Resin-Based Classification of Plastic Bottles Waste using CNN Architectures

The large use of plastic causes plastic waste problems in the environment. Recycling has become one of the solutions to deal with plastic pollution. Considering plastic waste is required to be separated based on the type of resin in the recycling process, the sorting process becomes more challenging. In this research, we aim to evaluate and compare the accuracy of the classification of plastic bottle waste by plastic resin type based on images using three CNN (Convolutional Neural Networks) architecture models. The CNN architectures used are ResNet50, DenseNet201, and InceptionResNetV2. We focus on PET and HDPE plastic bottles as the target object, and we develop the dataset by collecting about 1000 images for each class. The result shows that with a 99.4% accuracy rate, ResNet50 shows a better performance compared to DenseNet201 and InceptionResNetV2.