2025
1.
Dawood Ahmed; Ranjan Sapkota; Martin Churuvija; Manoj Karkee
Estimating optimal crop-load for individual branches in apple tree canopies using YOLOv8 Journal Article
In: Computers and Electronics in Agriculture, vol. 229, pp. 109697, 2025, ISSN: 0168-1699.
Abstract | Links | BibTeX | Tags:
@article{ahmed_estimating_2025,
title = {Estimating optimal crop-load for individual branches in apple tree canopies using YOLOv8},
author = {Dawood Ahmed and Ranjan Sapkota and Martin Churuvija and Manoj Karkee},
url = {https://www.sciencedirect.com/science/article/pii/S0168169924010883},
doi = {10.1016/j.compag.2024.109697},
issn = {0168-1699},
year = {2025},
date = {2025-02-01},
urldate = {2025-02-01},
journal = {Computers and Electronics in Agriculture},
volume = {229},
pages = {109697},
abstract = {Shortage of labor in fruit crop production has become a significant challenge in recent years. Therefore, mechanized and automated machines have emerged as promising alternatives to labor-intensive orchard operations such as harvesting, pruning, and thinning. One of the key aspects of the automated machines in accomplishing these tasks is their ability to identify tree canopy parts such as trunk and branches and estimate their geometric and topological parameters such as branch diameter, branch length, branch angles, and spacing between branches. By utilizing geometric parameters such as branch diameter, length, and orientation, researchers then can develop automated pruning and thinning systems that make more effective decisions to achieve optimal fruit yield and quality by accurately estimating the desired crop-load. In this study, we propose a machine vision system for estimating one of the canopy parameters in apple orchards: branch diameter. This parameter was used to estimate the optimal number of fruit that individual branches could bear in a commercial orchard, which provides a basis for robotic pruning, flower thinning, and fruitlet thinning so that desired fruit yield and quality could be achieved. Utilizing color and depth information collected with an RGB-D sensor (Azure Kinect DK, Microsoft, Redmond, WA), a YOLOv8-based instance segmentation technique was developed to identify trunks and branches of apple trees in the dormant season. We then applied a Principal Component Analysis (PCA) technique to estimate branch orientation, which was subsequently utilized to estimate branch diameter. The estimated branch diameter was used to calculate the Limb Cross Sectional Area (LCSA), which was then used to estimate optimal crop-load, as a larger LCSA indicates a higher potential fruit-bearing capacity of the branch. With this approach, Root Mean Squared Error (RMSE) for branch diameter estimation was calculated to be 2.06 mm (relative RMSE 10.82%) and the same for crop-load estimation (Number of fruits per branch) to be 3.93 (relative RMSE 22.25%). Our study demonstrated a promising workflow with a high level of performance in identifying and sizing branches of apple trees in a dynamic orchard environment and integrating farm management practices into automated decision-making for optimizing crop-load in apple orchards.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Shortage of labor in fruit crop production has become a significant challenge in recent years. Therefore, mechanized and automated machines have emerged as promising alternatives to labor-intensive orchard operations such as harvesting, pruning, and thinning. One of the key aspects of the automated machines in accomplishing these tasks is their ability to identify tree canopy parts such as trunk and branches and estimate their geometric and topological parameters such as branch diameter, branch length, branch angles, and spacing between branches. By utilizing geometric parameters such as branch diameter, length, and orientation, researchers then can develop automated pruning and thinning systems that make more effective decisions to achieve optimal fruit yield and quality by accurately estimating the desired crop-load. In this study, we propose a machine vision system for estimating one of the canopy parameters in apple orchards: branch diameter. This parameter was used to estimate the optimal number of fruit that individual branches could bear in a commercial orchard, which provides a basis for robotic pruning, flower thinning, and fruitlet thinning so that desired fruit yield and quality could be achieved. Utilizing color and depth information collected with an RGB-D sensor (Azure Kinect DK, Microsoft, Redmond, WA), a YOLOv8-based instance segmentation technique was developed to identify trunks and branches of apple trees in the dormant season. We then applied a Principal Component Analysis (PCA) technique to estimate branch orientation, which was subsequently utilized to estimate branch diameter. The estimated branch diameter was used to calculate the Limb Cross Sectional Area (LCSA), which was then used to estimate optimal crop-load, as a larger LCSA indicates a higher potential fruit-bearing capacity of the branch. With this approach, Root Mean Squared Error (RMSE) for branch diameter estimation was calculated to be 2.06 mm (relative RMSE 10.82%) and the same for crop-load estimation (Number of fruits per branch) to be 3.93 (relative RMSE 22.25%). Our study demonstrated a promising workflow with a high level of performance in identifying and sizing branches of apple trees in a dynamic orchard environment and integrating farm management practices into automated decision-making for optimizing crop-load in apple orchards.
2024
2.
Ranjan Sapkota; Zhichao Meng; Manoj Karkee
Synthetic meets authentic: Leveraging LLM generated datasets for YOLO11 and YOLOv10-based apple detection through machine vision sensors Journal Article
In: Smart Agricultural Technology, vol. 9, pp. 100614, 2024, ISSN: 2772-3755.
