Automated Waste Sorting Using Convolutional Neural Network
Hieu Huynh, Phu-Thinh Pham-Hoai, Anh-Kiet Tran, Thanh-Dat Nguyen
Abstract
The waste classification has become a crucial mission for sustaining worldwide economic growth and preserving the environment. Using deep learning to sort solid waste automatically is necessary since it could minimize the time taken to categorize a large amount of rubbish manually and health risks created by working with polluted waste. In this study, we take advantage of several Convolutional Neural Networks such as VGG, Resnet, Efficientnet, etc. to solve this problem. The test accuracies achieved by training Resnet101, EfficientNet-B0, and EfficientNet-B1 on the dataset of 6640 images are 92.43%, 90.02%, and 91.53% respectively. We also build an ensemble model on the base of these three models, which attains an accuracy of 94.11%. The dataset is from the Trashnet dataset of Stanford and images collected on the Internet. This approach can be potentially applied to real-life environmental problems.