Artificial Intelligence in Agriculture最新文献

Adaptive compensation method for navigation positioning errors considering the vibration characteristics of a combine harvester1 考虑联合收割机振动特性的导航定位误差自适应补偿方法

IF 12.4 Q1 AGRICULTURE, MULTIDISCIPLINARY

Artificial Intelligence in Agriculture

Pub Date : 2026-01-03 DOI: 10.1016/j.aiia.2025.12.002

Wenbo Wei , Hui Wang , Maohua Xiao , Yejun Zhu , Dongfang Li , Xiaomei Xu , Peiqi Guo , Chenshuo Xie , He Zheng

Positioning accuracy directly affects the operational performance and stability of navigation systems. However, in complex field environments, severe vibrations during combine harvester operation can significantly exacerbate positioning errors. These vibrations have become a key limiting factor for navigation accuracy improvements. This study proposes an adaptive compensation method for navigation positioning errors that considers the vibration characteristics of combine harvesters. The proposed method aims to enhance anti-interference positioning accuracy during fully autonomous harvesting operations. First, a hybrid Whale Optimization Algorithm (WOA)–Light Gradient Boosting Machine (LightGBM) model was developed to identify operational stages. This model was trained using vibration data collected at 100 Hz over 15 s, yielding 1500 time-frame samples. Then, a regression prediction model for positioning errors was established using an eXtreme Gradient Boosting (XGBoost)–Multilayer Perceptron (MLP) framework. This model was built from GNSS/INS error data recorded at 5 Hz over 175 s, resulting in 875 time-frame samples. Finally, GNSS data compensated with the predicted positioning errors were fused with Inertial Navigation System (INS) data using an error-state Kalman filter (ESKF) within the same local planar coordinate system to achieve adaptive error compensation. Experimental results showed high identification accuracy and strong error prediction performance, with a maximum lateral deviation of 0.039 m in the straight-line path tracking test. During autonomous harvesting, the system maintained strong stability with an average lateral deviation of 0.087 m. This study provides a new approach for achieving high-precision positioning of combine harvesters under vibration disturbances. It also offers valuable insights for the development of anti-interference positioning technologies in other ground-based agricultural vehicles.

定位精度直接影响导航系统的工作性能和稳定性。然而，在复杂的野外环境中，联合收割机运行过程中的剧烈振动会显著加剧定位误差。这些振动已经成为限制导航精度提高的关键因素。提出了一种考虑联合收割机振动特性的导航定位误差自适应补偿方法。该方法旨在提高全自主采集过程中的抗干扰定位精度。首先，建立了一种混合鲸鱼优化算法(WOA) -光梯度增强机（LightGBM）模型来识别操作阶段。该模型使用在100赫兹下超过15秒收集的振动数据进行训练，产生1500个时间框架样本。然后，利用极端梯度增强(XGBoost) -多层感知器（MLP）框架建立了定位误差的回归预测模型。该模型是根据GNSS/INS误差数据建立的，记录频率为5 Hz，时间为175 s，得到875个时间框架样本。最后，利用误差状态卡尔曼滤波（ESKF）在同一局部平面坐标系内将经预测定位误差补偿的GNSS数据与惯性导航系统（INS）数据融合，实现自适应误差补偿。实验结果表明，该方法具有较高的识别精度和较强的误差预测能力，在直线路径跟踪试验中，最大横向偏差为0.039 m。在自主采收过程中，系统保持了较强的稳定性，平均横向偏差为0.087 m。该研究为联合收割机在振动干扰下实现高精度定位提供了新的途径。这也为其他陆基农用车辆抗干扰定位技术的发展提供了宝贵的见解。

{"title":"Adaptive compensation method for navigation positioning errors considering the vibration characteristics of a combine harvester1","authors":"Wenbo Wei , Hui Wang , Maohua Xiao , Yejun Zhu , Dongfang Li , Xiaomei Xu , Peiqi Guo , Chenshuo Xie , He Zheng","doi":"10.1016/j.aiia.2025.12.002","DOIUrl":"10.1016/j.aiia.2025.12.002","url":null,"abstract":"<div><div>Positioning accuracy directly affects the operational performance and stability of navigation systems. However, in complex field environments, severe vibrations during combine harvester operation can significantly exacerbate positioning errors. These vibrations have become a key limiting factor for navigation accuracy improvements. This study proposes an adaptive compensation method for navigation positioning errors that considers the vibration characteristics of combine harvesters. The proposed method aims to enhance anti-interference positioning accuracy during fully autonomous harvesting operations. First, a hybrid Whale Optimization Algorithm (WOA)–Light Gradient Boosting Machine (LightGBM) model was developed to identify operational stages. This model was trained using vibration data collected at 100 Hz over 15 s, yielding 1500 time-frame samples. Then, a regression prediction model for positioning errors was established using an eXtreme Gradient Boosting (XGBoost)–Multilayer Perceptron (MLP) framework. This model was built from GNSS/INS error data recorded at 5 Hz over 175 s, resulting in 875 time-frame samples. Finally, GNSS data compensated with the predicted positioning errors were fused with Inertial Navigation System (INS) data using an error-state Kalman filter (ESKF) within the same local planar coordinate system to achieve adaptive error compensation. Experimental results showed high identification accuracy and strong error prediction performance, with a maximum lateral deviation of 0.039 m in the straight-line path tracking test. During autonomous harvesting, the system maintained strong stability with an average lateral deviation of 0.087 m. This study provides a new approach for achieving high-precision positioning of combine harvesters under vibration disturbances. It also offers valuable insights for the development of anti-interference positioning technologies in other ground-based agricultural vehicles.</div></div>","PeriodicalId":52814,"journal":{"name":"Artificial Intelligence in Agriculture","volume":"16 1","pages":"Pages 695-710"},"PeriodicalIF":12.4,"publicationDate":"2026-01-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145975954","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}

引用次数: 0

Maize phenological stage recognition via coordinated UAV and UGV multi-view sensing and deep learning 基于无人机和UGV协同多视角感知和深度学习的玉米物候期识别

