The Netherlands holds a prominent position as the leading global producer and exporter of flower bulbs, subject to stringent phytosanitary requirements. Among these, a crucial demand involves the absence of viruses in bulb batches, with Tulip Breaking Virus (TBV) and Tulip Virus X (TVX) posing significant challenges to the tulip industry. The research spans four consecutive years, employing standard RGB and multispectral cameras to explore their potential in automated virus detection in the field. The initial year involves pre-research, testing previous data with current deep learning software. Subsequent years focus on expanding datasets representative of real-world scenarios, utilizing field setups. The newly curated dataset is used to train a model with the goal of maximizing detection accuracy. Laboratory testing with spectral cameras assesses their capability to identify TVX-infected tulips without visible symptoms. Results indicate the feasibility of using deep-learning models and cameras to identify TBV and TVX in tulips, with 40% accuracy in detecting annotated symptoms using colour cameras. In a laboratory setting, detection accuracy reaches 86%, though distinguishing infected tulips without visible symptoms remains challenging. The spectral camera in the field, demonstrates an 86% accuracy, highlighting the potential of additional spectral bands for TVX and TBV recognition. The hand-held spectrometer achieves a prediction accuracy of 93%, showing promise in virus detection.
Original languageDutch
Place of PublicationWageningen
PublisherWageningen Plant Research
Number of pages33
Publication statusPublished - 2023

Publication series

NameRapport / Stichting Wageningen Research, Wageningen Plant Research, Business unit Glastuinbouw

Cite this