Gert Kootstra
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Gert Kootstra has a total of 101 co-authors. Here are the top co-authors by number of shared publications:
- Eldert J. van Henten
Farm Technology, Department of Plant Sciences, Wageningen University and Research, 6700 AA Wageningen, The Netherlands
17 shared publications - Akshay Kumar Burusa
Agricultural Biosystems Engineering, Wageningen University and Research, 6708 PB Wageningen, The Netherlands
5 shared publications - David Rapado-Rincon
Agricultural Biosystems Engineering, Wageningen University & Research, Wageningen, The Netherlands
4 shared publications - Rick van Essen
Agricultural Biosystems Engineering Wageningen University and Research Wageningen The Netherlands
4 shared publications - Frits K. van Evert
Wageningen University & Research
3 shared publications - Haris Ahmad Khan
NTNU - Norwegian University of Science and Technology
3 shared publications
Latest Publications
- Open Access
- 11 August 2026
- Published by Elsevier in International Journal of Applied Earth Observation and Geoinformation
- Vol. 153
Deep spatial-spectral fusion of UAV RGB and hyperspectral imagery for potato plant disease detection
- 11 August 2026
- Published by Elsevier in International Journal of Applied Earth Observation and Geoinformation
- Vol. 153
Open AccessDeep spatial-spectral fusion of UAV RGB and hyperspectral imagery for potato plant disease detection
- 30 July 2026
- Published by Elsevier in Biosystems Engineering
- Vol. 271
From point clouds to plant traits: investigation of a pipeline for phenotyping of tomato plants using skeletonisation
- 30 July 2026
- Published by Elsevier in Biosystems Engineering
- Vol. 271
From point clouds to plant traits: investigation of a pipeline for phenotyping of tomato plants using skeletonisation
- Open Access
- 30 June 2026
- Published by Elsevier in SSRN Electronic Journal
Balancing Structural Complexity and Light Absorption Accuracy: A Case Study on a 3D Cucumber Plant Model
Accurate simulation of canopy light absorption increasingly relies on detailed 3D plant models. Yet it remains unclear which structural details are necessary and which can be simplified without compromising accuracy. Here, we used a 3D scan-derived, parameterized cucumber model to quantify how morphological parameterization and leaf geometric fidelity affect ray-tracing-based light absorption and photosynthesis simulations. At the whole-plant scale, sensitivity analysis showed that leaf area, internode angle, and internode length were the dominant determinants of light absorption, causing deviations of up to 40%. In contrast, petiole traits and leaf orientation generally changed total plant absorption by less than 2%, indicating limited influence of these parameters on plant-level outputs. Sensitivity also depended on light environment: in the morning when solar elevation is low, top-canopy traits became more influential because they intercepted more direct radiation, whereas under cloudy diffuse light, the effects of changes in leaf orientation and position were relatively small. Leaf geometric fidelity had a strong scale-dependent effect. Coarse leaf meshes with 2-10 faces introduced less than 10% error in total plant absorption, but errors exceeded 40% for individual leaves. A 20-face model balanced accuracy and computational cost at the plant level, and biologically meaningful face allocation further improved accuracy. Mixed-fidelity simulations showed that higher leaf detail was most beneficial in the middle canopy. These results provide practical rules for matching 3D plant model complexity to the intended simulation scale and output, enabling efficient yet accurate structural models for greenhouse digital-twin applications.
- 30 June 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessBalancing Structural Complexity and Light Absorption Accuracy: A Case Study on a 3D Cucumber Plant Model
Accurate simulation of canopy light absorption increasingly relies on detailed 3D plant models. Yet it remains unclear which structural details are necessary and which can be simplified without compromising accuracy. Here, we used a 3D scan-derived, parameterized cucumber model to quantify how morphological parameterization and leaf geometric fidelity affect ray-tracing-based light absorption and photosynthesis simulations. At the whole-plant scale, sensitivity analysis showed that leaf area, internode angle, and internode length were the dominant determinants of light absorption, causing deviations of up to 40%. In contrast, petiole traits and leaf orientation generally changed total plant absorption by less than 2%, indicating limited influence of these parameters on plant-level outputs. Sensitivity also depended on light environment: in the morning when solar elevation is low, top-canopy traits became more influential because they intercepted more direct radiation, whereas under cloudy diffuse light, the effects of changes in leaf orientation and position were relatively small. Leaf geometric fidelity had a strong scale-dependent effect. Coarse leaf meshes with 2-10 faces introduced less than 10% error in total plant absorption, but errors exceeded 40% for individual leaves. A 20-face model balanced accuracy and computational cost at the plant level, and biologically meaningful face allocation further improved accuracy. Mixed-fidelity simulations showed that higher leaf detail was most beneficial in the middle canopy. These results provide practical rules for matching 3D plant model complexity to the intended simulation scale and output, enabling efficient yet accurate structural models for greenhouse digital-twin applications.