IF 12.4 Q1 AGRICULTURE, MULTIDISCIPLINARY

Artificial Intelligence in Agriculture

Pub Date : 2025-12-30 DOI: 10.1016/j.aiia.2025.12.004

Jibo Yue , Haikuan Feng , Yiguang Fan , Yang Liu , Chunjiang Zhao , Guijun Yang

Crop phenological stages, marked by key events such as germination, leaf emergence, flowering, and senescence, are critical indicators of crop development. Accurate, dynamic monitoring of these stages is essential for crop breeding management. This study introduces a novel multi-view sensing strategy based on coordinated unmanned aerial vehicles (UAVs) and unmanned ground vehicles (UGVs), designed to capture diverse canopy perspectives for phenological stage recognition in maize. Our approach integrates multiple data streams from top-down and internal-horizontal views, acquired via UAV and UGV platforms, and consists of three main components: (i) Acquisition of maize canopy height data, top-of-canopy (TOC) digital images, canopy multispectral images, and inside-of-canopy (IOC) digital images using a UAV- and UGV-based multi-view system; (ii) Development of a multi-modal deep learning framework, MSRNet (maize-phenological stages recognition network), which fuses physiological features from the UAV and UGV sensor modalities, including canopy height, vegetation indices, TOC maize leaf images, and IOC maize cob images; (iii) Comparative evaluation of MSRNet against conventional machine learning and deep learning models. Across 12 phenological stages (V2–R6), MSRNet achieved 84.5 % overall accuracy, outperforming conventional machine learning and single-modality deep learning benchmarks by 3.8–13.6 %. Grad-CAM visualizations confirmed dynamic, stage-specific attention, with the network automatically shifting focus from TOC leaves during vegetative growth to IOC reproductive organs during grain filling. This integrated UAV and UGV strategy, coupled with the dynamic feature selection capability of MSRNet, provides a comprehensive, interpretable workflow for high-throughput maize phenotyping and precision breeding.

作物物候阶段是作物发育的关键指标，以发芽、出芽、开花和衰老等关键事件为标志。准确、动态地监测这些阶段对作物育种管理至关重要。本研究提出了一种基于无人机和地面无人机的多视角玉米物候期识别方法。该方法集成了通过无人机和UGV平台获取的自上而下和内部水平视图的多个数据流，并由三个主要部分组成：(i)使用基于无人机和UGV的多视图系统获取玉米冠层高度数据、冠层顶部（TOC）数字图像、冠层多光谱图像和冠层内部（IOC）数字图像；（ii）开发多模态深度学习框架MSRNet（玉米物候阶段识别网络），该框架融合了无人机和UGV传感器模式的生理特征，包括冠层高度、植被指数、TOC玉米叶片图像和IOC玉米穗轴图像；（iii） MSRNet与传统机器学习和深度学习模型的比较评估。在12个物候阶段（V2-R6）中，MSRNet的总体准确率达到84.5%，比传统机器学习和单模式深度学习基准高出3.8 - 13.6%。Grad-CAM可视化证实了动态的、特定阶段的注意力，网络自动将注意力从营养生长期间的TOC叶片转移到籽粒灌浆期间的IOC生殖器官。这种集成无人机和UGV策略，加上MSRNet的动态特征选择能力，为高通量玉米表型和精确育种提供了一个全面的、可解释的工作流程。

{"title":"Maize phenological stage recognition via coordinated UAV and UGV multi-view sensing and deep learning","authors":"Jibo Yue , Haikuan Feng , Yiguang Fan , Yang Liu , Chunjiang Zhao , Guijun Yang","doi":"10.1016/j.aiia.2025.12.004","DOIUrl":"10.1016/j.aiia.2025.12.004","url":null,"abstract":"<div><div>Crop phenological stages, marked by key events such as germination, leaf emergence, flowering, and senescence, are critical indicators of crop development. Accurate, dynamic monitoring of these stages is essential for crop breeding management. This study introduces a novel multi-view sensing strategy based on coordinated unmanned aerial vehicles (UAVs) and unmanned ground vehicles (UGVs), designed to capture diverse canopy perspectives for phenological stage recognition in maize. Our approach integrates multiple data streams from top-down and internal-horizontal views, acquired via UAV and UGV platforms, and consists of three main components: (i) Acquisition of maize canopy height data, top-of-canopy (TOC) digital images, canopy multispectral images, and inside-of-canopy (IOC) digital images using a UAV- and UGV-based multi-view system; (ii) Development of a multi-modal deep learning framework, MSRNet (maize-phenological stages recognition network), which fuses physiological features from the UAV and UGV sensor modalities, including canopy height, vegetation indices, TOC maize leaf images, and IOC maize cob images; (iii) Comparative evaluation of MSRNet against conventional machine learning and deep learning models. Across 12 phenological stages (V2–R6), MSRNet achieved 84.5 % overall accuracy, outperforming conventional machine learning and single-modality deep learning benchmarks by 3.8–13.6 %. Grad-CAM visualizations confirmed dynamic, stage-specific attention, with the network automatically shifting focus from TOC leaves during vegetative growth to IOC reproductive organs during grain filling. This integrated UAV and UGV strategy, coupled with the dynamic feature selection capability of MSRNet, provides a comprehensive, interpretable workflow for high-throughput maize phenotyping and precision breeding.</div></div>","PeriodicalId":52814,"journal":{"name":"Artificial Intelligence in Agriculture","volume":"16 1","pages":"Pages 643-657"},"PeriodicalIF":12.4,"publicationDate":"2025-12-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145924243","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}

引用次数: 0

SpecColorNet: An interpretable multimodal deep learning approach for predicting SSC of multiple pears SpecColorNet：一种可解释的多模态深度学习方法，用于预测多个梨的SSC

IF 12.4 Q1 AGRICULTURE, MULTIDISCIPLINARY

Artificial Intelligence in Agriculture

Pub Date : 2025-12-30 DOI: 10.1016/j.aiia.2025.12.003

Xin Xu , Mengfei Yang , Jie Yang , Hao Yin , Yanyu Chen , Weijie Lan , Guoxiang Sun , Xiaochan Wang , Shaoling Zhang , Xiaolei Zhang

Visible/near-infrared spectroscopy provides a non-destructive approach for evaluating soluble solids content (SSC) of pears (Pyrus pyrifolia Nakai). However, variations among pear cultivars, especially in pear color, markedly affect spectral reflectance, thereby limiting the cross-cultivar SSC prediction. To address this, we developed SpecColorNet, a multimodal deep learning framework that integrates spectral and peel color data to accurately predict the SSC of six different pear cultivars simultaneously, and interpreted its decision-making mechanism with gradient-weighted class activation mapping++ (Grad-CAM++). The prediction accuracy of the multi-cultivar SpecColorNet was improved by 9.42 % compared to the multi-cultivar model based solely on spectral data with corresponding RMSEP values of 0.63, 0.60, 0.92, 0.70, 0.54 and 0.55°Brix for six different pear cultivars. In addition, the prediction accuracy of SpecColorNet was comparable to that of single-cultivar spectral models. The interpretation analysis indicated that the SpecColorNet successfully directed the model's attention to the 555–640 nm spectral region, where the most marked cross-cultivar differences occur, thereby improving its generalization and accuracy. Overall, this study proposed a multimodal approach for robust SSC prediction across diverse pear cultivars, which overcame the challenge of cultivar diversity and peel color variability to enable multi-cultivar models with enhanced robustness over pure spectroscopy-based methods.

