 ","pages":"Article 111385"},"PeriodicalIF":8.9,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145885846","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 assessment of mold contamination levels in wheat is crucial for ensuring its safe storage and quality. This study developed an olfactory analyzer based on a quartz crystal microbalance (QCM) gas sensor to quantify mold contamination levels in wheat during storage. Specifically, based on the characteristic volatiles identified in wheat during different stages of mold growth, namely, 1-octen-3-ol, 4-methyl-2-pentanone, and 3-octanone, anti-humidity molecularly imprinted polymer (MIP) composites were synthesized for modifying the QCM sensor. The prepared QCM sensor showed anti-humidity interference characteristics along with excellent sensitivity and selectivity for target molecules. In addition, the rational and precise air pathway design of the olfactory analyzer contributed to shorter response and recovery times. Using the number of mold colonies in wheat during different stages of mold growth as a physicochemical parameter for quantifying mold levels, we established a regression model to predict wheat mold contamination levels. Compared with the PLSR and RFR models, the NN model showed optimal predictive performance: the R2 and RMSE values of the testing set are 0.94 and 0.1998, respectively, and the RPD value is 4.15. Finally, the model was embedded into the olfactory analyzer. This study provides technical insights for developing high-performance gas sensors to detect early signs of mold contamination in wheat.
{"title":"Development of an olfactory analyzer based on hydrophobic quartz crystal microbalance sensors for predicting contamination levels in moldy wheat","authors":"Yuxin Hou , Shangjun Wang , Jianfei Xing , Linjie Zhou , Huaimeng Chen , Xiuying Tang","doi":"10.1016/j.compag.2025.111393","DOIUrl":"10.1016/j.compag.2025.111393","url":null,"abstract":"<div><div>Precise assessment of mold contamination levels in wheat is crucial for ensuring its safe storage and quality. This study developed an olfactory analyzer based on a quartz crystal microbalance (QCM) gas sensor to quantify mold contamination levels in wheat during storage. Specifically, based on the characteristic volatiles identified in wheat during different stages of mold growth, namely, 1-octen-3-ol, 4-methyl-2-pentanone, and 3-octanone, anti-humidity molecularly imprinted polymer (MIP) composites were synthesized for modifying the QCM sensor. The prepared QCM sensor showed anti-humidity interference characteristics along with excellent sensitivity and selectivity for target molecules. In addition, the rational and precise air pathway design of the olfactory analyzer contributed to shorter response and recovery times. Using the number of mold colonies in wheat during different stages of mold growth as a physicochemical parameter for quantifying mold levels, we established a regression model to predict wheat mold contamination levels. Compared with the PLSR and RFR models, the NN model showed optimal predictive performance: the R<sup>2</sup> and RMSE values of the testing set are 0.94 and 0.1998, respectively, and the RPD value is 4.15. Finally, the model was embedded into the olfactory analyzer. This study provides technical insights for developing high-performance gas sensors to detect early signs of mold contamination in wheat.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"243 ","pages":"Article 111393"},"PeriodicalIF":8.9,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145885847","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 : 2026-03-01Epub Date: 2025-12-27DOI: 10.1016/j.compag.2025.111377
Zhenyang Hui , Yating He , Shuanggen Jin , Wenbo Chen , Haiqing He , Yao Yevenyo Ziggah
Leaves play a crucial role in the growth of plants, both functionally and structurally. To meet the requirements of various levels of detail (LoDs) in leaf modeling for different applications, this paper introduces a self-adaptive 3D leaf modeling method aimed at enhancing LoDs representation. In this paper, a self-adaptive leaf axis determination method is first presented. According to the built leaf axis, feature points including contour points, inner corners, and outer corners are identified. Subsequently, based on these feature points, a multi-level veins generation model is proposed, extracting primary, secondary, and tertiary veins individually by leveraging the geometric and morphological traits of the leaf through a spatial colonization strategy. Hereafter, the three-dimensional leaf modeling achieves different LoDs by incorporating varying degrees of vein structures. To evaluate the effectiveness of the proposed method, both simulated and real datasets are utilized for testing. The simulated datasets consist of leaves from four different types, such as entire, toothed, disercted and digitate demonstrating that the method produces satisfactory results with small area deviation and distance residuals. In the real datasets, seven individual tomatoes with a total of 228 leaves are tested, showing that the proposed modeling approach aligns effectively with real data, with distance residuals mostly falling within -0.4 cm to 0.4 cm from real point clouds. Experimental results also reveal that higher levels of modeling lead to better outcomes due to increased detail from additional veins and feature points incorporated in the modeling process.