Abstract | Links | BibTeX | Tags:
@article{sapkota_synthetic_2024,
title = {Synthetic meets authentic: Leveraging LLM generated datasets for YOLO11 and YOLOv10-based apple detection through machine vision sensors},
author = {Ranjan Sapkota and Zhichao Meng and Manoj Karkee},
url = {https://www.sciencedirect.com/science/article/pii/S2772375524002193},
doi = {10.1016/j.atech.2024.100614},
issn = {2772-3755},
year = {2024},
date = {2024-12-01},
urldate = {2024-12-01},
journal = {Smart Agricultural Technology},
volume = {9},
pages = {100614},
abstract = {Training machine learning (ML) models for artificial intelligence (AI) and computer vision-based object detection process typically requires large, labeled datasets, a process often burdened by significant human effort and high costs associated with imaging systems and image acquisition. This research aimed to simplify image data collection for object detection in orchards by avoiding traditional fieldwork with different imaging sensors. Utilizing OpenAI's DALLE, a large language model (LLM) for realistic image generation, we generated and annotated a cost-effective dataset. This dataset, exclusively generated by LLM, was then utilized to train two state-of-the-art deep learning models: YOLOV10 and YOLO11. The YOLO11 model for apple detection was trained with its five configurations (YOLO11n, YOLO11 s, YOLO11 m, YOLO11l and YOLO11x), and YOLOv10 model with its six configurations (YOLOv10n, YOLOv10 s, YOLOv10 m, YOLOv10b, YOLOv10l and YOLOv10x), which was then tested with real-world (outdoor orchard) images captured by a digital (Nikon D5100) camera and a consumer RGB-D camera (Microsoft Azure Kinect). YOLO11 outperformed YOLOv10 as YOLO11x and YOLO11n exhibited superior precision of 0.917 and 0.916, respectively. Furthermore, YOLO11l demonstrated the highest recall among its counterparts, achieving a recall of 0.889. Likewise, the YOLO11n variant excelled in terms of mean average precision (mAP@50), achieving the highest value of 0.958. Validation tests against actual images collected through a digital camera (Nikon D5100) over Scilate apple variety in a commercial orchard environment showed a highest precision of 0.874 for YOLO11 s, recall of 0.877 for YOLO11l and mAP@50 of 0.91 for YOLO11x. Additionally, validation test against actual images collected through a Microsoft Azure camera over the same orchard showed a highest precision, recall and mAP@50 respectively of 0.924, 0.781 and 0.855 with YOLO11x. All variants of YOLO11 surprisingly demonstrated a pre-processing time of just 0.2 milliseconds (ms), which was faster than any variant of YOLOv10. The fastest inference time for the YOLO11n model using the training dataset generated by the language model was 3.2 ms, while YOLOv10n, fastest among YOLOv10 variants, had a longer inference time of 5.5 ms. Likewise, the fastest inference time for the sensor-based images was 7.1 ms (for Nikon D5100 camera images) and 4.7 ms (for Azure images) with YOLO11n. This study presents a pathway for generating large image datasets using LLM in challenging agricultural fields with minimal or no labor-intensive efforts in field data-collection, which could accelerate the development and deployment of computer vision and robotic technologies in orchard environments.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Training machine learning (ML) models for artificial intelligence (AI) and computer vision-based object detection process typically requires large, labeled datasets, a process often burdened by significant human effort and high costs associated with imaging systems and image acquisition. This research aimed to simplify image data collection for object detection in orchards by avoiding traditional fieldwork with different imaging sensors. Utilizing OpenAI's DALLE, a large language model (LLM) for realistic image generation, we generated and annotated a cost-effective dataset. This dataset, exclusively generated by LLM, was then utilized to train two state-of-the-art deep learning models: YOLOV10 and YOLO11. The YOLO11 model for apple detection was trained with its five configurations (YOLO11n, YOLO11 s, YOLO11 m, YOLO11l and YOLO11x), and YOLOv10 model with its six configurations (YOLOv10n, YOLOv10 s, YOLOv10 m, YOLOv10b, YOLOv10l and YOLOv10x), which was then tested with real-world (outdoor orchard) images captured by a digital (Nikon D5100) camera and a consumer RGB-D camera (Microsoft Azure Kinect). YOLO11 outperformed YOLOv10 as YOLO11x and YOLO11n exhibited superior precision of 0.917 and 0.916, respectively. Furthermore, YOLO11l demonstrated the highest recall among its counterparts, achieving a recall of 0.889. Likewise, the YOLO11n variant excelled in terms of mean average precision (mAP@50), achieving the highest value of 0.958. Validation tests against actual images collected through a digital camera (Nikon D5100) over Scilate apple variety in a commercial orchard environment showed a highest precision of 0.874 for YOLO11 s, recall of 0.877 for YOLO11l and mAP@50 of 0.91 for YOLO11x. Additionally, validation test against actual images collected through a Microsoft Azure camera over the same orchard showed a highest precision, recall and mAP@50 respectively of 0.924, 0.781 and 0.855 with YOLO11x. All variants of YOLO11 surprisingly demonstrated a pre-processing time of just 0.2 milliseconds (ms), which was faster than any variant of YOLOv10. The fastest inference time for the YOLO11n model using the training dataset generated by the language model was 3.2 ms, while YOLOv10n, fastest among YOLOv10 variants, had a longer inference time of 5.5 ms. Likewise, the fastest inference time for the sensor-based images was 7.1 ms (for Nikon D5100 camera images) and 4.7 ms (for Azure images) with YOLO11n. This study presents a pathway for generating large image datasets using LLM in challenging agricultural fields with minimal or no labor-intensive efforts in field data-collection, which could accelerate the development and deployment of computer vision and robotic technologies in orchard environments.
3.
