Abstract

Aiming at the current problems of complex and diverse waste varieties and low profits from secondary sorting in waste sorting and recycling, an image classification and recognition method of recycled waste based on deep residual neural networks is proposed. Based on the original ResNet50 neural network structure, by reducing the size of the convolution kernel and increasing the network width, the model training time is reduced and the model learning ability is improved. Combined with the characteristics of diverse varieties and different shapes of scrap images, the improved ResNet50 neural network is used to iteratively optimize the training set and the pictures in the training set are classified into five categories: glass, paper, cardboard, plastic and metal. The experimental results show that compared with other neural network structures, the classification accuracy of the improved ResNet50 proposed in this paper reaches 95.19%, and the network degradation problem caused by deepening the network layers is also avoided.

Keywords:Recycling waste classification; Image processing; Improved ResNet50 network; Depth residual neural network