{"title":"LeafLoDs: A Self-Adaptive 3-D leaf modeling with enhancing level of details expression","authors":"Zhenyang Hui , Yating He , Shuanggen Jin , Wenbo Chen , Haiqing He , Yao Yevenyo Ziggah","doi":"10.1016/j.compag.2025.111377","DOIUrl":"10.1016/j.compag.2025.111377","url":null,"abstract":"<div><div>Leaves play a crucial role in the growth of plants, both functionally and structurally. To meet the requirements of various levels of detail (LoDs) in leaf modeling for different applications, this paper introduces a self-adaptive 3D leaf modeling method aimed at enhancing LoDs representation. In this paper, a self-adaptive leaf axis determination method is first presented. According to the built leaf axis, feature points including contour points, inner corners, and outer corners are identified. Subsequently, based on these feature points, a multi-level veins generation model is proposed, extracting primary, secondary, and tertiary veins individually by leveraging the geometric and morphological traits of the leaf through a spatial colonization strategy. Hereafter, the three-dimensional leaf modeling achieves different LoDs by incorporating varying degrees of vein structures. To evaluate the effectiveness of the proposed method, both simulated and real datasets are utilized for testing. The simulated datasets consist of leaves from four different types, such as entire, toothed, disercted and digitate demonstrating that the method produces satisfactory results with small area deviation and distance residuals. In the real datasets, seven individual tomatoes with a total of 228 leaves are tested, showing that the proposed modeling approach aligns effectively with real data, with distance residuals mostly falling within -0.4 cm to 0.4 cm from real point clouds. Experimental results also reveal that higher levels of modeling lead to better outcomes due to increased detail from additional veins and feature points incorporated in the modeling process.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"243 ","pages":"Article 111377"},"PeriodicalIF":8.9,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145885969","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 : 2026-03-01Epub Date: 2025-12-27DOI: 10.1016/j.compag.2025.111334
Xuechen Li , Alireza Sanaeifar , Nicholas Padilla , Cole Stover , Alec Kowalewski , Eric Watkins , Bryan Runck , Lang Qiao , Ce Yang
Accurate detection of winter damage in turfgrass is essential for proactive management but remains difficult because early-stage injury is faint, irregular, and easily confused with background noise. These characteristics create two major challenges: limited availability of reliable training data and the need for a segmentation model that is highly sensitive to subtle features. To address the data limitation, this study employs a Conditional Deep Convolutional Generative Adversarial Network (cDCGAN) to generate synthetic, high-fidelity vegetation index (VI) maps. Compared with raw spectral bands, VIs are more robust to noise and enhance both dataset diversity and model generalization. To meet the segmentation challenge, we introduce a Transformer-based model with a novel Adaptive Attention Decoder (AAD), which dynamically refines feature representations to improve detection of low-contrast, spatially irregular damage. Field experiments conducted on golf courses in central Oregon, USA, from 2022 to 2023 demonstrate that the proposed pipeline outperforms other advanced deep learning models, achieving an mIoU of 82.47%, an accuracy of 97.85%, a recall of 85.62%, and an F1-score of 88.30%. Overall, this research presents a problem-driven framework that integrates targeted data augmentation with an improved segmentation architecture, offering a robust and accurate solution for early detection of winter damage in precision turfgrass management.
