Applications of machine vision for grapevine phenotyping

Abstract

Computer vision systems are powerful tools to assess canopy status, fruit grading, and yield in agriculture. This work shows the application of machine vision techniques for grapevine phenotyping under laboratory and field conditions. The first approach aimed at determining automatically the number of flowers per inflorescence using a low cost camera under field conditions. Strong and significant relationships, with R² above 0.8 in Tempranillo, Graciano and Carignan cultivars were observed between actual and automated estimation of inflorescence flower numbers. The second approach aimed at analysing cluster morphology using image analysis under laboratory conditions. Berry weight, berry number per cluster and cluster weight were estimated using several algorithms of image processing. The best results (R² between 0.69 and 0.95 in berry detection and between 0.65 and 0.97 in cluster weight estimation) were achieved using the Canny algorithm. The model’s capability based on image analysis to predict berry weight was R²=0.84. The third approach aimed at working outdoors, developing in-field systems capable of assessing the leaf area and yield of the vineyard by acquisition and processing of digital images. Our algorithms allowed the total leaf area (R²=0.81; p<0.001) and yield (R²=0.73; p=0.002) per vine to be estimated. In conclusion, computer vision techniques can be applied in viticulture for plant phenotyping and vineyard monitoring. A large set of samples can be automatically measured, saving time and providing more objective and precise information.