Pub Date : 2025-04-10DOI: 10.1109/TAFE.2025.3558104
{"title":"IEEE Circuits and Systems Society Information","authors":"","doi":"10.1109/TAFE.2025.3558104","DOIUrl":"https://doi.org/10.1109/TAFE.2025.3558104","url":null,"abstract":"","PeriodicalId":100637,"journal":{"name":"IEEE Transactions on AgriFood Electronics","volume":"3 1","pages":"C3-C3"},"PeriodicalIF":0.0,"publicationDate":"2025-04-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10962279","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143817922","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-04-10DOI: 10.1109/TAFE.2025.3558100
{"title":"IEEE Circuits and Systems Society Publication Information","authors":"","doi":"10.1109/TAFE.2025.3558100","DOIUrl":"https://doi.org/10.1109/TAFE.2025.3558100","url":null,"abstract":"","PeriodicalId":100637,"journal":{"name":"IEEE Transactions on AgriFood Electronics","volume":"3 1","pages":"C2-C2"},"PeriodicalIF":0.0,"publicationDate":"2025-04-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10962283","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143821549","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-04-02DOI: 10.1109/TAFE.2025.3551072
Li Zhang;Qianyue Liang;Vijay John;Hong Chen;Shanjun Li;Weifu Li;Yaohui Chen
Citrus psyllids are common pests that feed on the sap of citrus trees, leading to yellowing, deformation, and potentially tree death in severe cases. Effective identification and monitoring of these pests are crucial for the health and sustainable development of the citrus industry. Rapid and accurate detection enables farmers to control citrus psyllid infestations promptly, thereby protecting their crops and ensuring industry sustainability. In this article, we utilize a custom-built pest-trapping device to capture the psyllids and upload the image to a server via the Internet of Things. We captured 420 images with a resolution of 3820 × 2160 using the device. These images, containing various types of pests, were utilized for model experimentation and training. On the server, the diffusion transformer (DiT) is utilized to increase the training data, addressing challenges such as limited sample size and class imbalance. A small object detection head is integrated into YOLOv10 to enhance the capture of shallow features in psyllid images. In addition, the soft nonmaximum suppression method is applied to resolve overlapping issues in counting the psyllids. Finally, the results are uploaded to an app, allowing users to stay informed about citrus pest conditions in real time. The experimental results indicate that DiTs-generated images achieved scores of 76.79, 0.29, and 1.68 in the Frechet inception distance, learned perceptual image patch similarity, and multiscale structural similarity metrics, respectively, outperforming the commonly used DDPM model by 8.51, 0.18, and 0.34, respectively. The improved YOLOv10 model, trained with the expanded DiTs dataset, reached a recall, F1-score, and precision of 90.55%, 92.18%, and 93.88%, respectively, demonstrating outstanding performance across all metrics. This approach enables fully automated recognition of citrus psyllids, facilitating real-time detection and contributing to the protection of citrus crops.
{"title":"Intelligent Psyllid Monitoring Based on DiTs-YOLOv10-SOD","authors":"Li Zhang;Qianyue Liang;Vijay John;Hong Chen;Shanjun Li;Weifu Li;Yaohui Chen","doi":"10.1109/TAFE.2025.3551072","DOIUrl":"https://doi.org/10.1109/TAFE.2025.3551072","url":null,"abstract":"Citrus psyllids are common pests that feed on the sap of citrus trees, leading to yellowing, deformation, and potentially tree death in severe cases. Effective identification and monitoring of these pests are crucial for the health and sustainable development of the citrus industry. Rapid and accurate detection enables farmers to control citrus psyllid infestations promptly, thereby protecting their crops and ensuring industry sustainability. In this article, we utilize a custom-built pest-trapping device to capture the psyllids and upload the image to a server via the Internet of Things. We captured 420 images with a resolution of 3820 × 2160 using the device. These images, containing various types of pests, were utilized for model experimentation and training. On the server, the diffusion transformer (DiT) is utilized to increase the training data, addressing challenges such as limited sample size and class imbalance. A small object detection head is integrated into YOLOv10 to enhance the capture of shallow features in psyllid images. In addition, the soft nonmaximum suppression method is applied to resolve overlapping issues in counting the psyllids. Finally, the results are uploaded to an app, allowing users to stay informed about citrus pest conditions in real time. The experimental results indicate that DiTs-generated images achieved scores of 76.79, 0.29, and 1.68 in the Frechet inception distance, learned perceptual image patch similarity, and multiscale structural similarity metrics, respectively, outperforming the commonly used DDPM model by 8.51, 0.18, and 0.34, respectively. The improved YOLOv10 model, trained with the expanded DiTs dataset, reached a recall, F1-score, and precision of 90.55%, 92.18%, and 93.88%, respectively, demonstrating outstanding performance across all metrics. This approach enables fully automated recognition of citrus psyllids, facilitating real-time detection and contributing to the protection of citrus