Comparison of grape flower counting using patch-based instance segmentation and density-based estimation with Convolutional Neural Networks

Information on flower number per grapevine inflorescence is critical for grapevine genetic improvement, early yield estimation and vineyard management. Previous approaches to automize this process by traditional image processing techniques, are unsuccessful in the improvement of a universal system that can be applied to inflorescences with diverse morphology. In this paper, we validated the efficiency of advanced deep learning-based approaches for automatic flower number estimation. The results were analyzed on an inflorescence dataset of 204 images from four different cultivars during various growth stages. The algorithm developed on patch-based instance segmentation using Mask R-CNN, produced counting results highly correlated to manual counts (R-2 = 0.96). Practically constant MAPE values among different cultivars (from 5.50% to 8.45%), implying a high robustness in this method. Achieving the fastest counting (0.33 second per image of size 512 x 512) with slightly lower counting accuracy (R-2 = 0.91), the method based on object density-map using U-Net turned out to be suitable for real-time flower counting systems.