可见/近红外光谱技术为梨（Pyrus pyrifolia Nakai）可溶性固形物含量测定提供了一种无损的方法。然而，梨品种之间的差异，特别是梨颜色的差异，会显著影响光谱反射率，从而限制了跨品种SSC的预测。为了解决这个问题，我们开发了一个多模态深度学习框架SpecColorNet，该框架集成了光谱和果皮颜色数据，同时准确预测了6个不同梨品种的SSC，并使用梯度加权类激活映射++ (grad-cam++)解释了其决策机制。多品种SpecColorNet的预测精度比单纯基于光谱数据的多品种模型提高了9.42%，对应的RMSEP值分别为0.63、0.60、0.92、0.70、0.54和0.55°Brix。此外，SpecColorNet的预测精度与单品种光谱模型相当。解释分析表明，SpecColorNet成功地将模型的注意力集中在555-640 nm光谱区域，该区域是品种间差异最显著的区域，从而提高了模型的泛化和准确性。总体而言，本研究提出了一种多模态的梨品种SSC鲁棒预测方法，克服了品种多样性和果皮颜色变异的挑战，使多品种模型比纯光谱方法具有更强的鲁棒性。

{"title":"SpecColorNet: An interpretable multimodal deep learning approach for predicting SSC of multiple pears","authors":"Xin Xu , Mengfei Yang , Jie Yang , Hao Yin , Yanyu Chen , Weijie Lan , Guoxiang Sun , Xiaochan Wang , Shaoling Zhang , Xiaolei Zhang","doi":"10.1016/j.aiia.2025.12.003","DOIUrl":"10.1016/j.aiia.2025.12.003","url":null,"abstract":"<div><div>Visible/near-infrared spectroscopy provides a non-destructive approach for evaluating soluble solids content (SSC) of pears (<em>Pyrus pyrifolia</em> Nakai). However, variations among pear cultivars, especially in pear color, markedly affect spectral reflectance, thereby limiting the cross-cultivar SSC prediction. To address this, we developed SpecColorNet, a multimodal deep learning framework that integrates spectral and peel color data to accurately predict the SSC of six different pear cultivars simultaneously, and interpreted its decision-making mechanism with gradient-weighted class activation mapping++ (Grad-CAM++). The prediction accuracy of the multi-cultivar SpecColorNet was improved by 9.42 % compared to the multi-cultivar model based solely on spectral data with corresponding RMSEP values of 0.63, 0.60, 0.92, 0.70, 0.54 and 0.55°Brix for six different pear cultivars. In addition, the prediction accuracy of SpecColorNet was comparable to that of single-cultivar spectral models. The interpretation analysis indicated that the SpecColorNet successfully directed the model's attention to the 555–640 nm spectral region, where the most marked cross-cultivar differences occur, thereby improving its generalization and accuracy. Overall, this study proposed a multimodal approach for robust SSC prediction across diverse pear cultivars, which overcame the challenge of cultivar diversity and peel color variability to enable multi-cultivar models with enhanced robustness over pure spectroscopy-based methods.</div></div>","PeriodicalId":52814,"journal":{"name":"Artificial Intelligence in Agriculture","volume":"16 1","pages":"Pages 619-629"},"PeriodicalIF":12.4,"publicationDate":"2025-12-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145924241","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}

引用次数: 0

CTGNN: UAV-satellite cross-domain transfer learning for monitoring oat growth in China’s key production areas CTGNN：用于监测中国重点产区燕麦生长的无人机-卫星跨域迁移学习

IF 12.4 Q1 AGRICULTURE, MULTIDISCIPLINARY

Artificial Intelligence in Agriculture

Pub Date : 2025-12-29 DOI: 10.1016/j.aiia.2025.12.006

Pengpeng Zhang , Bing Lu , Jiali Shang , Changwei Tan , Shuchang Sun , Zhuo Xu , Junyong Ge , Yadong Yang , Huadong Zang , Zhaohai Zeng

Modern agricultural production necessitates real-time, precise monitoring of crop growth status to optimize management decisions. While remote sensing technologies offer multi-scale observational capabilities, conventional crop monitoring models face two critical limitations: (1) the independent retrieval of individual physiological traits, which overlooks the dynamic coupling between structural and physiological traits, and (2) inadequate cross-platform model transferability (e.g., from UAV images to satellite images), hindering the scaling of field-level precision to regional applications. To address these challenges, we proposed a deep learning-based framework, Cross-Task Growth Neural Network (CTGNN). This framework employed a dual-stream architecture to process spectral features for Leaf Area Index (LAI) and Soil Plant Analysis Development (SPAD), while using cross-trait attention mechanisms to capture their interactions. We further assessed the knowledge transfer capabilities of the model by comparing two transfer learning strategies—Transfer Component Analysis (TCA) and Domain-Adversarial Neural Networks (DANN)—in facilitating the adaptation of UAV-derived (1.3 cm/pixel) data to satellite-scale (3 m/pixel) monitoring. Validation using UAV-satellite synergetic datasets from extensively field-tested oat cultivars in China's Bashang Plateau demonstrates that CTGNN significantly reduces the prediction errors for LAI and SPAD compared with independent trait models, with RMSE reductions of 6.4–14.4 % and 10.5–15.6 %, respectively. In a cross-domain transfer learning scenario, the CTGNN model with the DANN strategy requires only 5 % of satellite-labeled data for fine-tuning to achieve regional-scale monitoring (LAI: R2 = 0.769; SPAD: R2 = 0.714). This framework provides a novel approach for the collaborative inversion of multiple crop growth traits, while its UAV-satellite cross-scale transfer capability facilitates optimal decision-making in oat variety breeding and cultivation technique dissemination, particularly in arid and semi-arid regions.

现代农业生产需要实时、精确地监测作物生长状况，以优化管理决策。虽然遥感技术提供了多尺度观测能力，但传统的作物监测模型面临两个关键限制：(1)单个生理性状的独立检索，忽略了结构与生理性状之间的动态耦合；(2)模型跨平台可移植性不足（例如，从无人机图像到卫星图像），阻碍了田间精度向区域应用的扩展。为了解决这些挑战，我们提出了一个基于深度学习的框架，交叉任务生长神经网络（CTGNN）。该框架采用双流架构处理叶面积指数（LAI）和土壤植物分析发展（SPAD）的光谱特征，同时利用交叉性状注意机制捕捉它们的相互作用。通过比较两种迁移学习策略——迁移成分分析（TCA）和域对抗神经网络（DANN），我们进一步评估了模型的知识迁移能力，以促进无人机衍生（1.3厘米/像素）数据对卫星尺度（3米/像素）监测的适应。利用中国巴上高原广泛田间试验的燕麦品种的无人机-卫星协同数据集进行验证，与独立性状模型相比，CTGNN显著降低了LAI和SPAD的预测误差，RMSE分别降低了6.4 - 14.4%和10.5 - 15.6%。在跨域迁移学习场景下，采用DANN策略的CTGNN模型只需要5%的卫星标记数据进行微调即可实现区域尺度的监测（LAI: R2 = 0.769; SPAD: R2 = 0.714）。该框架为多种作物生长性状的协同反演提供了一种新的方法，而其无人机-卫星跨尺度转移能力为燕麦品种育种和栽培技术推广提供了最优决策，特别是在干旱和半干旱地区。