crops.","PeriodicalId":100637,"journal":{"name":"IEEE Transactions on AgriFood Electronics","volume":"3 1","pages":"286-294"},"PeriodicalIF":0.0,"publicationDate":"2025-04-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143817988","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Accurate analysis of livestock feed quality is critical for enhancing productivity and supporting sustainable farming practices. This study explores the integration of red, green, and blue (RGB) and near-infrared (NIR) imaging modalities, leveraging their complementary strengths, RGB for physical properties and NIR for chemical compositions, to predict key nutritional metrics. A novel knowledge distillation model was developed to transfer insights from a complex teacher model to a simpler student model. The approach involved training three types of models: single-channel (RGB or NIR), double-channel (RGB and NIR), and knowledge distillation models. Key evaluation metrics, including mean squared error (MSE), mean absolute error, and root MSE, validated the model's predictive accuracy. Experimental results demonstrated that the knowledge distillation model significantly outperformed both single- and double-channel models, achieving a 91.86% reduction in the MSE compared to RGB single-channel models, an 89.68% reduction compared to NIR single-channel models, and a 63.43% improvement over double-channel models. This study provides a robust, efficient, and cost-effective solution for feed quality assessment, highlighting the transformative potential of multimodal imaging and machine learning in precision agriculture.
{"title":"Advanced Multimodal Prediction of Components of Livestock Feed Materials Using Knowledge Distillation","authors":"Owoeye Babatunde Oluwabukunmi;Akomolafe Ayobami Joseph;Miraculous Udurume;Judith Nkechinyere Njoku;Cosmas Ifeanyi Nwakanma;Senorpe Asem-Hiablie;Rammohan Mallipeddi;Tusan Park;Daniel Dooyum Uyeh","doi":"10.1109/TAFE.2025.3548949","DOIUrl":"https://doi.org/10.1109/TAFE.2025.3548949","url":null,"abstract":"Accurate analysis of livestock feed quality is critical for enhancing productivity and supporting sustainable farming practices. This study explores the integration of red, green, and blue (RGB) and near-infrared (NIR) imaging modalities, leveraging their complementary strengths, RGB for physical properties and NIR for chemical compositions, to predict key nutritional metrics. A novel knowledge distillation model was developed to transfer insights from a complex teacher model to a simpler student model. The approach involved training three types of models: single-channel (RGB or NIR), double-channel (RGB and NIR), and knowledge distillation models. Key evaluation metrics, including mean squared error (MSE), mean absolute error, and root MSE, validated the model's predictive accuracy. Experimental results demonstrated that the knowledge distillation model significantly outperformed both single- and double-channel models, achieving a 91.86% reduction in the MSE compared to RGB single-channel models, an 89.68% reduction compared to NIR single-channel models, and a 63.43% improvement over double-channel models. This study provides a robust, efficient, and cost-effective solution for feed quality assessment, highlighting the transformative potential of multimodal imaging and machine learning in precision agriculture.","PeriodicalId":100637,"journal":{"name":"IEEE Transactions on AgriFood Electronics","volume":"3 1","pages":"272-285"},"PeriodicalIF":0.0,"publicationDate":"2025-03-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143817914","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-03-19DOI: 10.1109/TAFE.2025.3546731
You Wang;Huayan Hu;Shangru Wu;Ya Xiong
Laser weeding technology offers an effective alternative to traditional chemical and mechanical methods, providing precision, low cost, and environmental benefits. However, automatic targeting of weeds using lasers often encounters positional errors, particularly in dynamic weeding modes, which can significantly reduce weed removal efficiency. In addition, the operational efficiency of laser weeding is influenced by multiple factors, and the coupling effects of these factors require further investigation. This article examines the impact of laser power, incident angle, and spot size on the weeding efficiency of four common weed species under static conditions, considering the presence of positioning errors in laser targeting. To address these targeting errors, four weeding patterns were proposed: zigzag, triangular, horizontal, and vertical error compensation trajectories. Among these, the horizontal error compensation trajectory proved to be the most efficient, yielding stable and reliable results. In addition, a laser spot size adjustment device was designed to vary the spot diameter between 1–4 mm. Through four exploratory experiments and one validation experiment, the optimal combination of weeding parameters was identified: the horizontal weeding pattern, maximum laser power, an incidence angle of 80$^{circ }$, and a 2 mm spot diameter. This combination achieved optimal compensation with position errors under 2 mm. Validation experiments demonstrated that under these conditions, the average cutting times for chenopodium album, polygonum hydropiper, setaria viridis, and eleusine indica were 0.411 s, 0.308 s, 0.419 s, and 0.384 s, respectively, highlighting the efficiency and stability of this laser weeding model.
