Seal seam inspection of food packaging
Hyperspectral imaging as solution for the detection of bubbles and impurities in the sealed seam of plastic packaging for sausage and ham
Ensuring that food packaging only contains what belongs inside, was the task a manufacturer of ham and bacon approached us with.
Since even the smallest impurities in the seam, which are invisible to the human eye, can lead to leaking packaging and spoilage of the goods and can result in expensive recalls, a camera-based inspection of the ham packages was selected as solution for the future.
In a first attempt, we wanted to tackle this task with colour cameras and different illuminations, but since these cameras only work with colour comparisons, the results were not satisfactory. For this reason, hyperspectral imaging was chosen, as this technology also detects contaminations outside the visible light spectrum, in our case product residues or melted grease in the hot-sealed area of the packaging, which remains hidden from both the human eye and other inspection technologies.
Hyperspectral cameras, on the other hand, reliably identify materials based on their biological and chemical composition and can reliably detect bubbles or inclusions even through printed foils.
In our application, a HSI-camera, mounted at a height of approximately 1m above the conveyor, captures 6 packages per second. In the following step the images are analysed using pvSealInspect, a software developed by phil-vision. First, the seal seam is extracted from the rest of the package and then the area of the seam is explicitly inspected.
Label detection
In addition, a label recognition was to be integrated into the application. For reading barcodes and text a monochrome camera and white illumination are therefore part of the system.
Visualisation of results
The results of the test are visualised in red or green. Below the coloured OK or NOK test results, statistics are displayed showing the total number of packages checked and the number of good and bad packages. In addition, in the case of bad packages, a distinction is made between the test criteria "bad seam" and "bad label" and the percentage error rate in relation to the total quantity is displayed. As additional information, the last hundred OK and NOK at this position are counted and recorded to identify any serial defects that may occur at a particular position. What exactly constitutes a defect can vary from pack to pack, but in our application, packs should only be ejected if several bubbles or foreign bodies such as fat are detected in the sealing seam. Single bubbles below a certain size are not considered a defect.
The integration of the SPECIM-FX camera we selected was easy. A challenge was the processing of the large 12-bit spectral images that we optimised for calculation on the GPU, i.e., by filtering out different spectral signatures based on the chemical composition and visualisation using RGB colours. This massively simplifies the segmentation of defects.
Patrick Gailer, responsible for this application at phil-visionsummarises the insights gained from the project:
"With this first HSI application, we were able to gain a lot of experience with a technology that was new to us and we can imagine a lot of other interesting applications not only in production lines for food, but also applications like freshness checks of various natural products such as meat, fish, cheese, vegetables or fruit, or the inspection of packaged food for various foreign bodies such as plastic parts, wood, paper or hair. Since hyperspectral cameras can also detect and evaluate the composition (moisture, purity, fat, protein, etc.), there are countless possible applications in the food sector. In a special software developed from this project, we first determine in which spectral range errors lie and can then process the images with classic image processing tools or AI."
Your individual solution
We would be happy to discuss your ideas and requirements in detail and develop a special solution together with you, tailored to your individual needs.