{"title":"CTGNN: UAV-satellite cross-domain transfer learning for monitoring oat growth in China’s key production areas","authors":"Pengpeng Zhang , Bing Lu , Jiali Shang , Changwei Tan , Shuchang Sun , Zhuo Xu , Junyong Ge , Yadong Yang , Huadong Zang , Zhaohai Zeng","doi":"10.1016/j.aiia.2025.12.006","DOIUrl":"10.1016/j.aiia.2025.12.006","url":null,"abstract":"<div><div>Modern agricultural production necessitates real-time, precise monitoring of crop growth status to optimize management decisions. While remote sensing technologies offer multi-scale observational capabilities, conventional crop monitoring models face two critical limitations: (1) the independent retrieval of individual physiological traits, which overlooks the dynamic coupling between structural and physiological traits, and (2) inadequate cross-platform model transferability (e.g., from UAV images to satellite images), hindering the scaling of field-level precision to regional applications. To address these challenges, we proposed a deep learning-based framework, Cross-Task Growth Neural Network (CTGNN). This framework employed a dual-stream architecture to process spectral features for Leaf Area Index (LAI) and Soil Plant Analysis Development (SPAD), while using cross-trait attention mechanisms to capture their interactions. We further assessed the knowledge transfer capabilities of the model by comparing two transfer learning strategies—Transfer Component Analysis (TCA) and Domain-Adversarial Neural Networks (DANN)—in facilitating the adaptation of UAV-derived (1.3 cm/pixel) data to satellite-scale (3 m/pixel) monitoring. Validation using UAV-satellite synergetic datasets from extensively field-tested oat cultivars in China's Bashang Plateau demonstrates that CTGNN significantly reduces the prediction errors for LAI and SPAD compared with independent trait models, with RMSE reductions of 6.4–14.4 % and 10.5–15.6 %, respectively. In a cross-domain transfer learning scenario, the CTGNN model with the DANN strategy requires only 5 % of satellite-labeled data for fine-tuning to achieve regional-scale monitoring (LAI: R2 = 0.769; SPAD: R2 = 0.714). This framework provides a novel approach for the collaborative inversion of multiple crop growth traits, while its UAV-satellite cross-scale transfer capability facilitates optimal decision-making in oat variety breeding and cultivation technique dissemination, particularly in arid and semi-arid regions.</div></div>","PeriodicalId":52814,"journal":{"name":"Artificial Intelligence in Agriculture","volume":"16 1","pages":"Pages 630-642"},"PeriodicalIF":12.4,"publicationDate":"2025-12-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145924150","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}

引用次数: 0

LPNet: A lightweight progressive network for calyx-aware apple pose estimation in orchard environments LPNet：用于果园环境中花萼感知苹果姿态估计的轻量级渐进式网络

IF 12.4 Q1 AGRICULTURE, MULTIDISCIPLINARY

Artificial Intelligence in Agriculture

Pub Date : 2025-12-29 DOI: 10.1016/j.aiia.2025.12.005

Wenbei Wang, Licheng Zhu, Bo Zhao, Jianbo Gao, Xuezhi Cui, Zhenhao Han, Siyuan Zhao

Robotic apple harvesting has gradually become a critical requirement for modern agriculture. However, complex orchard environments and limited computational resources pose significant challenges for achieving accurate and efficient vision-based pose estimation of apples. To address these challenges, this study proposes LPNet, a lightweight network tailored for calyx-aware apple pose estimation. LPNet introduces a Calyx-Aware Labeling Strategy (CALS) to improve annotation quality, employs an improved ShuffleNetV2 backbone with progressive channel expansion, and integrates Content-Aware ReAssembly of FEatures (CARAFE) with Bi-directional Feature Pyramid Network (BiFPN) in the neck, enabling compact yet expressive multi-scale feature processing. Furthermore, an Axis-Aligned Soft Geometric Constraint (ASGC) is incorporated to reinforce spatial symmetry and training stability through geometry-aware learning. The head predicts five calyx-centric keypoints, which are then processed through a geometric solver to determine the apple's 2D orientation for harvesting guidance. Experimental results show that LPNet achieves 93.6 % [email protected] at a low computational cost of only 22.2 GFLOPs while maintaining a high inference rate of 158.7 FPS, outperforming representative models such as YOLOv12m-pose, HRFormer, and RTMPose. These results demonstrate that LPNet achieves an effective trade-off between accuracy and efficiency, laying a solid foundation for the future development of practical vision systems in autonomous apple harvesting robots.

苹果收获机器人已逐渐成为现代农业的关键要求。然而，复杂的果园环境和有限的计算资源对实现准确有效的基于视觉的苹果姿态估计提出了重大挑战。为了应对这些挑战，本研究提出了LPNet，这是一种专为萼感知苹果姿态估计量身定制的轻量级网络。LPNet引入了萼感知标注策略（calx - aware Labeling Strategy， CALS）来提高标注质量，采用改进的ShuffleNetV2骨干网进行渐进式通道扩展，并在颈部集成了内容感知特征重组（CARAFE）和双向特征金字塔网络（BiFPN），实现了紧凑而又具有表达能力的多尺度特征处理。此外，该方法还引入了一种轴向软几何约束（ASGC），通过几何感知学习增强空间对称性和训练稳定性。头部预测五个以花萼为中心的关键点，然后通过几何求解器处理这些关键点，以确定苹果的二维方向，以指导收获。实验结果表明，LPNet以22.2 GFLOPs的低计算成本达到了93.6% [email protected]，同时保持了158.7 FPS的高推理率，优于YOLOv12m-pose、HRFormer和RTMPose等代表性模型。这些结果表明，LPNet实现了精度和效率之间的有效权衡，为未来自主苹果收获机器人实用视觉系统的发展奠定了坚实的基础。

