Viticulturists are one step closer to advanced computer vision
Vineyard owners can’t see their eyes anywhere, all the time. That is why land-based vision systems have become commercial scams.
Vision systems that quickly and accurately assess canopy size have long been around; Researchers at the University of Adelaide are developing vision systems that have more image recognition capabilities than ever before.
The researchers trained machine learning algorithms to identify important health and development indices from physical and thermal cameras-including plant area index (PAI), weight. of pruning, inflorescence and shoot number, number of buds, and leaf temperature of the plant.
This vision system will be available for growers soon as well as many libraries of images trained by algorithms.
VitiData is a library of more than 175,000 images of grapes at various stages of development. Both VitiData and the ground-based vision system itself became part of the larger VitiVisor project, funded by Wine Australia and Riverland Wine
By introducing VitiData to the open-source data-sharing platform Figshare, researchers are paying off their efforts to the larger viticulture and computer vision science community.
In the Information Age, data is our most valuable asset. VitiData is an extensive library of images collected from 14 different sites, with a total of 60 hectares of the five major grape varieties grown in Riverland.
Nearly 1,000 of the images were annotated by expert viticulturists on the ground-truth of the predictions made by the algorithms. It represents 420 hours of expert annotation, and results in a dataset of ‘training’ that will prove invaluable to companies looking to create or improve their ground vision capabilities.
Making datasets like VitiData publicly accessible and interpretable means they are more likely to be reused, providing greater benefits to the viticultural industry as a whole. In addition, it made access to the data of the very people who helped make it – Riverland grape growers.
To access the dataset, click here
Are you a Daily Wine News subscriber? If not, click here to join our mailing list. It’s free!