Root Guard AI: A Computer Vision Prototype for Early Detection of Root Disease in Hydroponic Systems

1. Abstract

Root diseases in hydroponic systems can spread rapidly through recirculating water and may remain unnoticed until visible plant decline occurs. This proposal aims to develop and validate Root Guard AI, a low-cost computer vision prototype for early monitoring of root health in Deep Water Culture hydroponics. The system will periodically capture root images using a waterproof camera and controlled illumination, then estimate root health using visual indicators such as browning, color change, slimy texture, density loss, and structural degradation. Our student team includes members with backgrounds in computational science, data science, engineering, and life sciences. The first pilot will use plants only, without human or animal participants, and will initially induce non-pathogenic root stress by temporarily reducing aeration. The estimated timeline is 6–8 weeks: prototype assembly, image collection, annotation, model testing, and expert/customer feedback. We will share the final protocol, anonymized plant image dataset, metadata, analysis code, and negative results through an open repository after review.

2. Introduction

Hydroponics is a soilless cultivation method where plants grow in nutrient-rich water under controlled environmental conditions. It is widely used in greenhouses, vertical farms, and urban farming because it allows year-round production, efficient space use, and controlled crop quality.

Root health is especially important in hydroponic systems because the roots are directly exposed to the nutrient solution. In recirculating systems, root disease can spread quickly through shared water and may remain undetected until visible plant decline occurs. Previous reports have shown that root rot in hydroponic crops can cause serious production losses, including severe damage in commercial lettuce systems [1].

Current monitoring approaches often focus on leaf symptoms, nutrient solution parameters, or general plant appearance. However, these methods may not directly detect early changes in the root zone. Biological control agents and beneficial bacteria are also used in plant systems, but plant-associated bacteria can have complex roles, including both beneficial and harmful effects depending on the organism, host, and conditions [2,3].

Computer vision has already been used in plant phenotyping and root image analysis. For example, image-based methods have been developed for automated root disease severity assessment, and root segmentation techniques have been used for high-throughput plant phenotyping [4,5]. These studies suggest that visual root features can provide useful information for plant health assessment.

Based on this background, our rationale is that hydroponic systems provide a unique opportunity for direct root monitoring because the roots are physically visible. Therefore, a low-cost camera-based system may help detect early visual changes such as browning, texture degradation, density loss, and structural damage before severe plant decline occurs.

Hypotheses

H1 — Primary Outcome

A controlled root imaging system can detect measurable visual differences between healthy and stressed hydroponic roots using color, texture, density, and structural features.

Primary outcome: change in computer vision root health score between healthy/control roots and stressed/damaged roots.

H2 — Secondary Outcome

Combining computer vision with basic environmental metadata such as pH, temperature, and aeration status will improve interpretation of root health compared with image-only assessment.

Secondary outcome: improvement in classification accuracy, precision, recall, or F1 score when optional sensor/condition metadata are included.

H3 — Secondary Outcome

Hydroponic growers, controlled-environment agriculture experts, and plant researchers will consider root-level monitoring useful if the system is low-cost, non-destructive, and easy to install.

Secondary outcome: qualitative feedback from expert interviews and customer development responses.

3. Methods

Participants/Sample

This study does not involve human or animal participants. The experimental sample will consist of hydroponically grown plants, preferably lettuce or fast-growing leafy greens, because they are common in controlled hydroponic systems and develop visible root systems quickly.

The pilot study will use approximately 24–40 plants depending on available space and funding. Plants will be divided into:

healthy control group
induced root-stress group
optional recovery/monitoring group if enough plants are available

The primary experimental unit will be the plant, not the number of images, because repeated images from the same plant are not statistically independent.

Inclusion Criteria

Plants will be included if they:

are grown in the same hydroponic system type
have visible roots suitable for imaging
are grown under similar light, nutrient, and temperature conditions
survive until the baseline imaging stage

Exclusion Criteria

Plants will be excluded if they:

fail to germinate or establish roots
are physically damaged during handling
have poor image quality due to camera failure or obstruction
experience unrelated severe environmental stress

Recruitment Strategy

No human recruitment is required for the plant experiment. For customer development, feedback will be requested from hydroponic growers, greenhouse operators, agritech companies, plant pathology researchers, controlled-environment agriculture researchers, and computer vision researchers. These responses will be used only as qualitative feedback unless formal interview ethics approval becomes necessary.

Sample Size Justification

This is a pilot feasibility study, not a final agronomic validation trial. A sample of 24–40 plants is sufficient for initial prototype testing, image collection, method debugging, and estimating the direction and size of root health score differences. Since the expected visual differences between healthy and stressed roots may be large, the pilot will focus on feasibility, image quality, and model behavior rather than making final production-level performance claims. Results will be used to estimate effect size and design a larger powered study later.

Materials and Procedures

The prototype will be built around a small Deep Water Culture hydroponic system. The system will include:

reservoir/tank
net pots and plants
air pump and air stone
LED grow light
waterproof camera or root imaging module
controlled short-duration illumination
optional pH and temperature sensors
optional EC or dissolved oxygen monitoring if available

The system will periodically capture images of the root zone under controlled lighting. Images will be collected at baseline while plants are healthy. After baseline imaging, mild root stress will be induced in the experimental group by temporarily reducing aeration for a controlled period. This non-pathogenic stress method is safer and simpler than introducing live pathogens at the pilot stage.

Images will be labeled into categories such as:

healthy roots
early visual stress
browning/color change
severe stress/damage
unclear/low-quality image

Visual features of interest will include:

root whiteness or browning
slimy or degraded texture
root density
root branching/structure
contrast between roots and background
temporal change compared with previous images

A simple model will first be developed using image preprocessing and classical computer vision features. If the dataset size is sufficient, a lightweight machine learning or deep learning model may be tested. The first goal is not to build a perfect commercial AI model, but to test whether root-level visual monitoring gives meaningful early information.

