AI-powered image analysis for early fungal disease detection in cut rose crops

Overview

This solution leverages RGB and thermal image analysis combined with supervised machine learning algorithms to detect pre-symptomatic fungal diseases in protected cut rose crops. Early detection enables growers to intervene before visible symptoms spread, reducing crop losses and optimizing fungicide use. The approach targets powdery mildew and downy mildew, two of the most economically damaging diseases in commercial rose production.

The technology is being developed for deployment in commercial greenhouse environments, where continuous canopy monitoring can support integrated pest management programs. By identifying infections before they become visible to the naked eye, growers can reduce yield losses, minimize chemical inputs, and improve overall crop quality.

Technical specifications

Detection methodology:

• RGB image analysis processes canopy images to identify subtle color and texture variations associated with fungal infection
• Thermal imaging captures canopy temperature differences that indicate physiological stress caused by disease
• Environmental sensors (temperature and relative humidity) mounted in a grid pattern provide contextual data correlated with disease pressure

Image processing pipeline:

• Captured images are noise filtered and normalized
• RGB images are transformed into HSV color space for enhanced feature extraction
• Regions of interest (ROIs) are classified into three categories: areas without visible lesions, areas with visible lesions, and adjacent areas to visible lesions
• Supervised learning algorithms classify canopy images into healthy, powdery mildew-infected, or downy mildew-infected categories

Validation status:

• Laboratory validation confirmed that pixel intensity differs significantly between healthy and infected leaf areas (P<0.0001)
• Post hoc analysis confirmed that lesion areas exhibit significantly higher pixel intensity than non-lesion and adjacent areas (P<0.05)
• Field validation in commercial crops is the next step to confirm performance under real-world conditions

Technology readiness level

The technology is currently at an early-to-mid stage of development. Laboratory validation has demonstrated proof of concept for RGB-based detection of powdery mildew in rose leaflets sampled from commercial crops in Colombia. A 10-month field validation study is planned to evaluate combined RGB and thermographic imaging under commercial greenhouse conditions in the Bogotá Plateau region. This next phase will validate the approach in real-world production environments and refine the supervised learning models for operational deployment. The research team is actively seeking partnerships to support equipment acquisition and funding for key personnel during this validation stage.