Deep learning detects defects in the food industry

Food & Beverages | Deep Learning | Deep Learning | OCR & OCV

Automation specialist INNDEO has developed a sophisticated automation solution for quality control in the packaging industry. Thanks to machine vision and deep learning technologies, high speeds and defect detection rates are no longer a problem.

The Spanish company INNDEO specializes in the automation of quality inspections and offers high-quality machine vision solutions for the food industry with its INSPECTRA brand. INNDEO has developed the Thermoseal & Label Inspector solution to reliably inspect packaging and read labels. The device combines a wide range of sophisticated technologies such as high-speed and processing capture, hyperspectral vision, deep learning, and high-performance RGB.

Fully automated inspection for the packaging industry

The advantages of a fully automated solution based on machine vision are higher detection rates of packaging defects, cost savings, as well as the comprehensive digitalization of production processes to monitor and improve them. As inspection processes are often still carried out manually in practice, and therefore defective products reach the end customer, the goal of an automated solution must be to reliably detect all conceivable defects in packaging. 100 percent automation of quality control reduces costs on the one hand and introduces objective criteria for sorting the objects to be inspected on the other. In addition, quality and production data can be continuously digitized end-to-end and the corresponding indicators displayed in real time.

Although there are solutions for inspections using machine vision, these are often not robust enough, have too low detection rates, or are difficult to adapt to changes in the production lines. INNDEO precisely addresses these weaknesses with its machine vision solution. The goal was to identify quality defects in food packaging at a high production rate of up to two packs per second to enable inline rejection with processing times of just a few milliseconds per image. It was therefore essential to automate the application end-to-end using machine vision.

MVTec HALCON software with deep learning technologies for defect detection

Within the setup of the Thermoseal & Label Inspector, cameras take images of the objects to be inspected at various points within the production environment. The images are processed by the integrated machine vision software MVTec HALCON. HALCON uses various parameters to determine the relevant inspection area (region of interest / ROI) of the image. To do this, INNDEO uses high-resolution RGB vision technology to find the simplest sealed area defects, such as a piece of ham, as its color is easily distinguishable in a transparent tray. In addition, the company uses hyperspectral vision technology to detect more complex defects, such as defects in opaque or printed trays.

Deep learning is also used to detect certain defects and is able to interpret images with a higher recognition speed and efficiency than the human eye ever could. The system is able to learn during a training phase without the need for additional programming by the user. This technology allows the detection of wrinkles in sealing films, faults in the arrangement of the product in the tray, and the detection of quality defects that cannot be distinguished by standard machine vision algorithms.

Another application scenario for machine vision is the inspection of labeling and checking whether wrinkles have formed underneath the label. To detect the label, a corresponding configurable tool looks for a specific pattern. Once this is located, the inspection processes take place. The application uses the optical character recognition technologies integrated in HALCON, such as OCR (Optical Character Recognition) or Deep OCR, which combine text recognition functions with intelligent deep learning algorithms. Deep learning technologies and pattern matching of color tones are also used to detect anomalies in the applied labeling.

Software offers flexibility in interface integration

For the end customer, it is important that the inspection solution can be seamlessly integrated into the existing process environment. This allows the user, for example, to control the inspection system in their familiar environment. Interface integration proved to be one of the biggest challenges during implementation, as the various inspection parameters can be configured from a different system control and all images from the various cameras have to be analyzed in a very short time. This means that the machine vision software only has a very short time to decide whether packaging is faulty and needs to be rejected. The integration of the solution was simplified by the possibilities of MVTec HALCON. "The software offers various interfaces for many types of industrial cameras and the possibility to directly run scripts within a real application using the HDevEngine and debug it. We see the wide variety of image processing algorithms, convenient programming, and seamless integration with our software as further strengths of MVTec products," confirms Emilio de la Red Bellvis, Chief Innovation Officer at INNDEO.

On the hardware side, the application setup includes various components, such as several industrial PCs. These receive the images from the individual cameras and communicate with programmable logic controllers (PLCs). INNDEO was confronted with a further challenge. Due to the shortage of electronic components, flexible programming had to be developed. This accommodates different camera types, processing architectures, and graphics processors so that the hardware can vary depending on availability.

Quality control solution wins award

As a result, INSPECTRA achieved 100 percent of the project objectives with its Thermoseal Inspector solution and the integrated HALCON machine vision software. The quality defects in both the packaging and the product itself are thus avoided. This eliminates the cost of destroying, replacing, and transporting defective products. In addition, producers can guarantee their brand promise to consumers at all times thanks to the reliable and consistent quality. The biggest advantage, however, is that quality control can be automated throughout the entire production process. This reduces the labor costs associated with quality assurance, lowers the error rate, and eliminates subjectivity in the inspection criteria. Finally, the technology can detect errors that would remain hidden to the human eye.

The quality of the solution not only impresses customers: At the Meat Attraction 2022 trade fair, the Thermoseal Inspector even received an award as the most innovative technology for the ancillary industries of the meat industry.

Sign up to stay up to date on machine vision trends, success stories and tips & tricks

Machine vision for every taste

The Dutch company Kobobss ensures packaging quality in the food industry with a modular solution. Part of the solution is the machine vision software HALCON from MVTec Software GmbH. This enables the company to increase the efficiency of packaging management and significantly reduce the number of recalls.

Read more

Application areas for machine vision software

Take a look at the highlighted industries to get an idea, what you can do with MVTec products. 

Discover more