Valuable Waste Classification Modeling based on SSD-MobileNet

• Published in: 2020 - 5th International Conference on Information Technology (InCIT) pp. 228 - 232
• Main Authors: Thokrairak, Sorawit, Thibuy, Kittiya, Jitngernmadan, Prajaks
• Format: Conference Proceeding
• Language: English
• Published: IEEE 21-10-2020
• Subjects: Biological system modeling; Computational modeling; Convolution; Convolutional neural networks; Machine Learning; Metals; MobileNet; Plastic products; Training; Waste Classification
• DOI: 10.1109/InCIT50588.2020.9310928

One of the most challenging classifications is the waste classification. If this can be done automatically, it will have a great use in waste management industry in terms of time and cost reduction. Nowadays, 2 kinds of waste classification and separation exist, namely the manual waste classification and the automatic waste classification using various technologies. The first one can be done using human understanding and power, while the latter is in the search of suitable methods for waste classification automatically. This work aimed to optimize the waste classification using SSD-MobileNet, which is a Convolutional Neural Network architecture. The purposed waste are the plastic bottles, the glass bottles, and the metal cans. The number of image dataset used in this work is 952. The model has 24,000 steps of training and 9 hours in total. The loss value of the model is 0.2711. The model is tested on Raspberry Pi 4 with the average classification accuracy of 95% for plastic bottles, 82% for glass bottles, and 86% for metal cans. With these waste classification accuracies, this model can be used in an embedded system for waste classification.