Bhupinderjeet Singh; Tanvir Ferdousi; John T. Abatzoglou; Samarth Swarup; Jennifer C. Adam; Kirti Rajagopalan
Sensitivity of snow magnitude and duration to hydrology model parameters Journal Article
In: Journal of Hydrology, vol. 645, pp. 132193, 2024, ISSN: 0022-1694.
Abstract | Links | BibTeX | Tags: AI, Water
@article{singh_sensitivity_2024,
title = {Sensitivity of snow magnitude and duration to hydrology model parameters},
author = {Bhupinderjeet Singh and Tanvir Ferdousi and John T. Abatzoglou and Samarth Swarup and Jennifer C. Adam and Kirti Rajagopalan},
url = {https://www.sciencedirect.com/science/article/pii/S0022169424015890},
doi = {10.1016/j.jhydrol.2024.132193},
issn = {0022-1694},
year = {2024},
date = {2024-12-01},
urldate = {2024-12-01},
journal = {Journal of Hydrology},
volume = {645},
pages = {132193},
abstract = {Process-based hydrology models are critical for understanding streamflow and water supply under global change. However, these models require parameterization which introduces additional uncertainty into the models. The role that these parameters play in driving uncertainty is under-studied, especially for intermediary processes outside of streamflow. An important example in snowmelt dominant regions are intermediary processes related to snowpack accumulation and ablation. We examine the sensitivity of snow magnitude and duration to eleven parameters relevant to snow processes in the coupled crop-hydrology model VIC-CropSyst using a hybrid global\textendashlocal Distributed Evaluation of Local Sensitivity Analysis approach. With the Pacific Northwest US as a case study, our specific research questions are: (a) What is the sensitivity response of peak snow water equivalent (SWE) and snow duration and how does it vary in the parameter space? (b) What are the key drivers of the sensitivity response? and (c) Which of the most sensitive parameters can we immediately improve by leveraging existing data products? Both target variables were sensitive to less than four of the eleven parameters. We found that peak SWE was most sensitive to either the precipitation partitioning temperature threshold or the albedo of new snow, depending on the geography and associated interplay between hydro-meteorological factors. In contrast, snow duration was primarily sensitive to the albedo of new snow and the albedo decay coefficient during snowmelt. Machine learning explainability workflows applied on the sensitivity response explained the model behavior and determined key geographic and hydro-meteorological drivers of the sensitivity response. Regions where the significant precipitation co-occurred with near-freezing temperature exhibited higher sensitivity of peak SWE to precipitation partitioning. In contrast, much colder high-elevation regions that have a delayed snowmelt-driven runoff when downward shortwave radiation is higher, displayed more sensitivity to albedo parameters. We also noted differences in the list of key parameters and in the level of sensitivity between our work and the limited comparable existing work. This highlights the need for comprehensive sensitivity analysis of snow metrics to become a routine component of hydrology model application studies in addition to streamflow. This is critical for us to better understand model behavior, identify key model parameters that can benefit from more dynamic representation in the models, and strategically improve models to best support decision-makers.},
keywords = {AI, Water},
pubstate = {published},
tppubtype = {article}
}
Process-based hydrology models are critical for understanding streamflow and water supply under global change. However, these models require parameterization which introduces additional uncertainty into the models. The role that these parameters play in driving uncertainty is under-studied, especially for intermediary processes outside of streamflow. An important example in snowmelt dominant regions are intermediary processes related to snowpack accumulation and ablation. We examine the sensitivity of snow magnitude and duration to eleven parameters relevant to snow processes in the coupled crop-hydrology model VIC-CropSyst using a hybrid global–local Distributed Evaluation of Local Sensitivity Analysis approach. With the Pacific Northwest US as a case study, our specific research questions are: (a) What is the sensitivity response of peak snow water equivalent (SWE) and snow duration and how does it vary in the parameter space? (b) What are the key drivers of the sensitivity response? and (c) Which of the most sensitive parameters can we immediately improve by leveraging existing data products? Both target variables were sensitive to less than four of the eleven parameters. We found that peak SWE was most sensitive to either the precipitation partitioning temperature threshold or the albedo of new snow, depending on the geography and associated interplay between hydro-meteorological factors. In contrast, snow duration was primarily sensitive to the albedo of new snow and the albedo decay coefficient during snowmelt. Machine learning explainability workflows applied on the sensitivity response explained the model behavior and determined key geographic and hydro-meteorological drivers of the sensitivity response. Regions where the significant precipitation co-occurred with near-freezing temperature exhibited higher sensitivity of peak SWE to precipitation partitioning. In contrast, much colder high-elevation regions that have a delayed snowmelt-driven runoff when downward shortwave radiation is higher, displayed more sensitivity to albedo parameters. We also noted differences in the list of key parameters and in the level of sensitivity between our work and the limited comparable existing work. This highlights the need for comprehensive sensitivity analysis of snow metrics to become a routine component of hydrology model application studies in addition to streamflow. This is critical for us to better understand model behavior, identify key model parameters that can benefit from more dynamic representation in the models, and strategically improve models to best support decision-makers.
4.
Ranjan Sapkota; Achyut Paudel; Manoj Karkee
2024, (arXiv:2411.11285).