- 20 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Robotics and Automation Letters
- Vol. 11 (6) , 7102-7109
AgriGS-SLAM: Orchard Mapping Across Seasons via Multi-View Gaussian Splatting SLAM
Autonomous robots in orchards require real-time 3D scene understanding despite repetitive row geometry, seasonal appearance changes, and wind-driven foliage motion. We present AgriGS-SLAM, a Visual–LiDAR SLAM framework that couples direct LiDAR odometry and loop closures with multi-camera 3D Gaussian Splatting (3DGS) rendering. Batch rasterization across complementary viewpoints recovers orchard structure under occlusions, while a unified gradient-driven map lifecycle executed between keyframes preserves fine details and bounds memory. Pose refinement is guided by a probabilistic LiDAR-based depth consistency term, back-propagated through the camera projection to tighten geometry-appearance coupling. We validate the system on a field platform in apple and pear orchards across dormancy, flowering, and harvesting, using a standardized trajectory protocol that evaluates both training-view and novel-view synthesis to reduce 3DGS overfitting in evaluation. Across seasons and sites, AgriGS-SLAM delivers sharper, more stable reconstructions and steadier trajectories than recent state-of-the-art 3DGS-based SLAM methods while maintaining real-time performance on-tractor. While demonstrated in orchard monitoring, the approach can be applied to other outdoor domains requiring robust multimodal perception.
- 20 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Robotics and Automation Letters
- Vol. 11 (6) , 7102-7109
AgriGS-SLAM: Orchard Mapping Across Seasons via Multi-View Gaussian Splatting SLAM
Autonomous robots in orchards require real-time 3D scene understanding despite repetitive row geometry, seasonal appearance changes, and wind-driven foliage motion. We present AgriGS-SLAM, a Visual–LiDAR SLAM framework that couples direct LiDAR odometry and loop closures with multi-camera 3D Gaussian Splatting (3DGS) rendering. Batch rasterization across complementary viewpoints recovers orchard structure under occlusions, while a unified gradient-driven map lifecycle executed between keyframes preserves fine details and bounds memory. Pose refinement is guided by a probabilistic LiDAR-based depth consistency term, back-propagated through the camera projection to tighten geometry-appearance coupling. We validate the system on a field platform in apple and pear orchards across dormancy, flowering, and harvesting, using a standardized trajectory protocol that evaluates both training-view and novel-view synthesis to reduce 3DGS overfitting in evaluation. Across seasons and sites, AgriGS-SLAM delivers sharper, more stable reconstructions and steadier trajectories than recent state-of-the-art 3DGS-based SLAM methods while maintaining real-time performance on-tractor. While demonstrated in orchard monitoring, the approach can be applied to other outdoor domains requiring robust multimodal perception.
- Open Access
- 11 April 2026
- Published by Elsevier in SSRN Electronic Journal
Prediction-guided Viewpoint Planning: Using Prior Knowledge about the Plant Structure for Efficient Fruit Node Perception
Robots are needed in the agricultural industry to compensate for the shortage of manual labour and perform labour-intensive tasks such as fruit harvesting. However, robots still struggle to accurately perceive the fruit nodes that need to be cut due to high levels of occlusion in plants. Active vision algorithms can help robots to improve their perception by allowing them to reason about the scene and purposefully plan the camera viewpoints such that any occlusion is overcome. However, current active vision algorithms treat the entire unknown space equally and explore it uniformly, which could lead to long exploration times and extensive sampling to explore sufficiently. The search and perception of fruit nodes can be improved by using knowledge about the plant structure, such as the fact that fruits are most likely to be found along the stem and tend to appear at regular intervals. In this paper, we propose a prediction-guided active vision algorithm that uses prior knowledge about the plant structure to predict the location of fruit nodes and uses these predicted locations to plan the camera viewpoints. The proposed approach was evaluated on the task of perceiving the fruit nodes of real-world tomato plants in a greenhouse. The proposed algorithm could predict the location of fruit nodes with an F1-score of 0.98 using three camera viewpoints. Further, using these predictions, the prediction-guided active vision algorithm was able to perceive 89.2% of the fruit nodes using three viewpoints, which was 7.4% more nodes compared to an active vision planner that did not predict the location of fruit nodes, and 22.9% more nodes compared to a random planner. The results clearly demonstrate the advantage of incorporating prior knowledge about the plant structure for efficient search and perception of fruit nodes. Hence, this approach can improve the efficiency of automated harvesting in greenhouse crops.
