Image based Crop Monitoring Technologies in Protected Horticulture: A Review

Namal Jayasuriya, Yi Guo, Wen Hu, Oula GhannoumarXiv 2024 · DOI · arXiv

Introduction

As the global population is projected to reach 9.3 billion by 2050, the agricultural sector faces the daunting task of increasing food production while navigating the constraints of climate change and diminishing arable land. Protected cropping—encompassing greenhouses, glasshouses, and vertical farms—has emerged as a vital solution for year-round food security. However, these facilities are notoriously labor-intensive. While high-throughput phenotyping (HTP) has flourished in research settings, a significant technological gap remains when transitioning these systems to large-scale commercial operations.

This review explores the current state of image-based plant phenotyping, categorizing sensors and delivery platforms while identifying the specific challenges of monitoring tall, vertically supported crops like tomatoes and peppers in commercial environments.

The Scalability Gap: Category 1 vs. Category 2

A primary contribution of this research is the distinction between two categories of plants based on their physical architecture and mobility. Most existing HTP systems were designed for Category 1 plants: small, movable model organisms (e.g., Arabidopsis) or young seedlings grown in pots. These systems typically utilize a "plant-to-sensor" approach, where conveyor belts transport plants to a stationary imaging chamber.

In contrast, commercial production is dominated by Category 2 crops: tall, immobile, and vertically supported plants. A 3-meter-tall tomato vine cannot be placed on a conveyor belt. This creates a "scalability gap," where the sophisticated monitoring tools used in labs fail to translate to the physical realities of a commercial greenhouse. Bridging this gap requires a shift toward "sensor-to-plant" systems, where mobile platforms navigate the crop rows.

Sensor Technologies and Deployment Platforms

The paper synthesizes current literature into a comprehensive framework for sensor selection and deployment.

Imaging Modalities

Different sensors are required to capture the diverse range of plant traits:

  • RGB Imaging: The most cost-effective method for monitoring morphology and growth rates.
  • 3D Scanning: Essential for volumetric traits and structural analysis.
  • Thermal Imaging: Used to detect water stress and stomatal conductance.
  • Fluorescence Imaging: Provides insights into photosynthetic efficiency and early stress detection.
  • Hyperspectral Imaging: Capable of identifying biochemical content, such as nitrogen levels and sugar concentration (Brix).
  • Tomography: While promising for non-destructive root analysis (MRI, PET, CT), these methods remain largely confined to labs due to their size and safety requirements.

Deployment Strategies

The authors introduce a selection framework that maps these sensors to specific facility types. For single-layer or multilayer vertical farms, gantry systems are often the most efficient. However, for the row-based architecture of commercial glasshouses, mobile robots are the only viable solution. Interestingly, the review notes that drones—while successful in broad-acre farming—are often impractical in greenhouses because their "prop wash" (downward wind) disturbs the canopy, preventing the stable image capture necessary for 3D reconstruction.

Economic Justification

The adoption of expensive robotic monitoring is driven by economic necessity. Analyzing the Australian Horticulture Statistics 2021/2022, the review highlights that glasshouse tomatoes represent only 26% of production volume but claim 65% of the total market value due to their superior quality. In Australia, protected cropping contributes approximately 15.85% of the total vegetable production value. These high-value crops justify the capital expenditure (CAPEX) required for automated monitoring systems that can ensure optimal growth and early pest detection.

Key Insights and Bottlenecks

  • The Lighting Challenge: In semi-controlled greenhouses, varying ambient light creates shadows and bright spots that disrupt standard image processing. The authors suggest that night-time imaging with artificial illumination or the use of mobile "dark chambers" are the most effective workarounds.
  • The Occlusion Problem: In dense commercial canopies, fruits and flowers are often hidden behind foliage. Future systems must move beyond simple "drive-by" imaging toward robotic arms that can "reach in" or multi-angle vision systems that can see through the canopy.
  • Algorithmic Robustness: Current computer vision models often struggle with foreground-background separation, especially under the specific light spectrums (e.g., pinkish LED hues) used in vertical farms. There is a pressing need for models trained specifically on these spectral environments.

Conclusion and Future Work

The transition from research labs to commercial greenhouses is the next frontier for image-based crop monitoring. While sensor technology is maturing, the industry must now focus on the physical challenges of the commercial environment: managing occlusion, developing portable root-scanning technologies, and creating robust algorithms that can handle the unique lighting of protected horticulture. Solving these issues will be critical for the sustainable intensification of global food production.

Summary updated

BibTeX
@misc{jayasuriya2024image,
  author = {Namal Jayasuriya and Yi Guo and Wen Hu and Oula Ghannoum},
  title = {{Image based Crop Monitoring Technologies in Protected Horticulture: A Review}},
  year = {2024},
  eprint = {2401.13928},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2401.13928},
}
Figure