Abstract | Links | BibTeX | Tags: Computer Science - Artificial Intelligence, Computer Science - Computer Vision and Pattern Recognition
@misc{sapkota_zero-shot_2024,
title = {Zero-Shot Automatic Annotation and Instance Segmentation using LLM-Generated Datasets: Eliminating Field Imaging and Manual Annotation for Deep Learning Model Development},
author = {Ranjan Sapkota and Achyut Paudel and Manoj Karkee},
url = {http://arxiv.org/abs/2411.11285},
doi = {10.48550/arXiv.2411.11285},
year = {2024},
date = {2024-11-01},
urldate = {2024-11-01},
publisher = {arXiv},
abstract = {Currently, deep learning-based instance segmentation for various applications (e.g., Agriculture) is predominantly performed using a labor-intensive process involving extensive field data collection using sophisticated sensors, followed by careful manual annotation of images, presenting significant logistical and financial challenges to researchers and organizations. The process also slows down the model development and training process. In this study, we presented a novel method for deep learning-based instance segmentation of apples in commercial orchards that eliminates the need for labor-intensive field data collection and manual annotation. Utilizing a Large Language Model (LLM), we synthetically generated orchard images and automatically annotated them using the Segment Anything Model (SAM) integrated with a YOLO11 base model. This method significantly reduces reliance on physical sensors and manual data processing, presenting a major advancement in "Agricultural AI". The synthetic, auto-annotated dataset was used to train the YOLO11 model for Apple instance segmentation, which was then validated on real orchard images. The results showed that the automatically generated annotations achieved a Dice Coefficient of 0.9513 and an IoU of 0.9303, validating the accuracy and overlap of the mask annotations. All YOLO11 configurations, trained solely on these synthetic datasets with automated annotations, accurately recognized and delineated apples, highlighting the method's efficacy. Specifically, the YOLO11m-seg configuration achieved a mask precision of 0.902 and a mask mAP@50 of 0.833 on test images collected from a commercial orchard. Additionally, the YOLO11l-seg configuration outperformed other models in validation on 40 LLM-generated images, achieving the highest mask precision and mAP@50 metrics. Keywords: YOLO, SAM, SAMv2, YOLO11, YOLOv11, Segment Anything, YOLO-SAM},
note = {arXiv:2411.11285},
keywords = {Computer Science - Artificial Intelligence, Computer Science - Computer Vision and Pattern Recognition},
pubstate = {published},
tppubtype = {misc}
}
Currently, deep learning-based instance segmentation for various applications (e.g., Agriculture) is predominantly performed using a labor-intensive process involving extensive field data collection using sophisticated sensors, followed by careful manual annotation of images, presenting significant logistical and financial challenges to researchers and organizations. The process also slows down the model development and training process. In this study, we presented a novel method for deep learning-based instance segmentation of apples in commercial orchards that eliminates the need for labor-intensive field data collection and manual annotation. Utilizing a Large Language Model (LLM), we synthetically generated orchard images and automatically annotated them using the Segment Anything Model (SAM) integrated with a YOLO11 base model. This method significantly reduces reliance on physical sensors and manual data processing, presenting a major advancement in "Agricultural AI". The synthetic, auto-annotated dataset was used to train the YOLO11 model for Apple instance segmentation, which was then validated on real orchard images. The results showed that the automatically generated annotations achieved a Dice Coefficient of 0.9513 and an IoU of 0.9303, validating the accuracy and overlap of the mask annotations. All YOLO11 configurations, trained solely on these synthetic datasets with automated annotations, accurately recognized and delineated apples, highlighting the method's efficacy. Specifically, the YOLO11m-seg configuration achieved a mask precision of 0.902 and a mask mAP@50 of 0.833 on test images collected from a commercial orchard. Additionally, the YOLO11l-seg configuration outperformed other models in validation on 40 LLM-generated images, achieving the highest mask precision and mAP@50 metrics. Keywords: YOLO, SAM, SAMv2, YOLO11, YOLOv11, Segment Anything, YOLO-SAM
5.
Supriya Savalkar; Michael Pumphrey; Kimberly Campbell; Fabio Scarpare; Tanvir Ferdousi; Samarth Swarup; Claudio Stöckle; Kirti Rajagopalan
Earlier planting in a future climate fails to replicate historical production conditions for US spring wheat. Journal Article
In: 2024, (ISSN: 2693-5015).