- 11 April 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessPrediction-guided Viewpoint Planning: Using Prior Knowledge about the Plant Structure for Efficient Fruit Node Perception
Robots are needed in the agricultural industry to compensate for the shortage of manual labour and perform labour-intensive tasks such as fruit harvesting. However, robots still struggle to accurately perceive the fruit nodes that need to be cut due to high levels of occlusion in plants. Active vision algorithms can help robots to improve their perception by allowing them to reason about the scene and purposefully plan the camera viewpoints such that any occlusion is overcome. However, current active vision algorithms treat the entire unknown space equally and explore it uniformly, which could lead to long exploration times and extensive sampling to explore sufficiently. The search and perception of fruit nodes can be improved by using knowledge about the plant structure, such as the fact that fruits are most likely to be found along the stem and tend to appear at regular intervals. In this paper, we propose a prediction-guided active vision algorithm that uses prior knowledge about the plant structure to predict the location of fruit nodes and uses these predicted locations to plan the camera viewpoints. The proposed approach was evaluated on the task of perceiving the fruit nodes of real-world tomato plants in a greenhouse. The proposed algorithm could predict the location of fruit nodes with an F1-score of 0.98 using three camera viewpoints. Further, using these predictions, the prediction-guided active vision algorithm was able to perceive 89.2% of the fruit nodes using three viewpoints, which was 7.4% more nodes compared to an active vision planner that did not predict the location of fruit nodes, and 22.9% more nodes compared to a random planner. The results clearly demonstrate the advantage of incorporating prior knowledge about the plant structure for efficient search and perception of fruit nodes. Hence, this approach can improve the efficiency of automated harvesting in greenhouse crops.
- Open Access
- 22 February 2026
- Published by Elsevier in Smart Agricultural Technology
- Vol. 13
Enhancing annotations for 5D apple pose estimation through 3D gaussian splatting (3DGS)
- 22 February 2026
- Published by Elsevier in Smart Agricultural Technology
- Vol. 13
Open AccessEnhancing annotations for 5D apple pose estimation through 3D gaussian splatting (3DGS)
- Open Access
- 24 January 2026
- Published by Elsevier in Smart Agricultural Technology
- Vol. 13
Lameness detection in dairy cows using pose estimation and bidirectional LSTMs
- 24 January 2026
- Published by Elsevier in Smart Agricultural Technology
- Vol. 13
Open AccessLameness detection in dairy cows using pose estimation and bidirectional LSTMs
- Open Access
- 21 July 2025
- Published by Elsevier in Biosystems Engineering
- Vol. 257
Multi-stage image-based approach for fish detection and weight estimation
- 21 July 2025
- Published by Elsevier in Biosystems Engineering
- Vol. 257
Open AccessMulti-stage image-based approach for fish detection and weight estimation
- Open Access
- 26 June 2025
- Published by Elsevier in Computers and Electronics in Agriculture
- Vol. 237
UAV-based path planning for efficient localization of non-uniformly distributed weeds using prior knowledge: A reinforcement-learning approach
- 26 June 2025
- Published by Elsevier in Computers and Electronics in Agriculture
- Vol. 237
Open AccessUAV-based path planning for efficient localization of non-uniformly distributed weeds using prior knowledge: A reinforcement-learning approach
- Open Access
A drone that learns to efficiently find objects in agricultural fields: from simulation to the real world
Drones are promising for data collection in precision agriculture, however, they are limited by their battery capacity. Efficient path planners are therefore required. This paper presents a drone path planner trained using Reinforcement Learning (RL) on an abstract simulation that uses object detections and uncertain prior knowledge. The RL agent controls the flight direction and can terminate the flight. By using the agent in combination with the drone's flight controller and a detection network to process camera images, it is possible to evaluate the performance of the agent on real-world data. In simulation, the agent yielded on average a 78% shorter flight path compared to a full coverage planner, at the cost of a 14% lower recall. On real-world data, the agent showed a 72% shorter flight path compared to a full coverage planner, however, at the cost of a 25% lower recall. The lower performance on real-world data was attributed to the real-world object distribution and the lower accuracy of prior knowledge, and shows potential for improvement. Overall, we concluded that for applications where it is not crucial to find all objects, such as weed detection, the learned-based path planner is suitable and efficient.
Open AccessA drone that learns to efficiently find objects in agricultural fields: from simulation to the real world
Drones are promising for data collection in precision agriculture, however, they are limited by their battery capacity. Efficient path planners are therefore required. This paper presents a drone path planner trained using Reinforcement Learning (RL) on an abstract simulation that uses object detections and uncertain prior knowledge. The RL agent controls the flight direction and can terminate the flight. By using the agent in combination with the drone's flight controller and a detection network to process camera images, it is possible to evaluate the performance of the agent on real-world data. In simulation, the agent yielded on average a 78% shorter flight path compared to a full coverage planner, at the cost of a 14% lower recall. On real-world data, the agent showed a 72% shorter flight path compared to a full coverage planner, however, at the cost of a 25% lower recall. The lower performance on real-world data was attributed to the real-world object distribution and the lower accuracy of prior knowledge, and shows potential for improvement. Overall, we concluded that for applications where it is not crucial to find all objects, such as weed detection, the learned-based path planner is suitable and efficient.