Articles
Transitioning from laboratory research to real-world application: encouraging results in the practical utilization of hyperspectral images for field detection of grapevine viruses
Article number
1395_27
Pages
201 – 208
Language
English
Abstract
Vineyards are affected by grapevine viruses, resulting in negative consequences such as hindered fruit ripening, diminished grape quality, and reduced crop yield.
It is crucial to identify infected vines in order to prevent the spread of these viruses.
Remote sensing offers an ideal solution by measuring the physical and biological properties of vegetation that are influenced by the disease.
In this research, we assessed the potential of hyperspectral VIS/NIR imagery in detecting red-blotch and leafroll infected vines.
To accomplish this, we collected leaf samples in 2020, 2021 and 2022, utilizing PCR analysis to identify infections.
Hyperspectral images of leaves infected with grapevine leafroll and red blotch diseases were captured.
Simultaneously, but only for red-blotch infected vines, images of the vine canopy sides were captured using a tripod, as well as aerial images above the canopy using a drone.
To classify the infection in leaf images in the laboratory, convolutional neural network (CNN) and random forest (RF) models were compared.
To classify the vine images in the field, partial least square discriminant analysis (PLS-DA), RF, and support vector machine (SVM) were utilized.
Several ways to simplify the predictors number were tested using spectral binning and a recursive feature elimination (RFE). On leaf images, when binarily classifying infected vs. non-infected leaves, the CNN model reaches an overall maximum accuracy of 87.0%. On field images, the best overall accuracy reached 73.3% using only 23 bands of 8 nm width.
In the late season, with visible symptoms, the accuracy increased to 76.6% with 18 bands with 16 nm width and first preliminary analysis using drone images shows an accuracy of 86.8%. This method enables the detection of virus transmission and identification of potentially infected vines autonomously.
It is crucial to identify infected vines in order to prevent the spread of these viruses.
Remote sensing offers an ideal solution by measuring the physical and biological properties of vegetation that are influenced by the disease.
In this research, we assessed the potential of hyperspectral VIS/NIR imagery in detecting red-blotch and leafroll infected vines.
To accomplish this, we collected leaf samples in 2020, 2021 and 2022, utilizing PCR analysis to identify infections.
Hyperspectral images of leaves infected with grapevine leafroll and red blotch diseases were captured.
Simultaneously, but only for red-blotch infected vines, images of the vine canopy sides were captured using a tripod, as well as aerial images above the canopy using a drone.
To classify the infection in leaf images in the laboratory, convolutional neural network (CNN) and random forest (RF) models were compared.
To classify the vine images in the field, partial least square discriminant analysis (PLS-DA), RF, and support vector machine (SVM) were utilized.
Several ways to simplify the predictors number were tested using spectral binning and a recursive feature elimination (RFE). On leaf images, when binarily classifying infected vs. non-infected leaves, the CNN model reaches an overall maximum accuracy of 87.0%. On field images, the best overall accuracy reached 73.3% using only 23 bands of 8 nm width.
In the late season, with visible symptoms, the accuracy increased to 76.6% with 18 bands with 16 nm width and first preliminary analysis using drone images shows an accuracy of 86.8%. This method enables the detection of virus transmission and identification of potentially infected vines autonomously.
Authors
E. Laroche-Pinel, E. Sawyer, B. Corrales, K. Singh, K. Vasquez, M.L. Cooper, M. Fuchs, L. Brillante
Keywords
disease detection, hyperspectral images, machine learning model, grapevine
Groups involved
- Division Temperate Tree Nuts
- Division Vine and Berry Fruits
- Division Temperate Tree Fruits
- Division Precision Horticulture and Engineering
- Division Plant-Environment Interactions in Field Systems
- Division Tropical and Subtropical Fruit and Nuts
- Working Group Precision Management of Orchards and Vineyards
Online Articles (59)