driedPorciniMushroomSlices

The industrial dataset consists of dried porcini mushroom slices from the Kündig Group (Kündig Nahrungsmittel GmbH & Co. KG, An der Salzbrücke 22, 98617 Ritschenhausen, Germany; and W. Kündig & Cie AG, Stumpfenbachstr. 38, CH-8006 Zürich, Switzerland). These were distributed chaotically on a conveyor belt and were photographed at the Hochschule Bielefeld – University of Applied Sciences and Arts (HSBI) with an RGB line scan camera (JAI – SW-4000TL-10GE trilinear line scan – 3 x 4096 pixels (R, G, B) CMOS Sensor – Pixel size: 7,5 mm x 7,5 mm and a Nikon Nikkor 35mm f1.4 lens).

The setup is based on the company's current batch-wise visual sampling quality inspection, in which the raw goods are checked for all possible contaminants in order to assess quality. In the context of automating this inspection, the data set was created to test anomaly detection models. Considered as contaminants, which were found in production, are aluminum, plastic and glass fragments as well as dried onions and nuts with a texture similar to porcini mushrooms.

The data was recorded in 4096x1024 pixels format. In addition, these were divided into 1024x1024 pixels images (as well as 512x512 pixels) to match common benchmark datasets. Ground truth masks were created for the "not good" images with contaminants. When slicing the images, it was ensured that there were no "good" anomaly-free images among the "not good" test data. In addition to the "good" anomaly-free training images, "good" test images are also provided.