{"title":"Enhancing Weeding Efficiency: Addressing Targeting Positional Errors and Key Determinants of Cutting Efficiency in Laser Weeding Robots","authors":"You Wang;Huayan Hu;Shangru Wu;Ya Xiong","doi":"10.1109/TAFE.2025.3546731","DOIUrl":"https://doi.org/10.1109/TAFE.2025.3546731","url":null,"abstract":"Laser weeding technology offers an effective alternative to traditional chemical and mechanical methods, providing precision, low cost, and environmental benefits. However, automatic targeting of weeds using lasers often encounters positional errors, particularly in dynamic weeding modes, which can significantly reduce weed removal efficiency. In addition, the operational efficiency of laser weeding is influenced by multiple factors, and the coupling effects of these factors require further investigation. This article examines the impact of laser power, incident angle, and spot size on the weeding efficiency of four common weed species under static conditions, considering the presence of positioning errors in laser targeting. To address these targeting errors, four weeding patterns were proposed: zigzag, triangular, horizontal, and vertical error compensation trajectories. Among these, the horizontal error compensation trajectory proved to be the most efficient, yielding stable and reliable results. In addition, a laser spot size adjustment device was designed to vary the spot diameter between 1–4 mm. Through four exploratory experiments and one validation experiment, the optimal combination of weeding parameters was identified: the horizontal weeding pattern, maximum laser power, an incidence angle of 80<inline-formula><tex-math>$^{circ }$</tex-math></inline-formula>, and a 2 mm spot diameter. This combination achieved optimal compensation with position errors under 2 mm. Validation experiments demonstrated that under these conditions, the average cutting times for chenopodium album, polygonum hydropiper, setaria viridis, and eleusine indica were 0.411 s, 0.308 s, 0.419 s, and 0.384 s, respectively, highlighting the efficiency and stability of this laser weeding model.","PeriodicalId":100637,"journal":{"name":"IEEE Transactions on AgriFood Electronics","volume":"3 1","pages":"263-271"},"PeriodicalIF":0.0,"publicationDate":"2025-03-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143817915","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-02-17DOI: 10.1109/TAFE.2025.3536579
Moteaal Asadi Shirzi;Mehrdad R. Kermani
This article introduces a novel mechatronic system for coupling the stems of seedlings and plants to wooden stakes or ropes, a crucial process for supporting them during growth, transportation, and fruiting in plant propagation facilities and greenhouses. The stem–stake coupling device utilizes interconnected mechanisms and an impedance control method to adjust motor torque and speed, shaping metallic wire into clips of various shapes and dimensions, effectively securing plant stems to stakes or ropes. In a robotic system, a claw-shaped arm mechanism, a stereo camera, and real-time vision techniques are integrated into the stem–stake coupling device to identify the optimal coupling point and automate the coupling task. This innovation addresses the labor-intensive task of manual coupling, offering a scalable solution for growers through handheld devices or fully automated robotic systems. In the context of increasing labor shortages and rising costs, the technology offers a sustainable and efficient alternative with significant potential to enhance operational efficiency in greenhouses and propagation facilities.