{"title":"Winter damage diagnostic modeling based on synthetic vegetation indices from UAV-based multispectral imaging","authors":"Xuechen Li , Alireza Sanaeifar , Nicholas Padilla , Cole Stover , Alec Kowalewski , Eric Watkins , Bryan Runck , Lang Qiao , Ce Yang","doi":"10.1016/j.compag.2025.111334","DOIUrl":"10.1016/j.compag.2025.111334","url":null,"abstract":"<div><div>Accurate detection of winter damage in turfgrass is essential for proactive management but remains difficult because early-stage injury is faint, irregular, and easily confused with background noise. These characteristics create two major challenges: limited availability of reliable training data and the need for a segmentation model that is highly sensitive to subtle features. To address the data limitation, this study employs a Conditional Deep Convolutional Generative Adversarial Network (cDCGAN) to generate synthetic, high-fidelity vegetation index (VI) maps. Compared with raw spectral bands, VIs are more robust to noise and enhance both dataset diversity and model generalization. To meet the segmentation challenge, we introduce a Transformer-based model with a novel Adaptive Attention Decoder (AAD), which dynamically refines feature representations to improve detection of low-contrast, spatially irregular damage. Field experiments conducted on golf courses in central Oregon, USA, from 2022 to 2023 demonstrate that the proposed pipeline outperforms other advanced deep learning models, achieving an mIoU of 82.47%, an accuracy of 97.85%, a recall of 85.62%, and an F1-score of 88.30%. Overall, this research presents a problem-driven framework that integrates targeted data augmentation with an improved segmentation architecture, offering a robust and accurate solution for early detection of winter damage in precision turfgrass management.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"243 ","pages":"Article 111334"},"PeriodicalIF":8.9,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145842804","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 : 2026-03-01Epub Date: 2025-12-27DOI: 10.1016/j.compag.2025.111374
Pengfei Zhao , Xirui Zhang , Xiaojun Gao , Junchang Zhang , Bing Qi , Hua Li , Chao Ji , Silin Cao , Congju Shen
To address the issues of unstable filling performance and poor sowing quality caused by ineffective seed clearing in mechanical soybean metering devices under high-speed operation, a novel high-speed precision metering device featuring seed diversion and pending filling functions was developed. The design introduces the concept of ’directional diversion and seed pending’ in which the sloped seed inlet guides the seeds directionally, while the double-lug hole structure enables orderly filling and temporary pending, thereby significantly enhancing the filling qualification index. Theoretical analysis was conducted to identify key structural parameters influencing seed filling, transportation, pending, and clearing. Orthogonal simulation experiments were performed to evaluate three critical parameters—perturbation angle, sidewall length, and bottom length—using the qualification index and pending index as optimization criteria. The results indicated that optimal seeding performance was achieved at a perturbation angle of − 12.07°, a sidewall length of 5.08 mm, and a bottom length of 15.28 mm. Bench validation experiments conducted at 6–10 km/h showed that the qualification index exceeded 98 %, while the pending index reached 93.47 %, representing an improvement of at least 3.9 percentage points over conventional brush-type metering devices. These results meet the operational requirements for high-speed precision seeding and offer new insights into the design of soybean metering devices.
{"title":"Computer-aided design and DEM-based simulation analysis of a diversion-type precision soybean metering device","authors":"Pengfei Zhao , Xirui Zhang , Xiaojun Gao , Junchang Zhang , Bing Qi , Hua Li , Chao Ji , Silin Cao , Congju Shen","doi":"10.1016/j.compag.2025.111374","DOIUrl":"10.1016/j.compag.2025.111374","url":null,"abstract":"<div><div>To address the issues of unstable filling performance and poor sowing quality caused by ineffective seed clearing in mechanical soybean metering devices under high-speed operation, a novel high-speed precision metering device featuring seed diversion and pending filling functions was developed. The design introduces the concept of ’directional diversion and seed pending’ in which the sloped seed inlet guides the seeds directionally, while the double-lug hole structure enables orderly filling and temporary pending, thereby significantly enhancing the filling qualification index. Theoretical analysis was conducted to identify key structural parameters influencing seed filling, transportation, pending, and clearing. Orthogonal simulation experiments were performed to evaluate three critical parameters—perturbation angle, sidewall length, and bottom length—using the qualification index and pending index as optimization criteria. The results indicated that optimal seeding performance was achieved at a perturbation angle of − 12.07°, a sidewall length of 5.08 mm, and a bottom length of 15.28 mm. Bench validation experiments conducted at 6–10 km/h showed that the qualification index exceeded 98 %, while the pending index reached 93.47 %, representing an improvement of at least 3.9 percentage points over conventional brush-type metering devices. These results meet the operational requirements for high-speed precision seeding and offer new insights into the design of soybean metering devices.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"243 ","pages":"Article 111374"},"PeriodicalIF":8.9,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145842816","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 : 2026-03-01Epub Date: 2025-12-27DOI: 10.1016/j.compag.2025.111382
Wenyi Cai , Fubo Qi , Lingyan Zha , Guanzheng Chen , Jingjin Zhang , Mengxuan Song , Hua Bao
The integration of electronic system into agricultural production can significantly enhance its efficiency and scalability. However, most of the current research focuses on the data acquisition and automated control. The development of expert-level, interpretable decision-making systems remains a challenge, primarily due to the prohibitive requirement for extensive domain-specific labeled data. In this manuscript, a novel agentic framework integrated with Large Language Models is proposed and demonstrated, using seedling assessment as a case study. The framework achieves high predictive accuracy, strong interpretability, and fast-adaption ability, offering a distinct advantage over methods that demand large labeled datasets. An agentic orchestration framework integrated with the Analytic Hierarchy Process and the reasoning ability of the Large Language Models is constructed to automatically derive the raw assessment rating. Based on a score calibration system using few-shot learning with three different types of lettuce, Butterhead, Grand Rapids, and Ramosa Hort, the final rating score can be derived with good prediction accuracy based on a small dataset (less than 20 labelled data). Additionally, three supplementary plant species (Sprout, Ball Brassica, and Rapa Brassica) are used to demonstrate the framework’s rapid adaptation capability. A field experiment guided by the agentic framework is conducted to prove that this seedling assessment system can be applied to help increase yield by more than 20 %. Our framework presents an important attempt towards an intelligent agricultural system that is capable to achieve expert-level and data-efficient decision making, thereby helping to bridge the critical gap between artificial intelligence research and practical agricultural application.