Outline analysis pipeline, preprocessing steps, statistical tests, and hypothesis-specific predictions. Describe how results will be interpreted regardless of outcome. Describe how deviations from the planned sample will be handled.

Planned Analysis

The analysis will follow this pipeline:

Image quality control
Cropping or segmentation of root region
Color normalization under controlled lighting
Extraction of color, texture, and density features
Annotation of root condition by team members and, if possible, plant/agro experts
Model training or rule-based scoring
Comparison of healthy vs stressed root images
Evaluation using accuracy, precision, recall, F1 score, and confusion matrix
Qualitative interpretation of false positives and false negatives

For H1, we expect stressed roots to show measurable changes in root color, texture, density, or structure compared with healthy roots. If no measurable difference is detected, this will indicate either that the stress method is not biologically suitable, the camera setup is insufficient, or visual features alone are not enough.

For H2, we expect optional sensor or condition metadata to improve interpretation. If metadata do not improve performance, this will support a simpler image-only prototype.

For H3, feedback will be analyzed qualitatively. Responses will be grouped into themes such as perceived problem importance, willingness to test, required accuracy, installation barriers, cost sensitivity, and best target niche.

If fewer plants are available than planned, the study will be reported honestly as a smaller feasibility pilot. No plant-level conclusions will be overstated. Repeated images from the same plant will be treated as repeated observations, not independent biological replicates.

Ethics and Data Management

The initial pilot does not involve human subjects, animals, or intentional pathogen inoculation. Therefore, formal human-subject ethics approval is not expected to be required for the plant-only prototype test. If future work includes structured interviews, personal data collection, or pathogen introduction, appropriate institutional ethics or biosafety review will be requested before starting.

No confidential farm data will be published. Plant images, imaging conditions, basic metadata, annotation guidelines, analysis scripts, and negative results will be shared through an open repository such as GitHub or Zenodo after internal review. Data will be released with clear documentation so that other researchers can reproduce or critique the method.

4. Pilot Data

We do not yet have final pilot data. Current feasibility is based on the prototype plan, literature support for image-based root assessment, and early expert feedback.

Our planned prototype includes a DWC system with a waterproof camera, illumination, aeration, and optional sensors. We have also received early validation feedback suggesting that the idea is feasible, but that the concept may need several iterations based on model and setup performance. The first practical recommendation was to grow plants, take root photos, collect even a small dataset, and verify interpretations with someone from an agro/life-science background.

The first pilot will therefore focus on:

confirming that roots can be imaged clearly
identifying stable lighting and camera position
collecting a small root image dataset
testing whether healthy and stressed roots are visually separable
obtaining expert feedback before claiming commercial readiness

5. Budget

Estimated pilot budget: USD 500–700

Item	Purpose	Estimated Cost
DWC reservoir/tank, net pots, tubing	Basic hydroponic prototype structure	USD 60
Seeds/seedlings, growing medium, nutrients	Plant growth and experimental sample	USD 50
Air pump, air stone, backup tubing	Aeration and root-zone oxygen control	USD 40
LED grow light and timer	Controlled plant growth conditions	USD 70
Waterproof camera / endoscope camera	Root image capture	USD 80
Controlled lighting module / LED strip / flashlight	Stable illumination during root imaging	USD 40
pH and temperature sensors	Basic environmental monitoring	USD 70
Optional EC or dissolved oxygen sensor	Additional nutrient/oxygen monitoring	USD 80
Cleaning materials, gloves, labels, consumables	Safe handling and experiment organization	USD 40
Data storage / annotation support / miscellaneous	Image storage, labeling, and analysis support	USD 40
Contingency	Unexpected replacement or extra materials	USD 70
Total Estimated Budget		USD 640

The minimum useful budget is approximately USD 500 if optional sensors are removed. A stronger version of the pilot would require USD 900–1,200 to add more replicates, better imaging hardware, and improved environmental monitoring.

6. References

[1] Pierz, L. D., Heslinga, D. R., Buell, C. R. & Haus, M. J. An image-based technique for automated root disease severity assessment using PlantCV. Appl. Plant Sci. 11, e11507 (2023). doi: 10.1002/aps3.11507.

[2] Raaijmakers, J. M. & Mazzola, M. Diversity and natural functions of antibiotics produced by beneficial and plant pathogenic bacteria. Annu. Rev. Phytopathol. 50, 403–424 (2012). doi: 10.1146/annurev-phyto-081211-172908.

[3] Seidenthal, K., Panjvani, K., Chandnani, R., Kochian, L. & Eramian, M. Iterative image segmentation of plant roots for high-throughput phenotyping. Sci. Rep. 12, 16563 (2022). doi: 10.1038/s41598-022-19754-9.

[4] Stanghellini, M. E., Kim, D. H., Rakocy, J., Gloger, K. & Klinton, H. First report of root rot of hydroponically grown lettuce caused by Pythium myriotylum in a commercial production facility. Plant Dis. 82, 831 (1998). doi: 10.1094/PDIS.1998.82.7.831D.

[5] Thomas, B. O., Lechner, S. L., Ross, H. C., Joris, B. R., Glick, B. R. & Stegelmeier, A. A. Friends and foes: Bacteria of the hydroponic plant microbiome. Plants 13, 3069 (2024). doi: 10.3390/plants13213069.

Root Guard AI: A Computer Vision Prototype for Early Detection of Root Disease in Hydroponic Systems

Root Guard AI: A Computer Vision Prototype for Early Detection of Root Disease in Hydroponic Systems

Funding Progress

Nonprofit Recipient

Topics

DOI