{"title":"LPNet: A lightweight progressive network for calyx-aware apple pose estimation in orchard environments","authors":"Wenbei Wang, Licheng Zhu, Bo Zhao, Jianbo Gao, Xuezhi Cui, Zhenhao Han, Siyuan Zhao","doi":"10.1016/j.aiia.2025.12.005","DOIUrl":"10.1016/j.aiia.2025.12.005","url":null,"abstract":"<div><div>Robotic apple harvesting has gradually become a critical requirement for modern agriculture. However, complex orchard environments and limited computational resources pose significant challenges for achieving accurate and efficient vision-based pose estimation of apples. To address these challenges, this study proposes LPNet, a lightweight network tailored for calyx-aware apple pose estimation. LPNet introduces a Calyx-Aware Labeling Strategy (CALS) to improve annotation quality, employs an improved ShuffleNetV2 backbone with progressive channel expansion, and integrates Content-Aware ReAssembly of FEatures (CARAFE) with Bi-directional Feature Pyramid Network (BiFPN) in the neck, enabling compact yet expressive multi-scale feature processing. Furthermore, an Axis-Aligned Soft Geometric Constraint (ASGC) is incorporated to reinforce spatial symmetry and training stability through geometry-aware learning. The head predicts five calyx-centric keypoints, which are then processed through a geometric solver to determine the apple's 2D orientation for harvesting guidance. Experimental results show that LPNet achieves 93.6 % [email protected] at a low computational cost of only 22.2 GFLOPs while maintaining a high inference rate of 158.7 FPS, outperforming representative models such as YOLOv12m-pose, HRFormer, and RTMPose. These results demonstrate that LPNet achieves an effective trade-off between accuracy and efficiency, laying a solid foundation for the future development of practical vision systems in autonomous apple harvesting robots.</div></div>","PeriodicalId":52814,"journal":{"name":"Artificial Intelligence in Agriculture","volume":"16 1","pages":"Pages 672-694"},"PeriodicalIF":12.4,"publicationDate":"2025-12-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145924242","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}

引用次数: 0

Integrating 3D detection networks and dynamic temporal phenotyping for wheat yield classification and prediction 结合三维检测网络和动态时间表型进行小麦产量分类和预测

IF 12.4 Q1 AGRICULTURE, MULTIDISCIPLINARY

Artificial Intelligence in Agriculture

Pub Date : 2025-12-06 DOI: 10.1016/j.aiia.2025.12.001

Honghao Zhou , Bingxi Qin , Qing Li , Wenlong Su , Shaowei Liang , Haijiang Min , Jingrong Zang , Shichao Jin , Dong Jiang , Jiawei Chen

Automated phenotyping of wheat growth stages from 3D point clouds is still limited. The study presents a concise framework that reconstructs multi-view UAS imagery into 3D point clouds (jointing to maturity) and performs plot-level phenotyping. A novel 3D wheat plot detection network—integrating spatial–channel coordinated attention and area attention modules—improves depth-direction feature recognition, and a point-cloud-density-based row segmentation algorithm enables planting-row-scale plot delineation. A supporting software system facilitates 3D visualization and automated extraction of phenotypic parameters. We introduce a dynamic phenotypic index of five temporal metrics (growth stage, slow growth stage, height/area reduction stage, maximum height/area difference stage, and height/area change rate) for growth-stage classification and yield prediction using static and time-series models. Experiments show strong agreement between predicted and measured plot heights (R² = 0.937); the detection net achieved AP_3D = 94.15 % and AP_BEV = 95.35 % in “easy” mode; and a Bi-LSTM incorporating dynamic traits reached 82.37 % prediction accuracy for leaf area and yield, a 6.14 % improvement over static-trait models. This workflow supports high-throughput 3D phenotyping and reliable yield estimation for precision agriculture.

小麦生长阶段的三维点云自动表型分析仍然有限。该研究提出了一个简洁的框架，将多视图UAS图像重建为3D点云（连接到成熟度），并执行情节水平表型。基于空间通道协调注意和区域注意模块的小麦三维地块检测网络改进了深度-方向特征识别，基于点云密度的行分割算法实现了种植行尺度的地块划分。一个支持的软件系统促进了表型参数的3D可视化和自动提取。采用静态和时间序列模型，引入生长期、慢生长期、高/面积减少期、最大高/面积差期和高/面积变化率五个时间指标的动态表型指数，用于生长期分类和产量预测。实验结果表明，预测值与实测值吻合较好（R2 = 0.937）；“easy”模式下检测网AP3D = 94.15%, APBEV = 95.35%；结合动态性状的Bi-LSTM对叶面积和产量的预测准确率达到82.37%，比静态性状模型提高6.14%。该工作流程支持高通量3D表型和可靠的精准农业产量估计。

{"title":"Integrating 3D detection networks and dynamic temporal phenotyping for wheat yield classification and prediction","authors":"Honghao Zhou , Bingxi Qin , Qing Li , Wenlong Su , Shaowei Liang , Haijiang Min , Jingrong Zang , Shichao Jin , Dong Jiang , Jiawei Chen","doi":"10.1016/j.aiia.2025.12.001","DOIUrl":"10.1016/j.aiia.2025.12.001","url":null,"abstract":"<div><div>Automated phenotyping of wheat growth stages from 3D point clouds is still limited. The study presents a concise framework that reconstructs multi-view UAS imagery into 3D point clouds (jointing to maturity) and performs plot-level phenotyping. A novel 3D wheat plot detection network—integrating spatial–channel coordinated attention and area attention modules—improves depth-direction feature recognition, and a point-cloud-density-based row segmentation algorithm enables planting-row-scale plot delineation. A supporting software system facilitates 3D visualization and automated extraction of phenotypic parameters. We introduce a dynamic phenotypic index of five temporal metrics (growth stage, slow growth stage, height/area reduction stage, maximum height/area difference stage, and height/area change rate) for growth-stage classification and yield prediction using static and time-series models. Experiments show strong agreement between predicted and measured plot heights (R<sup>2</sup> = 0.937); the detection net achieved AP<sub>3D</sub> = 94.15 % and AP<sub>BEV</sub> = 95.35 % in “easy” mode; and a Bi-LSTM incorporating dynamic traits reached 82.37 % prediction accuracy for leaf area and yield, a 6.14 % improvement over static-trait models. This workflow supports high-throughput 3D phenotyping and reliable yield estimation for precision agriculture.</div></div>","PeriodicalId":52814,"journal":{"name":"Artificial Intelligence in Agriculture","volume":"16 1","pages":"Pages 603-618"},"PeriodicalIF":12.4,"publicationDate":"2025-12-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145747432","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}

引用次数: 0

Optimizing herbicide reduction: A simulation approach using Artificial Intelligence and different nozzle configurations 优化除草剂减少：使用人工智能和不同喷嘴配置的模拟方法

IF 12.4 Q1 AGRICULTURE, MULTIDISCIPLINARY

Artificial Intelligence in Agriculture

Pub Date : 2025-12-01 DOI: 10.1016/j.aiia.2025.11.011

Renato H. Furlanetto , Ana C. Buzanini , Arnold W. Schumann , Nathan S. Boyd

Targeted application aims to minimize product usage by spraying only where needed. However, there is a lack of tools to evaluate the potential savings and the model's limitations. To address this, we developed a computer simulation using two YOLOv8x models (bounding boxes and segmentation masks) to assess spray volume reduction (SVR) and nozzle limitations across varying weed densities. We used Python, OpenCV, and Tkinter to analyze videos from row-middle trials at the University of Florida. The system tracks weeds crossing a horizontal detection line within a 70 cm row-middle width, dividing the frame into one to eight vertical zones representing nozzle distribution. When a weed is detected, the corresponding frame is saved. The system calculates the activation time in seconds by considering the total number of video frames. Spray volume calculations were based on manual measurements of a TeeJet 8001VS nozzle tip, which dispenses a known volume of liquid per second at 35 PSI. The activation time was multiplied by this rate to estimate the targeted spray volume. The broadcast application volume was calculated by multiplying the total video duration by the same tip output. The results showed that the models achieved up to 96 % accuracy (mAP) with no statistical difference. A polynomial model for low and medium weed densities demonstrated SVR of 74 % (six nozzles) and 57 % (seven nozzles). A linear model for high density achieved a 40 % reduction. The lowest reduction occurred with a single nozzle (4 % for medium, 2 % for high density). These findings demonstrate that nozzle density significantly impacted spray reduction at medium and high densities while low-density savings remained consistent.

