Artificial intelligence for cosmetics packaging

During the packaging process for sensitive and fragile products in the cosmetics sector, packaging plays an important part – lipsticks, for example, need optimum protection. But even before the last packaging step, quality control of the product that needs to be packed is indispensable in this case. SEA Vision uses artificial intelligence which is self-learning and continually improves.

During the visual inspection of products, the times when only the capabilities of the human eye were used, are long gone. Wherever possible, technology and machines are preferred which free staff members for other tasks. SEA Vision is making an effort to continually improve industrial image processing, which is used for packaging and automation processes.

A new application by the company intended for visual inspection of cosmetic products now uses technologies that utilise deep learning and neuronal networks. In order to create added value for companies, there is also work being done on the algorithms on which the software packets by SEA Vision are based. The new solution for inspecting lipsticks is still in development, but is already showing several of the options which will probably see forced development during the coming years. 

Lipsticks are among the most often sold cosmetics in the world. They come in different colours, shapes, formulas and combinations, and sometimes are embossed with logos and decorations. Many items therefore, that can be faulty. Among the most common faults in the body of a lipstick are product faults and deformities affecting the height of the stick or its shape. Aesthetically displeasing defects in the surface of the lipstick are also possible, as is an uneven distribution of colour. 

Faults like these are supposed to be detected by the new AI inspection system by SEA Vision for lipsticks. At the heart of this is the automation of pharmaceutical line clearance processes using neuronal algorithms. The goal of this technology is to improve the packaging of products. The system was developed by the SEA Vision Group in cooperation with a team from ARGO Visionand utilises the semantic segmentation of the parts of a lipstick – in this way, all faults can be found, pixel by pixel.

The segmented parts of the lipstick are divided into categories and each receive a label or a name. Every area of the image is classified according to a category and highlighted on the screen in colour. This allows the operator to directly view information about which parts need to be inspected.

The algorithm continues to learn by itself and thereby becomes able to identify more and more complex faults. This is made possible through a combination of real and synthetic images created by techniques for data expansion and neuronal generation. This base of proprietary datasets is complimented by the combination of different models and parameters which are learned over time.

As the artificial intelligence continues to learn, the inspection of the products continues to improve as well. The semantic segmentation technique based on deep learning is of essential importance. This is currently the standard for artificial intelligence and accelerates the development of abilities to analyse objects. The system becomes more and more precise and continues to develop itself as an increasing variety of scenarios can be taken into consideration. 

Kurzum: Die Automatisierung von Arbeitsschritten glänzt mit zahlreichen Vorzügen, stellt aber auch eine gewichtige Investition dar, gerade für Kleinunternehmer und Mittelständler. In Anbetracht zunehmend umworbener Fachkräfte werden sich aber wohl immer mehr Unternehmen für Automatisierungslösungen entscheiden, vor allem bei häufig anfallenden und leicht zu automatisierenden Arbeitsschritten.