Waste Classification System using a Hybrid Deep Transfer Learning Model

Waste accumulation in urban areas is of growing concern due to the problems of its collection, proper management, and disposal. Inefficient management and lack of separation of disposable waste negatively impacts humans. In addition, the mixing of disposable and non-disposable waste reduces the recycling rate. Therefore, there is a need for a smart waste management system that leads to the efficient disposal and recycling of waste in a sustainable and environmentally friendly manner. This paper presents a deep learning-based smart waste management system that can classify and segregate waste and help recycle it. Initially, a custom network is proposed to classify the waste into five categories: organic, plastic, metal, glass, and paper. The transfer learning approach reduces the training time and can perform better, and does not need much training data. Therefore, a transfer learning approach using trained MobileNetV2, Xception, EfficientNetB7, ResNet152V2, and DenseNet201 is presented in this paper. These models are evaluated using various quantitative parameters. It is observed that MobileNetV2 classified waste at 96% accuracy, whereas other models performed slightly better with 97% classification accuracy. However, the custom model achieved the least performance at 81.64%. Further analysis of transfer learning models showed that MobileNetV2 has minimum trainable parameters making it most suitable for waste classification. This model can be used to develop an intelligent bin that can segregate waste in real time, enhancing the recycling rate.