定向应用的目的是通过只在需要的地方喷洒来减少产品的使用。然而，缺乏工具来评估潜在的节省和模型的局限性。为了解决这个问题，我们开发了一个使用两个YOLOv8x模型（边界盒和分割掩模）的计算机模拟，以评估不同杂草密度下的喷雾体积减少（SVR）和喷嘴限制。我们使用Python、OpenCV和Tkinter来分析佛罗里达大学（University of Florida）中排试验的视频。该系统在一个70厘米宽的中线范围内跟踪穿过水平检测线的杂草，将框架划分为代表喷嘴分布的一到八个垂直区域。当检测到杂草时，保存相应的帧。系统通过考虑视频帧总数来计算激活时间（以秒为单位）。喷雾体积的计算是基于TeeJet 8001VS喷嘴尖端的手动测量，该喷嘴在35psi下每秒分配已知体积的液体。激活时间乘以这个速率来估计目标喷雾体积。通过将总视频持续时间乘以相同的提示输出来计算广播应用音量。结果表明，模型的准确率高达96%，无统计学差异。低、中杂草密度的多项式模型显示SVR分别为74%（6个喷嘴）和57%（7个喷嘴）。高密度的线性模型实现了40%的减少。单喷嘴的降低率最低（中等密度为4%，高密度为2%）。这些研究结果表明，在中高密度和低密度情况下，喷嘴密度显著影响了喷雾减少，而低密度情况下，喷嘴密度保持不变。

{"title":"Optimizing herbicide reduction: A simulation approach using Artificial Intelligence and different nozzle configurations","authors":"Renato H. Furlanetto , Ana C. Buzanini , Arnold W. Schumann , Nathan S. Boyd","doi":"10.1016/j.aiia.2025.11.011","DOIUrl":"10.1016/j.aiia.2025.11.011","url":null,"abstract":"<div><div>Targeted application aims to minimize product usage by spraying only where needed. However, there is a lack of tools to evaluate the potential savings and the model's limitations. To address this, we developed a computer simulation using two YOLOv8x models (bounding boxes and segmentation masks) to assess spray volume reduction (SVR) and nozzle limitations across varying weed densities. We used Python, OpenCV, and Tkinter to analyze videos from row-middle trials at the University of Florida. The system tracks weeds crossing a horizontal detection line within a 70 cm row-middle width, dividing the frame into one to eight vertical zones representing nozzle distribution. When a weed is detected, the corresponding frame is saved. The system calculates the activation time in seconds by considering the total number of video frames. Spray volume calculations were based on manual measurements of a TeeJet 8001VS nozzle tip, which dispenses a known volume of liquid per second at 35 PSI. The activation time was multiplied by this rate to estimate the targeted spray volume. The broadcast application volume was calculated by multiplying the total video duration by the same tip output. The results showed that the models achieved up to 96 % accuracy (mAP) with no statistical difference. A polynomial model for low and medium weed densities demonstrated SVR of 74 % (six nozzles) and 57 % (seven nozzles). A linear model for high density achieved a 40 % reduction. The lowest reduction occurred with a single nozzle (4 % for medium, 2 % for high density). These findings demonstrate that nozzle density significantly impacted spray reduction at medium and high densities while low-density savings remained consistent.</div></div>","PeriodicalId":52814,"journal":{"name":"Artificial Intelligence in Agriculture","volume":"16 1","pages":"Pages 592-602"},"PeriodicalIF":12.4,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145747431","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}

引用次数: 0

SCLFormer: A synergistic convolution-linear attention transformer for hyperspectral image classification of mechanical damage in maize kernels SCLFormer：一种用于玉米籽粒机械损伤高光谱图像分类的协同卷积-线性注意转换器

IF 12.4 Q1 AGRICULTURE, MULTIDISCIPLINARY

Artificial Intelligence in Agriculture

Pub Date : 2025-11-27 DOI: 10.1016/j.aiia.2025.11.010

Yiqiang Zheng , Jun Fu , Fengshuang Liu , Haiming Zhao , Jindai Liu

Classifying mechanical damage in maize kernels using hyperspectral imaging is crucial for food security and loss reduction. Existing methods are constrained by high computational complexity and limited precision in detecting subtle damages, such as pericarp damage and kernel cracks. To address these challenges, we introduce a novel algorithm, the Synergistic Convolution and Linear Attention Transformer (SCLFormer). By replacing traditional softmax attention with linear attention, we reduce computational complexity from quadratic to linear, thereby enhancing efficiency. Integrating convolutional operations into the encoder enriches local prior information for global feature modeling, improving classification accuracy. SCLFormer achieves an overall accuracy of 97.08 % in classifying maize kernel damage, with over 85 % accuracy for cracked kernel and pericarp damage. Compared to softmax attention, SCLFormer reduces training and testing times by 355.27 s (16.86 %) and 0.71 s (23.67 %), respectively. Additionally, we propose a modular hyperspectral image-level classification framework that can integrate existing pixel-level feature extraction networks to achieve classification accuracies exceeding 80 %, demonstrating the framework's scalability. SCLFormer, serving as the framework's dedicated feature extraction component, provides a robust solution for maize kernel damage classification and exhibits substantial potential for broader spatial-scale applications. This framework establishes a novel technical paradigm for hyperspectral image-wise classification of other agricultural products.

