In Field Detection of Downy Mildew Symptoms with Proximal Colour Imaging.

Area of Science:

• Agricultural Science
• Plant Pathology
• Computer Vision

Background:

• Grapevine downy mildew is a significant disease impacting viticulture.
• Accurate and early detection of foliar symptoms is crucial for disease management.
• Current detection methods may lack precision and scalability for in-field applications.

Purpose of the Study:

• To investigate the potential of on-board color imaging for grapevine downy mildew detection.
• To develop an algorithmic strategy for identifying various foliar symptom forms.
• To accurately estimate the area of affected grapevine tissues.

Main Methods:

• Utilized structure-color representations and probabilistic models of grapevine tissues.
• Developed image descriptors combining Local Structure Tensors (LST) with colorimetric statistics.
• Employed Log-Euclidean space mapping for descriptor modeling with Gaussian distributions.
• Implemented a seed growth segmentation process for pixel-wise classification.

Main Results:

• Achieved reliable detection of downy mildew symptoms and estimation of affected tissue area.
• Demonstrated high accuracy with an average of 83% pixel-wise precision.
• Reported an average pixel-wise recall of 76% in cross-validation.
• Validated the method on a dataset of 100 images with downy mildew and abiotic stresses.

Conclusions:

• The proposed on-board color imaging strategy effectively detects grapevine downy mildew.
• The algorithmic approach provides accurate symptom identification and area estimation.
• This technology holds promise for improved in-field disease monitoring in viticulture.