Source-linked AI summary
AGRICAM: A Track-Mounted Crop Pollination Monitoring Robot
Malika Nisal Ratnayake, Adel N. Toosi, James Cook, Romina Rader, Alan Dorin
TL;DR
Commercial-scale pollination monitoring needs greater spatial coverage without sacrificing fine-resolution observations. The paper presents AGRICAM, an autonomous track-mounted robot that combines sensing, localisation, connectivity, and computer vision, and demonstrates its use in commercial blueberry polytunnels for mapping pollinator activity and microclimate-associated variation.
Problem
Existing pollination monitoring must balance detailed observations against spatial coverage and labour, motivating scalable fine-resolution monitoring across protected cropping environments.
Method
AGRICAM autonomously traverses programmable track stations while collecting video, location, and microclimate data for computer-vision-based spatiotemporal pollinator analysis.
Results
The system captured spatial and temporal insect-activity variation in 80 m industrial polytunnels on a commercial farm, including 39 manually observed honeybees with 84.6% recall and 89.2% precision in validation videos.
Takeaways & Limitations
AGRICAM supports precision pollination management and data-informed decision-making at commercial scale.
Takeaways & Limitations
The prototype requires manual battery recharging, and battery capacity trades operating time against system mass and power demand.
Abstract
from arXiv · showhide
Insect pollination is critical for global food production, yet monitoring pollinators at commercial farm scale remains a challenge. Recent advances in computer vision and deep learning have enabled detailed analysis of pollinator behaviour, but monitoring must trade-off detail against spatial coverage and human or technological resources. This paper presents the Automated Guided Robot for Insect and Crop Activity Monitoring (AGRICAM), a purpose-built robotic system designed to meet the requirements of large-scale pollination monitoring in protected cropping systems. AGRICAM operates autonomously on low-cost, easily installed track for movement along crop rows, without disrupting farm operations or insect behaviour. The platform integrates two RGB cameras, microclimate sensors, GPS and RFID modules, motion sensors, and 4G cellular network connectivity for data transmission. A web interface enables remote device configuration and scheduling. The system autonomously captures video and image data of insects' locations and local environmental conditions. These are transferred to the cloud and analysed using computer vision models to quantify pollinator visitation and spatio-temporal activity variation. We deployed the system on a commercial blueberry farm to demonstrate and test its capability. It successfully mapped insect pollination patterns across 80 m long industrial polytunnels over 30 hours. This data enabled spatial analyses of insect activity we used to confirm a uniform pollinator distribution within polytunnels, as desired by the farm management team. The data also highlighted variation of insect activity associated with time of day and microclimate. AGRICAM therefore has been shown to be a scalable, automated crop pollination monitor that can support data-driven decisions to enhance pollination management, thereby improving crop productivity and food security.
1. Introduction
Pollination is vital to crop production, but existing monitoring methods are labour-intensive, indirect, or spatially limited. The paper introduces AGRICAM to provide scalable, fine-resolution monitoring across protected cropping environments.
- Pollination importance: 35% of global crop production volume depends on animal-mediated pollination, which supports nearly 75% of leading food crops.Pollination services also have an estimated annual global economic value of USD 235B–577B.
- Pollination importance: Pollinator populations face interacting stressors, making timely monitoring of pollinator–crop interactions important for sustaining crop productivity and long-term pollinator health.The paper notes that pollinator effectiveness differs between crops, so population size alone is insufficient.
- Monitoring gap: Conventional monitoring is time-consuming, labour-intensive, expertise-dependent, and vulnerable to observer bias and reduced reproducibility.Methods include transect walks, direct flower observations, and passive sampling.
- Monitoring gap: Hive traffic and wingbeat-based systems do not directly measure flower visitation, limiting interpretation for unfamiliar insects or non-target foraging.These approaches may estimate abundance or hive activity without documenting crop foraging.
- Monitoring gap: Fixed-camera systems can provide high-resolution flower observations but cover only small numbers of flower clusters, which is inadequate for 50–100 m polytunnels.Dynamic agricultural conditions, including changing illumination and plant movement, further complicate automated monitoring.
- Paper contribution: AGRICAM combines a suspended track-based robot, vision and microclimate sensing, RFID localisation, 4G connectivity, web control, and computer vision for high-resolution spatiotemporal monitoring.The system was designed for autonomous pollination monitoring in protected cropping and field validation on a commercial blueberry farm.
