Developing a microscope image dataset for fungal spore classification in grapevine using deep learning

Abstract

Grapevine trunk diseases (GTD) result from various fungi invading grapevine wood, leading to a decline in quality and yield. Accurate identification of fungal species is vital for effective disease management. Visual inspection through microscopy is a commonly used method, but distinguishing similar microorganisms within the same genus can be challenging. For precise identification, molecular methods are often required, despite being relatively costly and time-consuming. In this paper, we present a novel method for classifying four species of grapevine wood fungi using deep learning algorithms. We evaluate the performance of four different deep learning architectures, ResNet-50, VGG-16, MobileNet, and InceptionV3, in the classification of grapevine fungal spores from our microscope image dataset. During our tests, the proposed classification methodology achieved an accuracy of up to 97.40 %. Our approach can facilitate the development of more efficient and accurate methods for fungal species identification and has potential applications in viticulture and plant pathology research.