Recupel is handling the recycling of all Belgian electronic products. Different groups of products require different recycling procedures and Recupel needs to have a clear view on how much devices they are handling of each category. Currently this is a manual process during which Recupel samples ten thousands of devices each year and labels them manually.
We are using computer vision based on deep learning to automate the above sampling process completely and improve it by having better insights in the actual products, characteristics, etc. instead of only categories.
Examples of electronic appliance sub-category top-3 classification results:
We are not only interested in providing accurate results, but we would also like to understand what the model is doing. An example of how we can interpret the results is shown below. Notice that even though the model detects multiple appliances, it knows which one to classify.