{"title":"A Novel Technology for Semiautomatic and Automatic Stem–Stake Coupling of Seedlings and Plants","authors":"Moteaal Asadi Shirzi;Mehrdad R. Kermani","doi":"10.1109/TAFE.2025.3536579","DOIUrl":"https://doi.org/10.1109/TAFE.2025.3536579","url":null,"abstract":"This article introduces a novel mechatronic system for coupling the stems of seedlings and plants to wooden stakes or ropes, a crucial process for supporting them during growth, transportation, and fruiting in plant propagation facilities and greenhouses. The stem–stake coupling device utilizes interconnected mechanisms and an impedance control method to adjust motor torque and speed, shaping metallic wire into clips of various shapes and dimensions, effectively securing plant stems to stakes or ropes. In a robotic system, a claw-shaped arm mechanism, a stereo camera, and real-time vision techniques are integrated into the stem–stake coupling device to identify the optimal coupling point and automate the coupling task. This innovation addresses the labor-intensive task of manual coupling, offering a scalable solution for growers through handheld devices or fully automated robotic systems. In the context of increasing labor shortages and rising costs, the technology offers a sustainable and efficient alternative with significant potential to enhance operational efficiency in greenhouses and propagation facilities.","PeriodicalId":100637,"journal":{"name":"IEEE Transactions on AgriFood Electronics","volume":"3 1","pages":"254-262"},"PeriodicalIF":0.0,"publicationDate":"2025-02-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143818003","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Agriculture crop classification is helpful for agricultural production. However, it is challenging to classify crops from the agriculture image suffering from these problems: 1) Crops are often masked in complex backgrounds; 2) There is high similarity between crop categories. To address these problems, an adaptive masking attention network (AMA-Net) is proposed for agriculture crop identification from natural images, where the adaptive masking (AM) module is developed to distinguish the crop from the complex background by selectively eliminating redundant information of feature maps, and the fair attention module is devised to identify similar crops between categories by modeling the fine-grained features. Experiments conducted on the benchmark show the effectiveness and superiority of the proposed AMA-Net, achieving the performance of 96.65%, 96.65%, 97.13%, and 96.72% on the accuracy, precision, recall, and F1-score, respectively, which is better than other state-of-the-art methods.
{"title":"AMA-Net: Adaptive Masking Attention Network for Agricultural Crop Classification From UAV Images","authors":"Xu Wang;Deyi Wang;Zhaoshui He;Zhijie Lin;Shengli Xie","doi":"10.1109/TAFE.2025.3529724","DOIUrl":"https://doi.org/10.1109/TAFE.2025.3529724","url":null,"abstract":"Agriculture crop classification is helpful for agricultural production. However, it is challenging to classify crops from the agriculture image suffering from these problems: 1) Crops are often masked in complex backgrounds; 2) There is high similarity between crop categories. To address these problems, an adaptive masking attention network (AMA-Net) is proposed for agriculture crop identification from natural images, where the adaptive masking (AM) module is developed to distinguish the crop from the complex background by selectively eliminating redundant information of feature maps, and the fair attention module is devised to identify similar crops between categories by modeling the fine-grained features. Experiments conducted on the benchmark show the effectiveness and superiority of the proposed AMA-Net, achieving the performance of 96.65%, 96.65%, 97.13%, and 96.72% on the accuracy, precision, recall, and F1-score, respectively, which is better than other state-of-the-art methods.","PeriodicalId":100637,"journal":{"name":"IEEE Transactions on AgriFood Electronics","volume":"3 1","pages":"246-253"},"PeriodicalIF":0.0,"publicationDate":"2025-02-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143817955","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Continuous harvesting and storage of multiple fruits in a single operation allow robots to significantly reduce the travel distance required for repetitive back-and-forth movements. Traditional collision-free path planning algorithms, such as rapidly-exploring random tree (RRT) and A-star (A*), often fail to meet the demands of efficient continuous fruit harvesting due to their low search efficiency and the generation of excessive redundant points. This article presents the interactive local minima search algorithm (ILMSA), a fast path-planning method designed for the continuous harvesting of table-top grown strawberries. The algorithm featured an interactive node expansion strategy that iteratively extended and refined collision-free path segments based on local minima points. To enable the algorithm to function in 3-D, the 3-D environment was projected onto multiple 2-D planes, generating optimal paths on each plane. The best path was then selected, followed by integrating and smoothing the 3-D path segments. Simulations demonstrated that ILMSA outperformed existing methods, reducing path length by 21.5% and planning time by 97.1% compared to 3-D rapidly-exploring random tree, while achieving 11.6% shorter paths and 25.4% fewer nodes than the lowest point of the strawberry (LPS) algorithm in 3-D environments. In 2-D, ILMSA achieved path lengths 16.2% shorter than A*, 23.4% shorter than RRT, and 20.9% shorter than RRT-Connect, while being over 96% faster and generating significantly fewer nodes. In addition, ILMSA outperformed the partially guided Q-learning method, reducing path length by 36.7%, shortening planning time by 97.8%, and effectively avoiding entrapment in complex scenarios. Field tests confirmed ILMSA's suitability for complex agricultural tasks, having a combined planning and execution time and an average path length that were approximately 58% and 69%, respectively, of those achieved by the LPS algorithm.
