Maintaining a competitive edge in any industry requires that products are consistently produced. In the snack food industry, this requires that the texture, color, shape and general appearance remain uniform.
Image processing methods are attractive for achieving this in a rapid and non-destructive way.
Frito-Lay, a major snack-food producer was experiencing unacceptable variation in the seasoning level of several snack food products. The goal of this work was to use multivariate image analysis to quantify this seasoning level online to help improve understanding and control of the factors affecting seasoning level.
The technology developed was implemented online and enabled real-time feedback control of seasoning level. The closed-loop control system has been in place for several years and has enabled a 50% reduction in variation of seasoning level. This was a particularly challenging image analysis problem due to the random orientations and shapes of the snack food on the conveyor belt, leading to highly non-uniform lighting conditions. It would be very difficult for traditional image analysis methods to compete with the ProSensus algorithms for the rapid and robust separation of irrelevant (background) portions of the image and accurate prediction of seasoning.
The ProSensus Approach
ProSensus collected images of the snack food and analyzed them using specialized multivariate image analysis methods developed inhouse. Predicting the seasoning content first required an efficient method for separating the conveyor belt and other non-product portions of the image from the product portion. This was achieved using masking methods. A prediction model (based on PLS) was then built from features extracted from the product portion of the image to quantify seasoning content. The image below clearly shows that the model can segregate the background pixels and predict the seasoning level accurately.
The calculation is very rapid and robust, which was crucial for online implementation. Once online, the performance of the models was so good that a closed loop system was implemented to adjust the seasoning delivery rate depending on the calculated seasoning level. This system has produced significant savings in the years since it has been installed. There was a 50% reduction in seasoning content variation and also a reduced demand on lab personnel since seasoning content no longer needed to be analyzed routinely in the lab.
1. H. Yu, J.F. MacGregor (ProSensus), G. Haarsma and W. Bourg (Frito-Lay), “Digital Imaging for Online Monitoring and Control of Industrial Snack Food Processes”, Ind. Eng. Chem. Res., 42, 3036-3044, 2003.