Novel classification method of plastic wastes with optimal hyperparameter tuning of Inception_ResnetV2

Plastics have been used extensively over the past few decades. Prior to that, the use of plastics wasn't a major issue, but now we are polluting the ocean with approximately 12.7 million tons of plastic a year, and the damage we are doing to marine life and the ecosystem in general may soon be irreversible. This can be accredited largely to disposable plastics which have regrettably become ubiquitous. Therefore, the objective of this project is to properly classify waste through the use of deep-learning models with fine tuning. By doing this, they are complying with their legal duty of care. Overall, after the classification process, the Inception_resnet_V2 was separated and classified into three different classes: plastic, cardboard, and garbage. VGG19, VGG16, Inception_v3, Xception, and MobileNet were also used for the classification. Principal findings of our research concluded that there were only nominal differences in accuracy compared to the related works, which conducted binary classification. Also, for the optimizer, Nadam had the highest degree of accuracy, and was the highest when the learning rate was 0.01. As the CNN+Autoencoder model and the VGG16 model had a 35% difference in accuracy, we could determine the importance of pretrained models.