{"title":"A Fast Path-Planning Method for Continuous Harvesting of Table-Top Grown Strawberries","authors":"Zhonghua Miao;Yang Chen;Lichao Yang;Shimin Hu;Ya Xiong","doi":"10.1109/TAFE.2025.3528403","DOIUrl":"https://doi.org/10.1109/TAFE.2025.3528403","url":null,"abstract":"Continuous harvesting and storage of multiple fruits in a single operation allow robots to significantly reduce the travel distance required for repetitive back-and-forth movements. Traditional collision-free path planning algorithms, such as rapidly-exploring random tree (RRT) and A-star (A*), often fail to meet the demands of efficient continuous fruit harvesting due to their low search efficiency and the generation of excessive redundant points. This article presents the interactive local minima search algorithm (ILMSA), a fast path-planning method designed for the continuous harvesting of table-top grown strawberries. The algorithm featured an interactive node expansion strategy that iteratively extended and refined collision-free path segments based on local minima points. To enable the algorithm to function in 3-D, the 3-D environment was projected onto multiple 2-D planes, generating optimal paths on each plane. The best path was then selected, followed by integrating and smoothing the 3-D path segments. Simulations demonstrated that ILMSA outperformed existing methods, reducing path length by 21.5% and planning time by 97.1% compared to 3-D rapidly-exploring random tree, while achieving 11.6% shorter paths and 25.4% fewer nodes than the lowest point of the strawberry (LPS) algorithm in 3-D environments. In 2-D, ILMSA achieved path lengths 16.2% shorter than A*, 23.4% shorter than RRT, and 20.9% shorter than RRT-Connect, while being over 96% faster and generating significantly fewer nodes. In addition, ILMSA outperformed the partially guided Q-learning method, reducing path length by 36.7%, shortening planning time by 97.8%, and effectively avoiding entrapment in complex scenarios. Field tests confirmed ILMSA's suitability for complex agricultural tasks, having a combined planning and execution time and an average path length that were approximately 58% and 69%, respectively, of those achieved by the LPS algorithm.","PeriodicalId":100637,"journal":{"name":"IEEE Transactions on AgriFood Electronics","volume":"3 1","pages":"233-245"},"PeriodicalIF":0.0,"publicationDate":"2025-01-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143821530","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-01-27DOI: 10.1109/TAFE.2025.3526963
Zijian Wang;Xiaofei An;Ling Wang;Jinshan Tang
To address the challenges of traditional manual maize seedling thinning, this article proposes an innovative approach that utilizes computer vision and deep learning for automated thinning. A zero-shot keypoint annotation algorithm, leveraging segment for anything model, is designed to label large datasets of maize seedling centers without requiring training samples. We also propose an improved hourglass network that significantly enhances seedling center positioning accuracy, enabling precise thinning decisions. Furthermore, a novel automatic thinning decision algorithm is devised to determine optimal removal strategies, ensuring ideal plant-to-plant spacing. The system's performance was evaluated against manually annotated data from 1020 images encompassing 2756 individual maize seedlings collected from farms. Impressively, the algorithm achieved a precision rate of 98.84%, confirming its ability to identify seedlings for removal while preserving healthy plants accurately. Evaluations of the keypoint detection network at a threshold of 0.2 yielded a percentage of correct keypoints of 97.66% and an object keypoint similarity of 0.87, surpassing existing methods and demonstrating the model's superior performance.
{"title":"Optimizing Maize Cultivation: A Vision-Based AI-Driven Methodology for Automated Seedling Thinning","authors":"Zijian Wang;Xiaofei An;Ling Wang;Jinshan Tang","doi":"10.1109/TAFE.2025.3526963","DOIUrl":"https://doi.org/10.1109/TAFE.2025.3526963","url":null,"abstract":"To address the challenges of traditional manual maize seedling thinning, this article proposes an innovative approach that utilizes computer vision and deep learning for automated thinning. A zero-shot keypoint annotation algorithm, leveraging segment for anything model, is designed to label large datasets of maize seedling centers without requiring training samples. We also propose an improved hourglass network that significantly enhances seedling center positioning accuracy, enabling precise thinning decisions. Furthermore, a novel automatic thinning decision algorithm is devised to determine optimal removal strategies, ensuring ideal plant-to-plant spacing. The system's performance was evaluated against manually annotated data from 1020 images encompassing 2756 individual maize seedlings collected from farms. Impressively, the algorithm achieved a precision rate of 98.84%, confirming its ability to identify seedlings for removal while preserving healthy plants accurately. Evaluations of the keypoint detection network at a threshold of 0.2 yielded a percentage of correct keypoints of 97.66% and an object keypoint similarity of 0.87, surpassing existing methods and demonstrating the model's superior performance.","PeriodicalId":100637,"journal":{"name":"IEEE Transactions on AgriFood Electronics","volume":"3 1","pages":"224-232"},"PeriodicalIF":0.0,"publicationDate":"2025-01-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143821529","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：