AGROS disease control seed potatoes : report 2020-2023

Research output: Book/ReportReportProfessional

Abstract

The Netherlands are the world’s biggest producer of certified seed potatoes. In temperate climate regions, the main diseases of the seed potato crop are caused by viruses and bacterial infections (Dickeya and P. atrosepticum and P. c. subsp. Brasiliense). Farmers face the risk of declassification or rejection of this high value crop and spend a lot of effort to detect diseased plants and remove them before inspection by the Dutch General Inspection Service (NAK). In this project the potential of a vision based AI application that can detect these diseases in seed potatoes is explored. The objective was to develop a robust algorithm for both diseases, that can work under different conditions (weather, soil) and in different varieties. After that, testing of such an application under practical conditions was part of the project and combined with an interaction with seed potato farmers, as well as participants of the seed potato value chain, to investigate the needs and expectations. The major part of the project was related to the creation of an annotated data set, based on 4 years of data and different weather and soil conditions. Most of the data was collected on experimental fields with highly infected varieties. In the last year also collected data on other locations besides Lelystad (Tollebeek, Valthermond). The data analyses was done with the EfficientnetB_0 neural network. A model trained on data of 2020-2023 resulted in an accuracy of about 80%. Testing the model with different subsets, based on location or on different varieties show a similar level of precision. The best test use case was the one done at the NAK location in Tollebeek on a field with 200 different varieties in small plots. The precision level of the global model dropped to a level of 74%. This result shows that a further improvement of the algorithm by extension of the dataset is needed to improve the generalization of the model. A challenge for a vision based system is to detect symptoms of Erwinia that are sometimes located deeper in the crop foliage. That is why a technical setup was chosen with 3 cameras: one top view and 2 cameras under an angle of about 30 degrees. A first comparison between the model’s performance when using only the top view camera or using a combination of the top view camera and 2 cameras looking at the plant under a 30degree angle, show similar results when based on majority voting system. Additional research is needed to test different decision making approaches. The developments in the project were shared with farmers and representatives of seed potato value chain. Farmers were positive about the quality level of the algorithm presented. They are aware that creating a robust vision - AI base system will need extra efforts. Also meeting the quality level used by the Dutch General Inspection Service (NAK) needs extra work and testing. Although the classification algorithm needs improvement, farmers are interested in an early introduction of the new technology. Farmers also confirm the strong preference for early detection of sick plants.
Original languageEnglish
Place of PublicationWageningen
PublisherWageningen Plant Research
Number of pages30
DOIs
Publication statusPublished - 2024

Publication series

NameReport / Stichting Wageningen Research, Wageningen Plant Research, Business Unit Field Crops
No.WPR-OT-1091

Fingerprint

Dive into the research topics of 'AGROS disease control seed potatoes : report 2020-2023'. Together they form a unique fingerprint.

Cite this