2. Methodology
The paper’s methodology uses AGRICAM to acquire field video, location, and microclimate data, then transfers and processes them through AI-facilitated workflows for pollinator tracking and spatiotemporal analysis.
- Data-processing pipeline: AGRICAM collects video, location, and microclimate data in the field for automated insect-pollinator detection and tracking.Data are transferred through the cloud or USB storage before processing and analysis.
- Data-processing pipeline: The processing workflow analyses the collected data and produces visualisations of spatiotemporal microclimate and insect-behaviour patterns.The pipeline connects field acquisition, data transfer, AI-facilitated processing, and interpretation.
2.1. System Design
AGRICAM is an autonomous track-mounted monitoring robot designed for stable, programmable movement through protected crop rows while collecting position, video, and environmental data.
- Platform and track: AGRICAM rolls along a suspended, low-maintenance PVC track, avoiding ground contact and reducing interference with soil, harvesting paths, machinery, and farm operations.The water-resistant enclosure and suspended configuration support operation in protected cropping environments.
- Navigation and positioning: RFID tags provide precise station localisation independent of lighting or weather, while GNSS supplies global position data when available.A robot-mounted RFID reader uses a two-phase read during deceleration to reduce tag misreading.
- Navigation and positioning: When RFID reads fail, the robot estimates its station by incrementing or decrementing the previous station index according to travel direction within track limits.The estimate is bounded by S_min and S_max.
- Path planning and scheduling: Limit switches provide end-of-track safety, while automatic scheduling can skip an incomplete cycle and return the robot to a safe station at day’s end.These behaviours address navigation failures and inactivity-related exposure or security risks.
- Path planning and scheduling: Monitoring stations are programmable subsets of RFID tags, and each complete traversal from and back to a home station forms a monitoring cycle.Users define station spacing and cycle intervals to set spatial and temporal sampling resolution.
- Motion control: A single 12 V gear motor with compliant traction supports track traversal, while PWM enables gradual acceleration and deceleration at monitoring stations.The single-motor configuration is intended to keep energy consumption low and extend battery life.
- Integrated system: The system combines RGB imaging, microclimate sensing, RFID identification, GPS, motion sensing, and wireless data transmission in a farm-deployed monitoring platform.Figure 2 depicts the track installation, robot components, and commercial blueberry-farm deployment.
- Connectivity and control: 4G connectivity and a locally hosted web interface support remote monitoring, configuration, diagnostics, camera access, data download, and motion control.The interface includes status, path-map, diagnostics, and configuration modules.
2.2. Implementation and field evaluation
AGRICAM was deployed on two commercial blueberry rows to collect distributed environmental and video data, then process insect trajectories and flower visits for pollination analysis. Its workflow combined autonomous station-based sampling with computer-vision tracking, trajectory smoothing, and proximity-based visit classification.
- Field evaluation: Two independent rail tracks were installed in commercial blueberry polytunnels, monitoring rows approximately 90 m and 81 m long.The deployment occurred under standard commercial production conditions at Costa Berries Exchange in New South Wales, Australia.
- Data collection: 41 and 36 monitoring stations were positioned along Rows A and B at 2.25–2.75 m intervals, respectively.At each station, microclimate data were recorded every 15 s and 30 s video clips were captured during outward and return traversals.
- Data collection: AGRICAM recorded environmental and video data autonomously, storing footage onboard for later transfer and offline processing.A 256 GB SD card stored onboard data, after which recordings were transferred via a local area network.
- Video processing: Polytrack combined foreground–background segmentation with deep-learning object detection to track insect motion and detect blueberry flowers.Separate YOLOv12 models were trained for insect pollinators and flowers, while MOG2 parameters were tuned using videos from varied environmental and lighting conditions.
- Video processing: The insect detector achieved 97.5% recall on the test split at a confidence threshold of 0.001.The model was trained from scratch on 2,844 annotated honeybee images containing 3,271 labelled instances.
- Trajectory and visit analysis: Flower visits were defined by sustained assignment of an insect to the same proximate flower for at least T consecutive frames.Validated trajectories were represented by frame indices and image coordinates, with missing detections interpolated and coordinates smoothed before visit quantification.
3. Results
AGRICAM operated under field conditions, providing distributed video and microclimate measurements across blueberry rows. Polytrack validated insect and flower-visit detection, enabling fine-scale analyses of pollinator activity across space, time, and environmental conditions.