利用高光谱成像技术对玉米籽粒机械损伤进行分类对粮食安全和减少损失至关重要。现有方法在检测果皮损伤和果仁裂纹等细微损伤时，计算量大、精度低。为了解决这些挑战，我们引入了一种新的算法，即协同卷积和线性注意力转换器（SCLFormer）。通过将传统的softmax注意力替换为线性注意力，将计算复杂度从二次型降低到线性型，从而提高了效率。将卷积运算集成到编码器中，丰富了全局特征建模的局部先验信息，提高了分类精度。SCLFormer对玉米籽粒损伤分类的总体准确率为97.08%，对裂粒和果皮损伤分类的准确率超过85%。与softmax相比，SCLFormer的训练和测试时间分别减少了355.27 s（16.86%）和0.71 s（23.67%）。此外，我们提出了一个模块化的高光谱图像级分类框架，该框架可以集成现有的像素级特征提取网络，实现超过80%的分类精度，证明了框架的可扩展性。SCLFormer作为该框架的专用特征提取组件，为玉米籽粒损伤分类提供了一个强大的解决方案，并显示出更广泛的空间尺度应用潜力。该框架为其他农产品的高光谱图像分类建立了一种新的技术范式。

{"title":"SCLFormer: A synergistic convolution-linear attention transformer for hyperspectral image classification of mechanical damage in maize kernels","authors":"Yiqiang Zheng , Jun Fu , Fengshuang Liu , Haiming Zhao , Jindai Liu","doi":"10.1016/j.aiia.2025.11.010","DOIUrl":"10.1016/j.aiia.2025.11.010","url":null,"abstract":"<div><div>Classifying mechanical damage in maize kernels using hyperspectral imaging is crucial for food security and loss reduction. Existing methods are constrained by high computational complexity and limited precision in detecting subtle damages, such as pericarp damage and kernel cracks. To address these challenges, we introduce a novel algorithm, the Synergistic Convolution and Linear Attention Transformer (SCLFormer). By replacing traditional softmax attention with linear attention, we reduce computational complexity from quadratic to linear, thereby enhancing efficiency. Integrating convolutional operations into the encoder enriches local prior information for global feature modeling, improving classification accuracy. SCLFormer achieves an overall accuracy of 97.08 % in classifying maize kernel damage, with over 85 % accuracy for cracked kernel and pericarp damage. Compared to softmax attention, SCLFormer reduces training and testing times by 355.27 s (16.86 %) and 0.71 s (23.67 %), respectively. Additionally, we propose a modular hyperspectral image-level classification framework that can integrate existing pixel-level feature extraction networks to achieve classification accuracies exceeding 80 %, demonstrating the framework's scalability. SCLFormer, serving as the framework's dedicated feature extraction component, provides a robust solution for maize kernel damage classification and exhibits substantial potential for broader spatial-scale applications. This framework establishes a novel technical paradigm for hyperspectral image-wise classification of other agricultural products.</div></div>","PeriodicalId":52814,"journal":{"name":"Artificial Intelligence in Agriculture","volume":"16 1","pages":"Pages 565-577"},"PeriodicalIF":12.4,"publicationDate":"2025-11-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145693322","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}

引用次数: 0

Multi-arm robotic system and strategy for the automatic packaging of apples 苹果自动包装的多臂机器人系统及策略

IF 12.4 Q1 AGRICULTURE, MULTIDISCIPLINARY

Artificial Intelligence in Agriculture

Pub Date : 2025-11-26 DOI: 10.1016/j.aiia.2025.11.006

Yizhi Zhang , Liping Chen , Xin Li , Qicheng Li , Jiangbo Li

Packaging is a crucial step in the commercial distribution of apples after harvest. However, achieving fast and accurate automated packaging remains a challenge. This study proposed for the first time a highly integrated four-arm robotic system and an effective implementation strategy specifically designed for automatic packaging of apples. In order to achieve accurate and simultaneous detection of multi-channel apples in complex visual conditions, an enhanced target detection model (i.e. YOLOv8s-GUC) based on the YOLOv8s architecture was developed by combining with a global attention mechanism (GAM), UniRepLKNet large kernel convolution structure, and CARAFE lightweight upsampling module. The results demonstrated that the YOLOv8s-GUC model can significantly improve detection accuracy and generalization for apples and related small features (such as stems and calyxes), achieving an [email protected] of 99.5 %. Furthermore, considering the complexity and real-time constraints of task planning in multi-arm robotic operations, this study further proposed an intelligent scheduling algorithm based on a Deep Q-Network (DQN), which enables efficient collaboration and collision-free operation among the robotic arms through real-time state perception and online decision-making. The results of simulations and real-world experiments indicated that the developed multi-arm robotic packaging system and scheduling strategy had high operational stability and efficiency in apple packaging, with a success rate of 100 % and an average packaging time of less than one second per apple. This study provides an effective and reliable solution for automated apple packaging.

包装是苹果收获后商业分销的关键一步。然而，实现快速和准确的自动化包装仍然是一个挑战。本研究首次提出了一种专门为苹果自动包装设计的高度集成的四臂机器人系统和有效的实施策略。为了实现复杂视觉条件下多通道苹果的准确、同时检测，结合全局注意机制（GAM）、UniRepLKNet大核卷积结构和CARAFE轻量级上采样模块，开发了基于YOLOv8s架构的增强型目标检测模型（即YOLOv8s- guc）。结果表明，YOLOv8s-GUC模型可以显著提高苹果及其相关小特征（如茎和花萼）的检测精度和泛化，达到99.5%的[email protected]。此外，考虑到多臂机器人作业任务规划的复杂性和实时性约束，本研究进一步提出了一种基于深度q网络（Deep Q-Network， DQN）的智能调度算法，通过实时状态感知和在线决策，实现机械臂间高效协同、无碰撞作业。仿真和实际实验结果表明，所开发的多臂机器人包装系统和调度策略在苹果包装中具有较高的运行稳定性和效率，成功率为100%，平均每个苹果的包装时间小于1秒。本研究为苹果自动化包装提供了有效可靠的解决方案。

{"title":"Multi-arm robotic system and strategy for the automatic packaging of apples","authors":"Yizhi Zhang , Liping Chen , Xin Li , Qicheng Li , Jiangbo Li","doi":"10.1016/j.aiia.2025.11.006","DOIUrl":"10.1016/j.aiia.2025.11.006","url":null,"abstract":"<div><div>Packaging is a crucial step in the commercial distribution of apples after harvest. However, achieving fast and accurate automated packaging remains a challenge. This study proposed for the first time a highly integrated four-arm robotic system and an effective implementation strategy specifically designed for automatic packaging of apples. In order to achieve accurate and simultaneous detection of multi-channel apples in complex visual conditions, an enhanced target detection model (i.e. YOLOv8s-GUC) based on the YOLOv8s architecture was developed by combining with a global attention mechanism (GAM), UniRepLKNet large kernel convolution structure, and CARAFE lightweight upsampling module. The results demonstrated that the YOLOv8s-GUC model can significantly improve detection accuracy and generalization for apples and related small features (such as stems and calyxes), achieving an [email protected] of 99.5 %. Furthermore, considering the complexity and real-time constraints of task planning in multi-arm robotic operations, this study further proposed an intelligent scheduling algorithm based on a Deep Q-Network (DQN), which enables efficient collaboration and collision-free operation among the robotic arms through real-time state perception and online decision-making. The results of simulations and real-world experiments indicated that the developed multi-arm robotic packaging system and scheduling strategy had high operational stability and efficiency in apple packaging, with a success rate of 100 % and an average packaging time of less than one second per apple. This study provides an effective and reliable solution for automated apple packaging.</div></div>","PeriodicalId":52814,"journal":{"name":"Artificial Intelligence in Agriculture","volume":"16 1","pages":"Pages 578-591"},"PeriodicalIF":12.4,"publicationDate":"2025-11-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145693435","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}

