Computer Vision
We develop solutions based on computer vision and deep learning for any industry. We accompany you throughout the process, designing a tailored solution.
Objective: Determining quality and process parameters in bakery products
The PHOTONICS4BAKERY project, “Research on the Industrial Applicability of Photonic Technologies for Determining Quality Parameters and Process Optimization in the Bakery Sector,” aims to develop a system based on hyperspectral technology to automate certain quality control processes within the production of doughs and bakery products. The specific objectives include controlling the fermentation point of bread loaves, detecting foreign objects, and monitoring the composition of bakery doughs (salt, water, fat, and protein).
This project has received public funding under the AEI 2021 call from the Ministry of Industry, Trade, and Tourism, as part of the aid established to support innovative business clusters to improve the competitiveness of small and medium-sized enterprises.
The participating partners in PHOTONICS4BAKERY are the Galicia Food Cluster Association (CLUSAGA), the Castilla y León Food Industry Association Vitartis, Hornos de Lamastelle, Industriales Panaderos Agrupados, Lugar Da Veiga, Advanced Optical Technologies, and ATRIA.
ATRIA’s role in the PHOTONICS4BAKERY project has been the development of the artificial vision algorithm for detecting the parameters of interest.
Solution: NIR and VISNIR Hyperspectral Cameras to Control Parameters
Quality control in the food industry is a highly important process. New technologies allow for a much more efficient and controlled way of checking product quality. In this case, hyperspectral technology is used to detect foreign objects and monitor parameters such as fermentation and the composition of bakery products in a non-invasive, online manner.
We used NIR and VISNIR hyperspectral cameras to capture images of various products and samples of interest, aiming to generate a dataset that would later be used to develop classification algorithms. These samples contain matrix information at different wavelengths in each pixel of the image.
Additionally, we trained a classification model to assess the fermentation state of a loaf of bread, distinguishing between “under-fermented,” “fermented,” and “over-fermented.” Currently, this process is done manually by an expert.
Finally, we trained regression models for different dough composition parameters: fat, protein, sodium, salt, percentage of trimmings, water content, and kneading level.
The results obtained demonstrate that the technology is viable for detecting dough parameters and can accurately determine the fermentation stage of a loaf of bread. Furthermore, the technology is effective at detecting foreign objects on the material’s surface.