Abstract | Links | BibTeX | Tags:
@article{savalkar_earlier_2024,
title = {Earlier planting in a future climate fails to replicate historical production conditions for US spring wheat.},
author = {Supriya Savalkar and Michael Pumphrey and Kimberly Campbell and Fabio Scarpare and Tanvir Ferdousi and Samarth Swarup and Claudio St\"{o}ckle and Kirti Rajagopalan},
url = {https://www.researchsquare.com/article/rs-4940971/v1},
doi = {10.21203/rs.3.rs-4940971/v1},
year = {2024},
date = {2024-10-01},
urldate = {2024-10-01},
publisher = {Research Square},
abstract = {Global warming can increase crop heat stress exposure, adversely affecting crop yields and quality. Earlier planting is widely considered in climate change literature as a potential adaptation strategy by shifting the growing season to cooler periods. However, the effectiveness of earlier planting in achieving overall temperature exposures during the crop growth season that are at least as favorable as historical conditions remains unclear. Our objective is to comprehensively assess the potential effectiveness of earlier planting as an adaptation strategy by addressing two key questions: How effective is earlier planting in reducing exposure to excessively high temperatures across different growth stages? What are the associated tradeoffs in temperature exposure, and can historical conditions be matched with this adaptation strategy? We focus on major US spring wheat growing states as a case study, analyzing lethal, critical, suboptimal, and optimal temperature thresholds by growth stage to quantify the impact of earlier planting. Our findings indicate that while earlier planting does reduce exposure to critical and lethal high-temperature categories during some reproductive stages, it generally fails to replicate historical production conditions for the majority of the US spring wheat production region (contributing 85% of the current production). This trend persists across all future time frames and emission scenarios. The Pacific Northwest US presents an exception; however, even in this region, certain early and late growth stages may experience worse-than-historical conditions requiring management. The planting window with historically equivalent temperature exposures also narrows from 11 weeks to 1-7 weeks, presenting additional logistical challenges. Given that the Pacific Northwest US accounts for less than 15% of the national spring wheat production, at a national scale, current production levels are unlikely to be sustained by primarily relying on earlier planting as an adaptation strategy. This is a critical consideration, especially given that many climate change assessments frequently list earlier planting as an effective adaptation strategy without the capacity to fully explore its tradeoffs. Exploring additional adaptation strategies to maintain current national production levels is important.},
note = {ISSN: 2693-5015},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Global warming can increase crop heat stress exposure, adversely affecting crop yields and quality. Earlier planting is widely considered in climate change literature as a potential adaptation strategy by shifting the growing season to cooler periods. However, the effectiveness of earlier planting in achieving overall temperature exposures during the crop growth season that are at least as favorable as historical conditions remains unclear. Our objective is to comprehensively assess the potential effectiveness of earlier planting as an adaptation strategy by addressing two key questions: How effective is earlier planting in reducing exposure to excessively high temperatures across different growth stages? What are the associated tradeoffs in temperature exposure, and can historical conditions be matched with this adaptation strategy? We focus on major US spring wheat growing states as a case study, analyzing lethal, critical, suboptimal, and optimal temperature thresholds by growth stage to quantify the impact of earlier planting. Our findings indicate that while earlier planting does reduce exposure to critical and lethal high-temperature categories during some reproductive stages, it generally fails to replicate historical production conditions for the majority of the US spring wheat production region (contributing 85% of the current production). This trend persists across all future time frames and emission scenarios. The Pacific Northwest US presents an exception; however, even in this region, certain early and late growth stages may experience worse-than-historical conditions requiring management. The planting window with historically equivalent temperature exposures also narrows from 11 weeks to 1-7 weeks, presenting additional logistical challenges. Given that the Pacific Northwest US accounts for less than 15% of the national spring wheat production, at a national scale, current production levels are unlikely to be sustained by primarily relying on earlier planting as an adaptation strategy. This is a critical consideration, especially given that many climate change assessments frequently list earlier planting as an effective adaptation strategy without the capacity to fully explore its tradeoffs. Exploring additional adaptation strategies to maintain current national production levels is important.
6.
Basavaraj R. Amogi; Nisit Pukrongta; Lav R. Khot; Bernardita V. Sallato
Edge compute algorithm enabled localized crop physiology sensing system for apple (textitMalus domestica Borkh.) crop water stress monitoring Journal Article
In: Computers and Electronics in Agriculture, vol. 224, pp. 109137, 2024, ISSN: 0168-1699.
Abstract | Links | BibTeX | Tags: AI, Farm Ops
@article{amogi_edge_2024,
title = {Edge compute algorithm enabled localized crop physiology sensing system for apple (textitMalus domestica Borkh.) crop water stress monitoring},
author = {Basavaraj R. Amogi and Nisit Pukrongta and Lav R. Khot and Bernardita V. Sallato},
url = {https://www.sciencedirect.com/science/article/pii/S0168169924005283},
doi = {10.1016/j.compag.2024.109137},
issn = {0168-1699},
year = {2024},
date = {2024-09-01},
urldate = {2024-09-01},
journal = {Computers and Electronics in Agriculture},
volume = {224},
pages = {109137},
abstract = {Elevated air temperature (\>35 ℃) combined with intense solar radiation can cause heat stress related damage to apple (Malus domestica Borkh.) fruits (e.g., sunburn) and increase tree evapotranspiration demand. Current heat stress mitigation techniques (e.g., evaporative cooling and netting) may protect fruits but can skew the tree evapotranspiration rates, preventing precision under-tree irrigation. A detailed understanding of heat stress mitigation techniques on tree fruit water status is critical for optimized irrigation scheduling and reduced crop losses. This study aimed to quantify water stress using a localized edge-compute-enabled crop physiology sensing system (CPSS), developed previously for fruit heat stress management. The CPSS is capable of acquiring thermal infrared and RGB images of the scene at predetermined interval. In this study, the edge compute algorithm on CPSS was amended to estimate crop water stress index (CWSI). Developed algorithm was validated for its accuracy in predicting the crop water stress under four different heat stress mitigation techniques namely: conventional overhead sprinklers, foggers, netting, and combinations of foggers and netting. A CPSS node was deployed in each treatment for acquiring thermal infrared and RGB images. Acquired imagery data were used to estimate CWSI using the modified algorithm. The algorithm-estimated CWSI showed significant negative correlation with stem water potential measurements (r = -0.8, p \< 0.01). The heat stress mitigation techniques had varying effects on sensitivity of estimated CWSI. Algorithm estimated CWSI was most sensitive to changes in water stress under fogging (r = 0.76) and least sensitive under neeting (r = -0.65). Overall, the use of real-time CWSI estimates in conjunction with heat stress monitoring could help improve precision irrigation management, enabling timely actuation of the under tree drip irrigation in apple orchards.},
keywords = {AI, Farm Ops},
pubstate = {published},
tppubtype = {article}
}
Elevated air temperature (>35 ℃) combined with intense solar radiation can cause heat stress related damage to apple (Malus domestica Borkh.) fruits (e.g., sunburn) and increase tree evapotranspiration demand. Current heat stress mitigation techniques (e.g., evaporative cooling and netting) may protect fruits but can skew the tree evapotranspiration rates, preventing precision under-tree irrigation. A detailed understanding of heat stress mitigation techniques on tree fruit water status is critical for optimized irrigation scheduling and reduced crop losses. This study aimed to quantify water stress using a localized edge-compute-enabled crop physiology sensing system (CPSS), developed previously for fruit heat stress management. The CPSS is capable of acquiring thermal infrared and RGB images of the scene at predetermined interval. In this study, the edge compute algorithm on CPSS was amended to estimate crop water stress index (CWSI). Developed algorithm was validated for its accuracy in predicting the crop water stress under four different heat stress mitigation techniques namely: conventional overhead sprinklers, foggers, netting, and combinations of foggers and netting. A CPSS node was deployed in each treatment for acquiring thermal infrared and RGB images. Acquired imagery data were used to estimate CWSI using the modified algorithm. The algorithm-estimated CWSI showed significant negative correlation with stem water potential measurements (r = -0.8, p < 0.01). The heat stress mitigation techniques had varying effects on sensitivity of estimated CWSI. Algorithm estimated CWSI was most sensitive to changes in water stress under fogging (r = 0.76) and least sensitive under neeting (r = -0.65). Overall, the use of real-time CWSI estimates in conjunction with heat stress monitoring could help improve precision irrigation management, enabling timely actuation of the under tree drip irrigation in apple orchards.
