Citrus Under Protective Screening (CUPS) System

• CUPS is an advanced horticultural system that employs engineered protective screens to mitigate environmental stress and optimize citrus production.
• It integrates sensor networks, robotics, and high-performance computing to enable real-time monitoring, precise imaging, and automated interventions.
• Digital twin simulations and multimodal data fusion drive robust disease detection and dynamic resource management for improved crop outcomes.

Citrus Under Protective Screening (CUPS) is an advanced horticultural system in which citrus crops are cultivated under engineered protective coverings to mitigate environmental stressors, reduce disease pressure, and enhance microclimate control. Contemporary CUPS facilities leverage a combination of high-density sensor networks, robotics, automated imaging, and high-performance computing (HPC) to achieve precision monitoring and intervention capabilities in both experimental and commercial citrus production systems.

1. Sensor Networks and Edge Processing in CUPS

Modern CUPS facilities are instrumented with a diverse array of sensors, such as weather stations recording wind, temperature, and humidity, as well as imaging and microclimate sensors distributed throughout the protected environment. Sensor nodes, often built on low-power edge devices (e.g., Raspberry Pi), run lightweight agents (such as the CSPOT runtime) to collect and pre-process environmental telemetry. Data acquisition is continuous, supporting granular monitoring of the protected citrus canopy and structural health of the screening itself (Kurafeeva et al., 24 Sep 2025).

A distinctive aspect of CUPS deployments is the transmission of the sensor data over private 5G networks. These networks employ network slicing mechanisms to provide low-latency (under 100 ms uplink) and high-throughput channels between field sensors and remote HPC facilities. Sensor messaging protocols utilize log-based architectures: each measurement is appended to a persistent log, ensuring reliable data transfer in the presence of intermittent connectivity, which is typical of remote agricultural deployments.

2. Integration of Sensing, Robotics, and Analytical Workflows

High-resolution image acquisition—from stationary or mobile platforms—forms a critical component of plant health assessment under CUPS. For instance, digital image processing techniques exploit k-means clustering algorithms operated in LAB color space to segment affected and healthy regions within citrus leaves, supporting both disease diagnosis (e.g., citrus canker) and quantitative planimetry. The calculation of affected leaf area relies on the binary segmentation output, with metrics such as the percentage of diseased pixels (e.g., 17.41% in affected leaves) providing an automated severity index (Kumar et al., 2023).

Robotic platforms, such as wheeled mobile agents, perform both routine surveys and targeted interventions. Robotics integration is further enhanced in CUPS by using private 5G for looped communication—sensor events (such as anomalies in microclimate or detected screen breaches) trigger dispatch commands to robots, which then traverse the protected space using pre-defined route plans. The robots generate additional imaging (stereo RGB, NIR, depth, thermal) and navigational data (LiDAR, IMU, GNSS-RTK), contributing to multimodal datasets for concurrent mapping, localization, and monitoring tasks (Teng et al., 2023).

3. High-Performance Computing and Digital Twin Systems

A distinguishing feature of next-generation CUPS is the direct coupling between edge sensor networks and centralized HPC infrastructures via 5G. The xGFabric architecture, for example, enables real-time integration of in-situ sensor data with computational fluid dynamics (CFD) models running on HPC systems. The system uses a layered software architecture, with CSPOT for log-based event messaging and Laminar for strongly-typed dataflow. Laminar orchestrates pipelines that detect statistically significant changes in telemetry by monitoring sliding windows (e.g., 30-minute intervals), which trigger real-time HPC batch allocations via a Pilot interface.

CFD simulations (implemented, for example, in OpenFOAM) operate as a digital twin, parameterized by field sensor conditions (wind, humidity, temperature). Comparisons between predicted and measured parameters support fault detection: deviations may indicate screen breaches or emergent microclimate anomalies, motivating immediate robotics deployment or changes to irrigation and disease management (Kurafeeva et al., 24 Sep 2025).

Table: Core System Integration Components in CUPS

Component	Function	Key Enabling Technology
Sensor Network	Environmental telemetry acquisition	CSPOT runtime, log-based messaging
Robotics	Survey, imaging, intervention	5G teleoperation, route planning
Data Analysis	Disease quantification, health assessment	K-means clustering, NDVI computation
HPC/CFD	Microclimate simulation, anomaly detection	OpenFOAM, digital twin, Laminar dataflow

The relationship between these components enables continuous, in-the-loop monitoring and intervention, integrating both empirical and simulated data under the CUPS paradigm.