- System deployment evaluation: 251 seconds of mean daily monitoring time per station per row provided distributed coverage across field conditions.Monitoring averaged approximately 82–86 seconds per point during peak periods and 51 seconds during early morning and late afternoon periods.
- System deployment evaluation: Comparable video numbers across rows and sampling days indicated consistent acquisition after installation, apart from Day 1 Row A during 09:00–11:00.The robot operated without human intervention during the daily field monitoring windows.
- Detection validation: 84.6% recall and 89.2% precision were achieved for honeybee detection across 12 validation videos.Manual observation identified 39 honeybees; Polytrack correctly detected 33, with 6 false negatives and 4 false positives.
- Detection validation: 20 versus 21 total flower visits were recorded by manual and automated observations, respectively.Differences occurred across individual samples because dense inflorescences complicated flower detection and wind-driven movement could split one visit into multiple interactions.
- Spatiotemporal activity: 1,249 video sampling points yielded 1,971 honeybee tracks and 854 flower visits for fine-scale activity analysis.The processed recordings supported spatial and temporal characterisation of pollinator activity along the monitored rows.
- Microclimatic drivers: 22–23°C was associated with peak honeybee activity, while the highest insect counts generally coincided with intermediate humidity.Temperature ranged from 17°C to 30°C and relative humidity from 25% to 60% during monitoring.
4. Discussion
AGRICAM provided fine-resolution, automated pollination monitoring across commercial blueberry polytunnels while integrating with farm operations. Results showed broadly uniform spatial activity, clear temporal variation, and important prototype limitations in power, motion robustness, and flower detection.
- AGRICAM design and implementation: AGRICAM operated with minimal human intervention without disrupting routine farm activities, using an overhead track that is cheap, simple, and easy to retrofit.The overhead design avoids ground contact, reducing structural and soil-related impacts within polytunnels.
- AGRICAM design and implementation: Manual battery recharging limits the AGRICAM prototype, with endurance affected by tunnel length, slope, monitoring duration, sampling frequency, and battery mass.The authors identify autonomous charging as necessary for commercial-scale deployment.
- Automated video processing: Three-dimensional blueberry canopies challenge tracking because occlusion fragments trajectories, strong winds cause camera motion, and clustered flowers reduce bounding-box detection accuracy.The paper identifies occlusion handling, motion robustness, and accurate flower segmentation as priorities for future development.
- Pollination analysis: 2.25–2.5 m spatial intervals enabled high-resolution quantification of insect tracks and flower visits across blueberry polytunnels.The authors report that this detail would be impractical to obtain through manual transect walks or quadrat-based observations.
- Pollination analysis: Both insect tracks and flower visits were relatively uniform along the monitored polytunnels, while extended seasonal monitoring remains necessary to investigate potentially variable distributions.The authors attribute differences from previous work to possible site-specific and methodological differences.
- Pollination analysis: Honeybee tracks peaked between 11:00 and 13:00 before declining through the afternoon, and flower visits followed similar spatial and temporal patterns.Flower visits were consistently lower than honeybee tracks because some interactions were not clearly visible or were obscured by foliage.
5. Conclusion
AGRICAM is a scalable robotic platform for automated, high-resolution pollination monitoring under commercial conditions. Its data support improved pollination management and informed decision-making at commercial scale.
- AGRICAM integrates track-based mobility, vision sensing, RFID localisation, microclimate sensors, 4G connectivity, and a web interface for automated pollination monitoring.Its data are analysed through a computer vision and AI pipeline for spatio-temporal analysis of pollinator activity.
- 80 m industrial polytunnels were monitored over multiple days, capturing spatial and temporal variations in insect activity and relationships with microclimatic conditions.The system required minimal human intervention and did not disrupt routine farm activities.
- AGRICAM provides a scalable and practical solution for automated pollination monitoring under commercial conditions.
- The approach supports precision agriculture through improved pollination management and informed decision-making, contributing to improved food security at commercial scale.
CRediT authorship contribution statement
The CRediT statement assigns authorship across conceptualisation, data, analysis, funding, investigation, methodology, administration, resources, software, supervision, validation, and writing activities.
- The authors contributed across the study’s conceptual, methodological, analytical, operational, software, validation, and writing activities.
Declaration of competing interest
The authors disclose external financial support for several contributors and report no known competing financial interests or relevant personal relationships.
- Alan Dorin, Adel N. Toosi, James Cook, and Romina Rader received Australian Research Council support, while Rader also received University of New England support.
- The other authors declare no known competing financial interests or personal relationships that could have influenced the reported work.