7.
Ranjan Sapkota; Dawood Ahmed; Manoj Karkee
Comparing YOLOv8 and Mask R-CNN for instance segmentation in complex orchard environments Journal Article
In: Artificial Intelligence in Agriculture, vol. 13, pp. 84–99, 2024, ISSN: 2589-7217.
Abstract | Links | BibTeX | Tags: Artificial intelligence, Automation, Deep learning, Machine Learning, Machine vision, Mask R-CNN, Robotics, YOLOv8
@article{sapkota_comparing_2024,
title = {Comparing YOLOv8 and Mask R-CNN for instance segmentation in complex orchard environments},
author = {Ranjan Sapkota and Dawood Ahmed and Manoj Karkee},
url = {https://www.sciencedirect.com/science/article/pii/S258972172400028X},
doi = {10.1016/j.aiia.2024.07.001},
issn = {2589-7217},
year = {2024},
date = {2024-09-01},
urldate = {2024-09-01},
journal = {Artificial Intelligence in Agriculture},
volume = {13},
pages = {84\textendash99},
abstract = {Instance segmentation, an important image processing operation for automation in agriculture, is used to precisely delineate individual objects of interest within images, which provides foundational information for various automated or robotic tasks such as selective harvesting and precision pruning. This study compares the one-stage YOLOv8 and the two-stage Mask R-CNN machine learning models for instance segmentation under varying orchard conditions across two datasets. Dataset 1, collected in dormant season, includes images of dormant apple trees, which were used to train multi-object segmentation models delineating tree branches and trunks. Dataset 2, collected in the early growing season, includes images of apple tree canopies with green foliage and immature (green) apples (also called fruitlet), which were used to train single-object segmentation models delineating only immature green apples. The results showed that YOLOv8 performed better than Mask R-CNN, achieving good precision and near-perfect recall across both datasets at a confidence threshold of 0.5. Specifically, for Dataset 1, YOLOv8 achieved a precision of 0.90 and a recall of 0.95 for all classes. In comparison, Mask R-CNN demonstrated a precision of 0.81 and a recall of 0.81 for the same dataset. With Dataset 2, YOLOv8 achieved a precision of 0.93 and a recall of 0.97. Mask R-CNN, in this single-class scenario, achieved a precision of 0.85 and a recall of 0.88. Additionally, the inference times for YOLOv8 were 10.9 ms for multi-class segmentation (Dataset 1) and 7.8 ms for single-class segmentation (Dataset 2), compared to 15.6 ms and 12.8 ms achieved by Mask R-CNN's, respectively. These findings show YOLOv8's superior accuracy and efficiency in machine learning applications compared to two-stage models, specifically Mask-R-CNN, which suggests its suitability in developing smart and automated orchard operations, particularly when real-time applications are necessary in such cases as robotic harvesting and robotic immature green fruit thinning.},
keywords = {Artificial intelligence, Automation, Deep learning, Machine Learning, Machine vision, Mask R-CNN, Robotics, YOLOv8},
pubstate = {published},
tppubtype = {article}
}
Instance segmentation, an important image processing operation for automation in agriculture, is used to precisely delineate individual objects of interest within images, which provides foundational information for various automated or robotic tasks such as selective harvesting and precision pruning. This study compares the one-stage YOLOv8 and the two-stage Mask R-CNN machine learning models for instance segmentation under varying orchard conditions across two datasets. Dataset 1, collected in dormant season, includes images of dormant apple trees, which were used to train multi-object segmentation models delineating tree branches and trunks. Dataset 2, collected in the early growing season, includes images of apple tree canopies with green foliage and immature (green) apples (also called fruitlet), which were used to train single-object segmentation models delineating only immature green apples. The results showed that YOLOv8 performed better than Mask R-CNN, achieving good precision and near-perfect recall across both datasets at a confidence threshold of 0.5. Specifically, for Dataset 1, YOLOv8 achieved a precision of 0.90 and a recall of 0.95 for all classes. In comparison, Mask R-CNN demonstrated a precision of 0.81 and a recall of 0.81 for the same dataset. With Dataset 2, YOLOv8 achieved a precision of 0.93 and a recall of 0.97. Mask R-CNN, in this single-class scenario, achieved a precision of 0.85 and a recall of 0.88. Additionally, the inference times for YOLOv8 were 10.9 ms for multi-class segmentation (Dataset 1) and 7.8 ms for single-class segmentation (Dataset 2), compared to 15.6 ms and 12.8 ms achieved by Mask R-CNN's, respectively. These findings show YOLOv8's superior accuracy and efficiency in machine learning applications compared to two-stage models, specifically Mask-R-CNN, which suggests its suitability in developing smart and automated orchard operations, particularly when real-time applications are necessary in such cases as robotic harvesting and robotic immature green fruit thinning.
