Classification of Metal Objects Using Deep Neural Networks in Waste Processing Line (original) (raw)

Each year, a factory releases a lot of metal debris which is normally used in a recycling phase. In order to be effectively recycled, it is necessary to classify the debris into different classes. The sorting by hand takes a lot of times and effort. Other classification approaches which use color, size, weight, electrostatic, or magnetic features may not obtain high accuracy. It has a lack of technique to classify the metal debris. Thus, this paper proposes a framework for classification of metal debris which is spread on a conveyor belt. The framework employs deep neural networks. Four different deep neural network models were investigated and compared in our framework called the AlexNet model, the GoogleNet model, the VGGNet model, and the ResNet model to choose a suitable model for the framework. In addition, the experiments can also investigate and compare the operation of different deep neural network models in a practical application instead of using conventional academic benchmarks. Experimental results demonstrated that the proposed framework could be one solution to separate the metal debris. Especially, the AlexNet model had the highest accuracy among the four models.