Waste recycling is a major part of the life cycle of our goods. The principle of recycling is to reuse the waste. A major challenge that waste recycling facilities are facing is the disruptive wastes that are not well sorted. Sorting errors represent additional management costs. One example of sorting problem is differentiating types of plastic. The most known recyclable plastic type is Polyethylene Terephthalate (PET). One way to solve the sorting errors is going to the source of waste: people. By educating people we can achieve fewer errors.CleverTrash deals with a recycle bin enhanced with a waste recognition system that educates users to properly recycle their waste. It also provides statistics on the sorting quality in recycling bins thanks to an embedded system (Raspberry Pi and sensors). These statistics are provided to the manager to improve the waste collection strategy. We explore the performance of Convolutional Neural Networks (CNNs) inside the waste recognition system. In machine learning, CNNs are mostly specialized in image recognition. We explore the performance of CNNs inside the waste recognition system. We started the project from scratch. We gathered our own dataset by setting up a trash can incorporating an embedded video and lighting system. The camera takes pictures of the garbage being thrown away. A machine learning model classifies the images. The main challenge was to develop the best machine learning classifier model from the pictures and evaluate its performance on this use case. We compare seven types of CNN algorithms and produce the results in the present paper.Our results show that CNNs have difficulties distinguishing between PET bottles and glass bottles on a classification "PET versus Other (non-PET)". It suggests that CNNs are strong to classify different shapes but weak when shapes are similar. It led us to try a classification of "Bottles versus Other (non-Bottle)" which performs better than "PET versus Other". The best result of the former classification performed as well as the worst result of the latter classification.