8.
Dawood Ahmed, Ranjan Sapkota, Martin Churuvija, Matthew Whiting, Manoj Karkee
Slicing-Aided Hyper Inference for Enhanced Fruit Bud Detection and Counting in Apple Orchards during Dormant Season Proceedings Article
In: ASABE, 2024.
Abstract | Links | BibTeX | Tags:
@inproceedings{noauthor_slicing-aided_nodateb,
title = {Slicing-Aided Hyper Inference for Enhanced Fruit Bud Detection and Counting in Apple Orchards during Dormant Season},
author = {Dawood Ahmed, Ranjan Sapkota, Martin Churuvija, Matthew Whiting, Manoj Karkee},
url = {https://doi.org/10.13031/aim.202401055},
year = {2024},
date = {2024-08-02},
publisher = {ASABE},
abstract = {To meet the global food demand amidst the growing population, there‘s a critical need to augment productivity in the specialty crop industry. However, workforce shortage in agriculture presents a significant challenge to increasing agricultural productivity, emphasizing the need for automating various farm operations such as harvesting, canopy management, and crop-load management practices. A critical aspect of crop-load management is determining existing crop-load along with target yield, which can be used for optimizing crop-load in orchards. Buds, the initial phase of the flowering and fruiting process, are critical for early crop-load estimation (number of fruiting sites per branch or tree) and can facilitate informed pruning decisions to achieve the desired crop-load. Yet, the detection of these buds poses a huge challenge due to their miniature size and variable environmental conditions (including lighting) of commercial orchards. This research explores the performances of a Convolutional Neural Network (CNN) model and a Detection Transformer (DETR) model in detecting fruiting buds in apple orchards, using standard inference and a hyper inferencing technique. Additionally, the system quantifies buds and employs an association algorithm to correlate each bud to its respective branch, facilitating the estimation of the number of buds per branch. Despite the challenges due to the size of buds and environmental complexity, the proposed system demonstrated precision of 80.2 %, recall of 80.1 %, and f1-score of 80.15 % in detecting fruit buds, using a YOLOv8 model. The use of the Slicing Aided Hyper Inferencing Technique improved the overall recall score of the model. Additionally, the system quantified buds in a frame for the most salient branch using the branch-bud association with an accuracy of 73%. This study demonstrated a promising workflow in detecting and counting fruiting buds in a dynamic orchard environment to facilitate current manual pruning operation as well as future studies on automated pruning.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
To meet the global food demand amidst the growing population, there‘s a critical need to augment productivity in the specialty crop industry. However, workforce shortage in agriculture presents a significant challenge to increasing agricultural productivity, emphasizing the need for automating various farm operations such as harvesting, canopy management, and crop-load management practices. A critical aspect of crop-load management is determining existing crop-load along with target yield, which can be used for optimizing crop-load in orchards. Buds, the initial phase of the flowering and fruiting process, are critical for early crop-load estimation (number of fruiting sites per branch or tree) and can facilitate informed pruning decisions to achieve the desired crop-load. Yet, the detection of these buds poses a huge challenge due to their miniature size and variable environmental conditions (including lighting) of commercial orchards. This research explores the performances of a Convolutional Neural Network (CNN) model and a Detection Transformer (DETR) model in detecting fruiting buds in apple orchards, using standard inference and a hyper inferencing technique. Additionally, the system quantifies buds and employs an association algorithm to correlate each bud to its respective branch, facilitating the estimation of the number of buds per branch. Despite the challenges due to the size of buds and environmental complexity, the proposed system demonstrated precision of 80.2 %, recall of 80.1 %, and f1-score of 80.15 % in detecting fruit buds, using a YOLOv8 model. The use of the Slicing Aided Hyper Inferencing Technique improved the overall recall score of the model. Additionally, the system quantified buds in a frame for the most salient branch using the branch-bud association with an accuracy of 73%. This study demonstrated a promising workflow in detecting and counting fruiting buds in a dynamic orchard environment to facilitate current manual pruning operation as well as future studies on automated pruning.
9.
Mohsen Farajijalal, Samira Malek, Arash Toudeshki, Joshua H. Viers, Reza Ehsani
Data-Driven Model to Improve Mechanical Harvesters for Nut Trees Proceedings Article
In: ASABE, 2024.
