Detection of citrus black spot fungi Phyllosticta citricarpa & Phyllosticta capitalensis on UV-C fluorescence images using YOLOv8

Autor: Pappu Kumar Yadav, Thomas Burks, Jianwei Qin, Moon Kim, Megan M. Dewdney, Fartash Vasefi
Jazyk: angličtina
Rok vydání: 2024
Předmět:
Zdroj: Smart Agricultural Technology, Vol 9, Iss , Pp 100615- (2024)
Druh dokumentu: article
ISSN: 2772-3755
DOI: 10.1016/j.atech.2024.100615
Popis: Citrus Black Spot (CBS), caused by the pathogenic fungus Phyllosticta citricarpa, is a quarantine citrus disease, that has the potential to spread on nursery tress especially in regions like Florida and beyond. The early detection of this disease assumes paramount importance, especially during the asymptomatic phase of tree infection. This critical stage offers a window of opportunity for grove managers to implement preemptive measures, thereby mitigating the potential dissemination of the infection within orchards. In the present study, we elucidate the robust capabilities of the Contamination Sanitization Inspection-Disinfection Plus (CSI-D+) system, which integrates state-of-the-art fluorescence imaging technology, in tandem with the YOLOv8 deep learning framework. Our investigation is centered on the direct detection of conidia of the CBS-causing fungus P. citricarpa (Gc12) and its non-pathogenic counterpart P. capitalensis (Gm33), both prevalent on surfaces of infected citrus leaves across varying concentration gradients. Impressively, the CSI-D+ system (which is a new, handheld fluorescence-based imaging device developed to detect microbial contamination and disinfect surfaces rapidly) exhibits remarkable discriminatory acumen, achieving a noteworthy mean classification accuracy of 96.97 % for Gc12 fungus classification. This precision is complemented by an impressive F1-score of 96.35 %, coupled with a commendable mAP@50 score of 97.82 %. Furthermore, our inquiry extends to encompass the Gm33 variant, wherein the system maintains a commendable average classification accuracy of 96.17 %, alongside an F1-score of 88.76 %, and a mAP@50 of 91.64 %. Such pioneering systems bear substantial promise, serving as a rapid, non-invasive instrument for the early identification of incipient CBS infestations within citrus arboreal landscapes. In equipping grove managers with timely insights, these advancements stand to empower effective and timely intervention strategies, fortifying orchard resilience against the progression of this pathogenic menace.
Databáze: Directory of Open Access Journals