Machine vision learning for crop disease and quality parameters

Machine vision learning for crop disease and quality parameters

Detection of many crop diseases can be difficult due to a lack of visible symptoms (Figure 1). However, reflectance imaging has been useful in stress discrimination in a laboratory setting. The ultraviolet, visible, near infrared, shortwave infrared and thermal portions of the electromagnetic spectrum have been reported to be related to plant health and physiology. Reflectance imaging has been used to discriminate disease (Mewes, 2011; Thomas, 2017) and to differentiate diseases within single crops (Mahlein et al. 2010, 2013). However, many sensors used in crop-disease
research are developed in controlled, laboratory conditions: there is a requirement for sensors to be developed for commercial, field conditions.

Novel disease detection research is being conducted using sensors in visible and non-visible wavebands, to reduce the amount of sensor data required for disease discrimination. Machine learning techniques, which infer patterns in data without explicit instruction, are being implemented to reduce data requirements and potentially allow for commercially available and lower cost sensors for disease discrimination. The multiresolution capability of these sensing and processing approaches allows for selection of disease detection models potentially compatible with ground- and drone-based camera systems.

Files associated with this item cannot be displayed due to copyright restrictions.