February 8, 2024
Stora Enso is a leading global provider of renewable solutions in packaging, biomaterials, wooden construction, and paper. To ensure the quality of their paper products, Stora Enso follows the INGEDE method for grading wastepaper as a raw material for paper recycling. This method requires the classification of wastepaper into several categories: newspaper, office paper, magazine, white cardboard, grey cardboard, brown cardboard, etc.
As this classification is done manually on small samples, Stora Enso decided to investigate if this procedure could be automated and joined forces with Apixa to develop a vision system to perform this grading task.
By combining two state-of-the-art technologies, Stora Enso and Apixa succeeded in completing a pilot phase for a fully automated wastepaper grading system. A hyperspectral camera captures a spectral signal for each pixel. These spectral signals are fed to a deep-learning network that computes a full segmentation of the input image.
The pilot project involved the manufacturing of a special setup for massive data collection, tools for annotation and synthesizing training data, and design and tuning of the deep learning network. The results were impressive: an overall accuracy of 93.7% for binary grading (paper versus cardboard) and 87.8% for multi-class grading on a large-scale experiment with 20K fully annotated images.
Would you like to integrate vision solutions into your production? Reach out to us here.