Development of Vision Based on Recycling Material Identification for Reverse Vending Machine Platform

Abstract

Reverse Vending Machine (RVM) is one of interactive platform that can boost recycling activities. It works by providing reward to a user that return the recycle items to the machine. To achieve that, the RVM must be equipped with material identification capability so that each inserted recycle materials can be rewarded accordingly. In this project a vision based recycle material identification system was proposed using Convolutional Neural Network (CNN) concept and also implementing the Yolo object detection algorithm. In this thesis, the convolutional neural network (CNN) concept and Yolo algorithm will be implemented in detecting the sample images, and sample video for obtaining the efficiency of detection rate by using Yolo algorithm. The main approach of this thesis is to focus on development of vision based on material identification for reverse vending machine platform. Regarding on the aim of the project development, CNN is a better platform as it can learn discriminative patterns automatically from images by stacking convolutional layers. Besides that, CNN is being classified as one of the most powerful image classifiers and currently responsible for computer vision field in machine learning. The sample of images are required in this project as the images will undergoes the training and validation process which is one of the important parts of having convolutional neural network. The convolutional neural network in Reverse Vending Machine is for beverage containers recognition and sorting purpose. In general, the overall operation of the reverse vending machine will need the support from the CNN in order for the machine to function as a detecting recycling material. In this project, a dataset of recycling materials that contains around 20 images for each class which are PET bottle and aluminum can are used for training purpose. It is expected that the system would recognize the targeted objected when tested for image detection process, video detection process, and real time detection process.