Automatic diagnosis of olive quick decline syndrome and grapevine yellows for the agriculture industry

The agriculture industry is facing a crisis and must consider AI to reduce operational costs. One such area is pest and disease scouting. Delays cause further infection and growers must act quickly if a disease is discovered. However, scouting often requires external consultants and lab tests, time consuming and costly. Deep learning has the potential to detect diseases automatically from leaf clipping images. We achieve good results on two diseases of interest to the industry: Olive Quick Decline Syndrome (OQDS) and Grapevine Yellows (GY). The current challenge is overcoming data requirements of deep learning, often addressed with crowd sourcing. However, it can be difficult to crowd-source data of control pathogens. We demonstrate that it is possible to use data augmentation in place of crowd sourcing. The system has positive predictive values of 98.1% and 99.9% for detecting OQDS and GY respectively. Additionally, we discuss the potential for diagnosis-as-a-service, and open-source our data for greatest benefit to the AI community.