UV-spectrum remote UVA imaging for use in precision agriculture

Abstract

Floral UV-reflectance is considered an essential factor in plant-pollinator interactions. UVreflective pigments on reproductive structures allow pollinators to locate flowers from the air and differentiate conspecifics. This relationship is possible due to the ability of pollinators, such as bees, to see in the UV spectrum (300-400nm). Despite this well-documented visual signalling strategy of flowering plants, our literary review indicated that few crop species have had their UV spectral reflectance documented. Without considering UV reflectance, breeding efforts could render flowers cryptic to pollinators, decreasing yield and harvest quality. Strawberry cultivars were spectrally analyzed and compared to their wild counterpart. White-flowering cultivars showed higher pollinator visibility, whereas the red-flowering cultivar was cryptic. Bee vision (300-650nm) is adapted to detect flowers. This project mimicked the bee vision range in designing and creating the Nature Inspired Detector (NID) to detect strawberry flowers remotely. Two state-of-the-art AI algorithms were trained on a custom strawberry flower image dataset where YOLOv5 outperformed Faster R-CNN( mAP 0.978 vs. 0.912, respectfully). The NID was then field deployed on a UAV over a strawberry field. Results were comparable to a contemporary study, but the NID had a faster training time (0.3 vs. 5.5 hrs) and higher mAP (0.951 vs. 0.772).

Date	27 Aug 2023
Original language	American English
Awarding Institution	École de technologie supérieure
Supervisor	David St-Onge (Supervisor) & Robert Hausler (Co-supervisor)