Evaluation of Transfer Learning based Deep Learning architectures for Waste Classification

Today, the development and modernization have led to the generation of waste, which has become a problem for all the living beings and the environment, whether it is medical waste, of which 25% is hazardous, or household waste, which contains harmful plastic. In this work, a deep learning solution has been provided towards the classification of a few classes of wastes such as plastic, paper, metal, glass, cardboard, etc. In this work, transfer learning has been used and applied. This helps deep learning models to accomplish the classification task in the most accurate way. The models such as EfficientNet, ResNet34, Densenet121, ResNeXt-50 32x4d, Wide ResNet50_2, Densenet169 are used here in this work. Extensive experimentation was done with the different optimizers. The training such as Adam optimizer, RMSprop optimizer, and Adadelta was performed and the experimental results demonstrate that the Adam optimizer produced the best results over the other competing methods. The proposed work has achieved a test accuracy of 98.02% using ResNeXt-50 32x4d and 95.8% using Wide ResNet50_2 architecture.