4. Multimodal Sensing and Data Fusion

Multimodal datasets, such as those made available in the CitrusFarm corpus, include stereo RGB, depth, near-infrared, and thermal imaging, combined with navigational data from wheel odometry, LiDAR, IMU, and GNSS-RTK. Such datasets are foundational for crop monitoring, as they support computation of normalized difference vegetation index (NDVI): $$\text{NDVI} = \frac{\text{NIR} - \text{Red}}{\text{NIR} + \text{Red}}$$ which discriminates plant health and stress under variable light conditions imposed by protective screens (Teng et al., 2023). Thermal imaging, relatively insensitive to visual-band artifacts and lighting non-uniformity, identifies microclimate-driven stresses (e.g., water deficiency).

Data fusion is critical in CUPS because physical screening may degrade GPS, create occlusions, and induce lighting variations. Sensor fusion algorithms—such as Extended Kalman Filters (EKF) combining wheel odometry, IMU, LiDAR, and visual streams—provide robust localization and mapping. Adaptive pipelines weight sensor modalities based on signal confidence, dynamically compensating for the screening’s impact on data quality.

5. Automated Disease Detection and Quantitative Assessment

Digital planimetry for disease detection leverages both histogram comparison and unsupervised clustering to segment images of citrus leaves. The workflow involves the following steps (Kumar et al., 2023):

1. Image acquisition with controlled backgrounds to minimize reflectance artifacts.
2. Conversion to grayscale and binary forms for preliminary area estimation.
3. Color space mapping (RGB to LAB), followed by k-means clustering on a* and b* channels.
4. Segmentation into “unaffected” and “affected” regions at the pixel level.
5. Analysis of histograms and binary masks to compute affected area:
   • The disease severity metric is formalized as: Error (%) = (WP1 / TP) × 100, where WP1 and TP denote affected pixels and total pixels, respectively.
6. Resulting area quantifications directly inform the progression and intervention schedule for citrus canker and other foliar diseases.

A plausible implication is that integration of such image-based analysis with real-time crop monitoring enables earlier and more objective decision-making, especially when automated meters are deployed for continuous surveillance.

6. Dynamic Resource Management and Real-Time Interventions

Effective CUPS management relies on dynamic scaling of computational resources in response to observed field events. The HPC “Pilot” interface operates as a resource broker, using data-driven formulas to determine node allocation: $N_{\text{req}} = \max(1, D/\text{threshold})$ where $D$ is the incoming data volume. The interface launches pilots as needed, factoring in queue delays typical of batch-controlled systems (Kurafeeva et al., 24 Sep 2025).

Upon detection of irregularities (e.g., environmental outliers or simulated breaches), the architecture orchestrates real-time interventions. These may include:

• Dispatch of mobile robots to suspected damage sites, where on-board cameras provide visual confirmation before scheduling repairs.
• Automated actuation of irrigation or pesticide delivery units, guided by CFD simulation outcomes and sensor evidence of water stress or disease.

The orchestration of these interventions is mediated by Laminar dataflow, which abstracts from underlying synchronization and ensures high reliability via persistent logs and fault-tolerant runtime handling.

7. Challenges, Calibration, and Future Potential

The physical and optical effects introduced by protective screens—such as lighting inhomogeneity, reflections, and intermittent occlusions—necessitate algorithmic and operational adaptations. Image preprocessing (e.g., histogram equalization) and periodic re-calibration (potentially using the Kalibr toolbox for multi-camera and IMU calibration) are required to mitigate screening-induced measurement biases (Teng et al., 2023).

A plausible implication is that continuous accumulation of multimodal, calibrated datasets will enable the construction of predictive models for disease progression, energy efficiency, yield optimization, and dynamic screening maintenance. Furthermore, the CUPS paradigm, augmented with digital twin technology, holds potential for generalization to other high-value crops where controlled climate and disease management are essential.

In conclusion, Citrus Under Protective Screening represents a convergence of environmental control, real-time sensing, multimodal data fusion, high-performance simulation, and automated intervention, all coordinated via advanced network and software infrastructure. These integrated methodologies provide the foundation for robust, scalable, and precise citrus production in variable agroecological contexts.