引用次数: 0

PlantSegNeRF: A few-shot, cross-species method for plant 3D instance point cloud reconstruction via joint-channel NeRF with multi-view image instance matching PlantSegNeRF：一种基于联合通道NeRF和多视图图像实例匹配的植物三维实例点云重建方法

IF 12.4 Q1 AGRICULTURE, MULTIDISCIPLINARY

Artificial Intelligence in Agriculture

Pub Date : 2025-11-24 DOI: 10.1016/j.aiia.2025.11.009

Xin Yang , Ruiming Du , Hanyang Huang , Jiayang Xie , Pengyao Xie , Leisen Fang , Ziyue Guo , Nanjun Jiang , Yu Jiang , Haiyan Cen

Organ segmentation of plant point clouds is a prerequisite for the high-resolution and accurate extraction of organ-level phenotypic traits. Although the fast development of deep learning has boosted much research on segmentation of plant point clouds, the existing techniques for organ segmentation still face limitations in resolution, segmentation accuracy, and generalizability across various plant species. In this study, we proposed a novel approach called plant segmentation neural radiance fields (PlantSegNeRF), aiming to directly generate high-precision instance point clouds from multi-view RGB image sequences for a wide range of plant species. PlantSegNeRF performed two-dimensional (2D) instance segmentation on the multi-view images to generate instance masks for each organ with a corresponding instance identification (ID). The multi-view instance IDs corresponding to the same plant organ were then matched and refined using a specially designed instance matching (IM) module. The instance NeRF was developed to render an implicit scene containing color, density, semantic and instance information, which was ultimately converted into high-precision plant instance point clouds based on volume density. The results proved that in semantic segmentation of point clouds, PlantSegNeRF outperformed the commonly used methods, demonstrating an average improvement of 16.1 %, 18.3 %, 17.8 %, and 24.2 % in precision, recall, F1-score, and intersection over union (IoU) compared to the second-best results on structurally complex datasets. More importantly, PlantSegNeRF exhibited significant advantages in instance segmentation. Across all plant datasets, it achieved average improvements of 11.7 %, 38.2 %, 32.2 % and 25.3 % in mean precision (mPrec), mean recall (mRec), mean coverage (mCov), and mean weighted coverage (mWCov), respectively. Furthermore, PlantSegNeRF demonstrates superior few-shot, cross-species performance, requiring only multi-view images of few plants to train models applicable to specific or similar varieties. This study extends organ-level plant phenotyping and provides a high-throughput way to supply high-quality 3D data for developing large-scale artificial intelligence (AI) models in plant science.

植物点云的器官分割是高分辨率和准确提取器官水平表型性状的先决条件。尽管深度学习的快速发展促进了植物点云分割的研究，但现有的器官分割技术在分辨率、分割精度和跨植物物种的可泛化性等方面仍然存在局限性。在这项研究中，我们提出了一种新的方法，称为植物分割神经辐射场（PlantSegNeRF），旨在从多视角RGB图像序列中直接生成高精度的实例点云，用于广泛的植物物种。PlantSegNeRF对多视图图像进行二维（2D）实例分割，为每个器官生成具有相应实例标识（ID）的实例掩码。然后使用专门设计的实例匹配（IM）模块对同一植物器官对应的多视图实例id进行匹配和细化。开发实例NeRF来渲染包含颜色、密度、语义和实例信息的隐式场景，最终根据体积密度转换成高精度的植物实例点云。结果表明，在点云的语义分割中，PlantSegNeRF优于常用的方法，在精度、召回率、f1得分和交汇比（IoU）方面，与结构复杂数据集的第二好的结果相比，平均提高了16.1%、18.3%、17.8%和24.2%。更重要的是，PlantSegNeRF在实例分割方面表现出显著的优势。在所有植物数据集中，它在平均精度（mPrec）、平均召回率（mRec）、平均覆盖率（mCov）和平均加权覆盖率（mWCov）方面分别实现了11.7%、38.2%、32.2%和25.3%的平均改进。此外，PlantSegNeRF展示了优越的少拍、跨物种性能，只需要少数植物的多视图图像来训练适用于特定或类似品种的模型。该研究扩展了器官水平的植物表型，为植物科学中大规模人工智能（AI）模型的开发提供了高通量的高质量3D数据。

{"title":"PlantSegNeRF: A few-shot, cross-species method for plant 3D instance point cloud reconstruction via joint-channel NeRF with multi-view image instance matching","authors":"Xin Yang , Ruiming Du , Hanyang Huang , Jiayang Xie , Pengyao Xie , Leisen Fang , Ziyue Guo , Nanjun Jiang , Yu Jiang , Haiyan Cen","doi":"10.1016/j.aiia.2025.11.009","DOIUrl":"10.1016/j.aiia.2025.11.009","url":null,"abstract":"<div><div>Organ segmentation of plant point clouds is a prerequisite for the high-resolution and accurate extraction of organ-level phenotypic traits. Although the fast development of deep learning has boosted much research on segmentation of plant point clouds, the existing techniques for organ segmentation still face limitations in resolution, segmentation accuracy, and generalizability across various plant species. In this study, we proposed a novel approach called plant segmentation neural radiance fields (PlantSegNeRF), aiming to directly generate high-precision instance point clouds from multi-view RGB image sequences for a wide range of plant species. PlantSegNeRF performed two-dimensional (2D) instance segmentation on the multi-view images to generate instance masks for each organ with a corresponding instance identification (ID). The multi-view instance IDs corresponding to the same plant organ were then matched and refined using a specially designed instance matching (IM) module. The instance NeRF was developed to render an implicit scene containing color, density, semantic and instance information, which was ultimately converted into high-precision plant instance point clouds based on volume density. The results proved that in semantic segmentation of point clouds, PlantSegNeRF outperformed the commonly used methods, demonstrating an average improvement of 16.1 %, 18.3 %, 17.8 %, and 24.2 % in precision, recall, F1-score, and intersection over union (IoU) compared to the second-best results on structurally complex datasets. More importantly, PlantSegNeRF exhibited significant advantages in instance segmentation. Across all plant datasets, it achieved average improvements of 11.7 %, 38.2 %, 32.2 % and 25.3 % in mean precision (mPrec), mean recall (mRec), mean coverage (mCov), and mean weighted coverage (mWCov), respectively. Furthermore, PlantSegNeRF demonstrates superior few-shot, cross-species performance, requiring only multi-view images of few plants to train models applicable to specific or similar varieties. This study extends organ-level plant phenotyping and provides a high-throughput way to supply high-quality 3D data for developing large-scale artificial intelligence (AI) models in plant science.</div></div>","PeriodicalId":52814,"journal":{"name":"Artificial Intelligence in Agriculture","volume":"16 1","pages":"Pages 546-564"},"PeriodicalIF":12.4,"publicationDate":"2025-11-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145623318","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}

引用次数: 0