Good news for seed potato growers. After collecting data for 4 years, the trained disease-search algorithm appears to be able to recognize diseased plants. This emerged during the practical demo this summer, which was well attended by partners, growers en trading houses connected to the AGROS project.
High scores in recall and precision
In recent years (up until 2023) all kinds of images of plants have been collected by so-called RGB camera’s. These were mainly images of plants of different breeds with Erwinia infection and various viruses. A portion of these images was used to train and test an artificial intelligence (AI) disease-search algorithm under different conditions. The results appear promising.
The disease-search algorithm was able to achieve a recall score of 0.86 in the Erwinia infected plants. This means that 86% of the diseased plants were found. With a precision score of 0.96 (96% of the plants were correctly classified as diseased or healthy), this score was also high. In virus-infected plants the recall score was 0.97 and the precision score 1.00 (2022). A year earlier the precision score was 0.80
Tests in plots with other varieties did not yet score optimally
In 2023, images were also recorded at locations other than just at WUR Field Crops in Lelystad. For example, images of about 200 different varieties were recorded in a field at the Dutch General Inspection Service (NAK). When the algorithm was tested on these images, the results were slightly lower. 97% of healthy plants were classified as healthy, but 36% of diseased plants were not classified as diseased.
More research and working towards a commercial application
The above picture was confirmed in last summer’s field demonstration. Participants could see – in real time – how a plant was rated. The parties present drew the conclusion that it is now important to work quickly towards a commercial application. Those present also indicated the need for further research into generalizing the algorithm (seeing if the algorithm works well in more varieties and under various conditions). Growers also urged more research on finding diseased plants as early as possible. This is expected to contribute greatly to preventing further infection of a field.