L. Zabawa: Counting grapevine berries in images via semantic segmentation

Laura Zabawa is a Ph.D. candidate and research assistant at the Institute of Geodesy and Geoinformation at the University of Bonn since 2018. In the scope of the project NoViSys (https://www.zukunft-weinbau.de/), her research focuses on high throughput, image-based phenotyping in viticulture, using images which were recorded in the field.

Phenotyping in viticulture is restricted to on-site analysis due to the perennial nature of grapevine. This leads to time and labour intensive procedures, which are traditionally performed by skilled experts and extrapolated from small samples to plots. Thereby different grapevine varieties and training systems, e.g. vertical shoot positioning (VSP) and semi minimal pruned hedges (SMPH) pose different challenges.

We propose an image-based deep learning approach, which works on data directly collected in the field with a phenotyping platform. A convolutional neural network detects single berries in images by performing a semantic segmentation and counting each berry with a connected component algorithm. We compare our results with the Mask- RCNN, a state-of-the-art network for instance segmentation and with a regression approach for counting. The accuracy for the detection of green berries in canopy with our approach is 94.0% in the VSP and 85.6% in the SMPH.