Using Artificial Intelligence to Drive Disease Identification in Potatoes

The South African potato agriculture industry is worth around 7.5 billion rand and employs an estimated 45 000 permanent and seasonal labourers. One major issue this industry faces is the prevalence of diseases such as Black Scurf, Silver Scurf and Powdery Scab. These diseases are a much-talked-about issue, given their negative impact on the commercial value of potatoes and their effect on the industry as a whole. However, one of the most prominent issues facing farmers is the inability to contain large outbreaks without sacrificing a significant portion of potentially healthy yields. While it is possible to manually and routinely examine crops to catch outbreaks before their exponential spread, this method is expensive and inefficient. New and innovative ways to rapidly identify diseases in the field and packhouse are needed to expedite this examination process and allow farmers to respond promptly to potential outbreaks.

In order to address the problem, the head of the MakerSpace Unit collaborated with a professor from the Faculty of Natural and Agricultural Sciences to investigate a machine learning model, which is the automated creation of advanced algorithms to train a computer to describe pre-analysed data correctly. From the investigations, models were developed using convolutional neural networks, a specialised form of machine learning inspired by biology with the collaboration of experts in the field that could be tested and enhanced further.

To address accessibility concerns to this tool, a web interface was developed to facilitate the collection and analysis of potato images from the field. The Potato AI platform can analyse provided images and, using the pre-analysed data and training, can determine if the potato is diseased or not. Furthermore, all images submitted to the platform are captured for future manual review to add to and refine the model further and improve its accuracy. The platform can currently identify Black Scurf, Silver Scurf and Powdery Scab, with an average accuracy of 84%.

This project is an ideal example of the cross-disciplinary research and service offering available to all students and researchers who visit the University of Pretoria's Department of Library Service’s MakerSpace.