Development of a high-performance platform based on computational vision for grape leaf rust phenotyping

Abstract

Fungal diseases are one of the major causes of economic losses in agriculture. Here we present a high-throughput system for automatic phenotyping of foliar discs inoculated with Asian Leaf Rust (ALR), caused by Neophysopella tropicalis (Ono). Foliar discs with 10 mm diameter were inoculated with ALR and photographed by an automatic capture image system, called “BlackBird”. The images were manually phenotyped using an image editor software and used to train a Mask R-CNN model on framework Detectron-2 to collect leaf rust severity, number of pustules, and pustule mean area. After training, we manually phenotyped a subset of images and compared these data with our trained Mask R-CNN model. Convolutional Neural Networks (CNNs) were also trained and used to classify these images. The mean difference between real and automatic severity, calculated by Mask R-CNN and CNN, was 0,37% and 19,30%, demonstrating that the Mask R-CNN model has great accuracy for image classification. To our knowledge, this is the first high-performance platform for phenotyping grapevine leaf rust. This methodology can also be used to phenotype rust from other hosts and be adapted to other pathogens.

References

BIERMAN, A. et al. Ahigh-throughput phenotyping system using machine vision to quantify severity of grapevine powdery mildew. Plant Phenomics, 2019.

CADLE-DAVIDSON, L. et al. Lessons from a phenotyping center revealed by the genome-guided mapping of powdery mildew resistance loci. Phytopathology, v. 106, n. 10, p. 1159-1169, 2016.

GIRSHICK, R. Fast R-CNN. In: Proceedings of the IEEE International Conference on Computer Vision, 2015, p. 1440-1448.

GOMES, B. R. et al. Assessment of grapevine germoplasm collection for resistance to grape leaf rust (Phakopsora euvitis) using a leaf disc assay. Euphytica, v. 215, p. 1-11, 2019.

HE, K. et al. Deep residual learning for image recognition. In: Proceedings of the IEEE conference on computer vision and pattern recognition. 2016. p. 770-778.

MALAGOL, N. et al. Ahigh-throughput ResNet CNN approach for automated grapevine leaf hair quantification. Scientific Reports, v. 15, n. 1, p. 1590, 2025.

ONO, Y. Taxonomy of the Phakopsora ampelopsidis species complex on vitaceous hosts in Asia including a new species, P. euvitis. Mycologia, v. 92, n. 1, p. 154-173, 2000.

PRIMIANO, I. V. et al. Asian grapevine leaf rust caused by Phakopsora euvitis: an important disease in Brazil. Plant Pathology, v. 66, n. 5, p. 691-701, 2017.

SANTOS, R. F.; PRIMIANO, I. V.; AMORIM, L. Identification and pathogenicity of Neophysopella species associated with Asian grapevine leaf rust in Brazil. Plant Pathology, v. 70, n. 1, p. 74-86, 2021.

Acknowledgements

We are grateful to the Coordination for the Improvement of Higher Education Personnel (CAPES) and CNPq for scholarships awarded to A.C.M. and R.S. This research was financially supported by the National Council for Scientific and Technological Development (CNPq), Brazil (Grant: CNPq/409471/2021-6).