将电子系统集成到农业生产中，可以显著提高农业生产的效率和可扩展性。然而，目前的研究大多集中在数据采集和自动控制方面。开发专家级的、可解释的决策系统仍然是一个挑战，主要是由于对广泛的特定领域标记数据的限制要求。本文以幼苗评估为例，提出并论证了一种集成了大型语言模型的新型代理框架。该框架具有较高的预测精度、较强的可解释性和快速适应能力，与需要大型标记数据集的方法相比具有明显的优势。构建了结合层次分析法和大型语言模型推理能力的代理编排框架，实现了原始评价等级的自动生成。基于使用三种不同类型的莴苣（Butterhead， Grand Rapids和Ramosa Hort）的few-shot学习的评分校准系统，可以基于小数据集（少于20个标记数据）获得具有良好预测精度的最终评级分数。此外，还使用了三种补充植物（芽甘蓝、球甘蓝和油菜）来证明该框架的快速适应能力。在机构框架指导下进行了田间试验，证明该育苗评价系统可帮助增产20%以上。我们的框架提出了对智能农业系统的重要尝试，该系统能够实现专家级和数据高效的决策，从而有助于弥合人工智能研究与实际农业应用之间的关键差距。
{"title":"Few-shot and interpretable agentic framework based on large language models for data-efficient plant phenotyping","authors":"Wenyi Cai , Fubo Qi , Lingyan Zha , Guanzheng Chen , Jingjin Zhang , Mengxuan Song , Hua Bao","doi":"10.1016/j.compag.2025.111382","DOIUrl":"10.1016/j.compag.2025.111382","url":null,"abstract":"<div><div>The integration of electronic system into agricultural production can significantly enhance its efficiency and scalability. However, most of the current research focuses on the data acquisition and automated control. The development of expert-level, interpretable decision-making systems remains a challenge, primarily due to the prohibitive requirement for extensive domain-specific labeled data. In this manuscript, a novel agentic framework integrated with Large Language Models is proposed and demonstrated, using seedling assessment as a case study. The framework achieves high predictive accuracy, strong interpretability, and fast-adaption ability, offering a distinct advantage over methods that demand large labeled datasets. An agentic orchestration framework integrated with the Analytic Hierarchy Process and the reasoning ability of the Large Language Models is constructed to automatically derive the raw assessment rating. Based on a score calibration system using few-shot learning with three different types of lettuce, Butterhead, Grand Rapids, and Ramosa Hort, the final rating score can be derived with good prediction accuracy based on a small dataset (less than 20 labelled data). Additionally, three supplementary plant species (Sprout, Ball Brassica, and Rapa Brassica) are used to demonstrate the framework’s rapid adaptation capability. A field experiment guided by the agentic framework is conducted to prove that this seedling assessment system can be applied to help increase yield by more than 20 %. Our framework presents an important attempt towards an intelligent agricultural system that is capable to achieve expert-level and data-efficient decision making, thereby helping to bridge the critical gap between artificial intelligence research and practical agricultural application.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"243 ","pages":"Article 111382"},"PeriodicalIF":8.9,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145842813","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