Collagen is the main component of connective tissue and is found in all mammals. It is widely used in medical and food applications, one of which is sausage casing manufacturing.
Collagen casings are less expensive to produce, more consistent and easier to produce than natural casings made of gut.
The collagen skins are inspected for defects in the production line. The inspection is done by transmitting a light through the flattened skins. A camera captures the resulting images which are then passed to a computer system for analysis.
Black Speck Defect Larger than 1mm in Length.
The software application treats folds or creases present during the inspection of the casing as defects. This results in wastage and lowers the efficiency of the production line. A typical false-positive is shown in the image below.
False Positive: Sausage Skin Folding Rejected by the In-House Vision System.
The challenge is to devise an algorithm that can detect defects in the casings reliably and accurately, and not reject a folded sausage skin that is still usable.
Dr Paul and his team developed an algorithm that probes the color deviation and fluctuation for each pixel in the collagen skins. Based on the analysis results, the image is accepted or rejected.
The technique was able to locate and identify punctual defects in the skin. It did not reject the images showing creases. It did not work as well when a black speck was present within a fold: the system accepted the image. The researchers are working on refinements of the technique to avoid this problem.
This project reduced wastage, by reducing the rejection rate of otherwise usable products. The technique is more flexible than the algorithm used in the software that Devro currently uses and can be incorporated into any programming environment.
Testing showed an average 26% increase in the ability to detect false positives, reducing the operating costs of the production line.
Christopher D. Williams, Manoranjan Paul and Tanmoy Debnath. "Enhancing Automated Defect Detection in Collagen Based Manufacturing by Employing a Smart Machine Vision Technique", to be published in Lecture Notes in Computer Science (LNCS) as a post-conference publication.