Deep learning and X-ray imaging for non-destructive internal quality inspection of foods in inline application

Most commercial quality inspection systems for foods are based on external quality only and the evaluation of the internal quality is mostly lacking. Internal disorders develop before or during storage or processing and are often impossible to detect externally. In manual destructive inspection, a number of randomly selected foods of each batch are examined. The whole batch is often discarded when the number of foods with disorders is greater than a certain amount. The disadvantages of this are financial losses because batches with high incidence of disorders still can contain a significant number ofunaffected products. Also, definitely not all foods with disorders are discarded, leading to a waste of resources further down the chain and a negative effect on consumer trust with respect to a specific brand and retail store. The aim of this project is to develop a cost-effective non- destructive internalquality grading system to deliver high quality foods by combining 3D measurement and X-ray radiography with efficient deep learning and parameterized shape models to extract information from products in an online application on a conveyor belt at commercial speeds. The successful principle was demonstrated in the SBO project TomFood and patented. In this project a prototype system is designed, optimized image analysis protocols are developed and different food quality control applications are tested as a leverage towards successful technology transfer to industry (licensing, collaborative projects).

Keywords:X-ray, internal defects, sensors for food, artificial intelligence, online sorting

Disciplines:Agrofood mechatronics