Abstract | Links | BibTeX | Tags:
@inproceedings{noauthor_data-driven_nodateb,
title = {Data-Driven Model to Improve Mechanical Harvesters for Nut Trees},
author = {Mohsen Farajijalal, Samira Malek, Arash Toudeshki, Joshua H. Viers, Reza Ehsani},
url = {https://doi.org/10.13031/aim.202400858},
year = {2024},
date = {2024-08-02},
publisher = {ASABE},
abstract = {The economic impact of California‘s agricultural sector is the most important in the United States. California produces more than 80% of the world‘s almonds. Currently, the trunk shaker harvesting machine is the most widely used for harvesting almond trees in California. During the harvest, the operator, based on their experience, set the shaking parameters such as duration and shaking frequency. Manual adjustments by operators lead to variability in fruit removal, and in some cases, it could cause tree damage. This study aimed to develop a data-driven mathematical model required to build an intelligent tree shaker machine capable of optimizing shaking parameters autonomously. A sensor system to monitor force distribution throughout the tree canopy was designed and implemented. Data was collected from an almond orchard in California during the summer of 2023. A quadratic mathematical model was developed using machine learning regression methods to estimate relationships between trunk diameter, acceleration, and shaking duration. The findings show that trunk diameter positively correlates with acceleration and shaking time duration. Our research demonstrates the potential for intelligent harvesting machines that can adjust parameters based on real-time sensor data, ultimately improving fruit removal efficiency, and potentially reducing tree damage.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
The economic impact of California‘s agricultural sector is the most important in the United States. California produces more than 80% of the world‘s almonds. Currently, the trunk shaker harvesting machine is the most widely used for harvesting almond trees in California. During the harvest, the operator, based on their experience, set the shaking parameters such as duration and shaking frequency. Manual adjustments by operators lead to variability in fruit removal, and in some cases, it could cause tree damage. This study aimed to develop a data-driven mathematical model required to build an intelligent tree shaker machine capable of optimizing shaking parameters autonomously. A sensor system to monitor force distribution throughout the tree canopy was designed and implemented. Data was collected from an almond orchard in California during the summer of 2023. A quadratic mathematical model was developed using machine learning regression methods to estimate relationships between trunk diameter, acceleration, and shaking duration. The findings show that trunk diameter positively correlates with acceleration and shaking time duration. Our research demonstrates the potential for intelligent harvesting machines that can adjust parameters based on real-time sensor data, ultimately improving fruit removal efficiency, and potentially reducing tree damage.
10.
Bhupinderjeet Singh; Mingliang Liu; John Abatzoglou; Jennifer Adam; Kirti Rajagopalan
Dynamic precipitation phase partitioning improves modeled simulations of snow across the Northwest US Journal Article
In: EGUsphere, pp. 1–24, 2024, (Publisher: Copernicus GmbH).
Abstract | Links | BibTeX | Tags:
@article{singh_dynamic_2024,
title = {Dynamic precipitation phase partitioning improves modeled simulations of snow across the Northwest US},
author = {Bhupinderjeet Singh and Mingliang Liu and John Abatzoglou and Jennifer Adam and Kirti Rajagopalan},
url = {https://egusphere.copernicus.org/preprints/2024/egusphere-2024-2284/},
doi = {10.5194/egusphere-2024-2284},
year = {2024},
date = {2024-08-01},
urldate = {2024-08-01},
journal = {EGUsphere},
pages = {1\textendash24},
abstract = {\<p\>\<strong class="journal-contentHeaderColor"\>Abstract.\</strong\> While the importance of dynamic precipitation phase partitioning to get accurate estimates of rain versus snow amounts has been established, hydrology models rely on simplistic static temperature-based partitioning. We evaluate model skill changes for a suite of snow metrics between static and dynamic partitioning. We used the VIC-CropSyst coupled crop hydrology model across the Pacific Northwest US as a case study. We found that transition to the dynamic method resulted in a better match between modeled and observed (a) peak snow water equivalent (SWE) magnitude and timing (textasciitilde50 % mean error reduction), (b) daily SWE in winter months (reduction of relative bias from -30 % to -4 %), and (c) snow-start dates (mean reduction in bias from 7 days to 0 days) for a majority of the observational snow telemetry stations considered (depending on the metric, 75 % to 88 % of stations showed improvements). However, there was a degradation in model-observation agreement for snow-off dates, likely because errors in modeled snowmelt dynamics\—which cannot be resolved by changing the precipitation partitioning\—become important at the end of the cold season. Additionally, the transition from static to dynamic partitioning resulted in an 8 % mean increase in the snowmelt contribution to runoff. These results emphasize that the hydrological modeling community should transition to incorporating dynamic precipitation partitioning so we can better understand model behavior, improve model accuracies, and better support management decision support for water resources.\</p\>},
note = {Publisher: Copernicus GmbH},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
<p><strong class="journal-contentHeaderColor">Abstract.</strong> While the importance of dynamic precipitation phase partitioning to get accurate estimates of rain versus snow amounts has been established, hydrology models rely on simplistic static temperature-based partitioning. We evaluate model skill changes for a suite of snow metrics between static and dynamic partitioning. We used the VIC-CropSyst coupled crop hydrology model across the Pacific Northwest US as a case study. We found that transition to the dynamic method resulted in a better match between modeled and observed (a) peak snow water equivalent (SWE) magnitude and timing (textasciitilde50 % mean error reduction), (b) daily SWE in winter months (reduction of relative bias from -30 % to -4 %), and (c) snow-start dates (mean reduction in bias from 7 days to 0 days) for a majority of the observational snow telemetry stations considered (depending on the metric, 75 % to 88 % of stations showed improvements). However, there was a degradation in model-observation agreement for snow-off dates, likely because errors in modeled snowmelt dynamics—which cannot be resolved by changing the precipitation partitioning—become important at the end of the cold season. Additionally, the transition from static to dynamic partitioning resulted in an 8 % mean increase in the snowmelt contribution to runoff. These results emphasize that the hydrological modeling community should transition to incorporating dynamic precipitation partitioning so we can better understand model behavior, improve model accuracies, and better support management decision support for water resources.</p>
