A double branches binary neural network with the application for garbage classification

Articles in Press

Document Type : Article

Authors

School of Mechatronic Engineering and Automation, Foshan University, Foshan 528000, China

Abstract

With the development of artificial intelligence, automatic garbage classification technology gradually replaces the traditional manual sorting method. Deep neural networks are popular in the field of artificial intelligence, however, it faces the problems of a number of layers, millions of parameters, the heavy computation and storage, which inevitably limits its application for garbage classification in practice. In order to improve the efficiency of waste classification, a garbage image classification model based on double branches binary neural network (DBBNN) is proposed in this paper. In DBBNN, an improved network architecture with an extra compensation module is designed to offset the information loss. Based on hinge loss function, an improved network loss function named HP-loss is proposed. Combined with the exponential decreasing learning rate, the DBBNN model is trained to meet the requirements of garbage classification task. In order to illustrate the performance of the proposed model, comparative experiments on CIFAR-10 and GIGO public datasets have been done for seven different models. Then, DBBNN is applied for automatic garbage classification on our dataset of garbage objects. The experimental results illustrate that the proposed DBBNN exceeds other four compared models in terms of classification accuracy.

Keywords

Main Subjects

Scientia Iranica

Articles in Press, Accepted Manuscript
Available Online from 24 June 2024

Files

History

  • Receive Date: 28 July 2023
  • Revise Date: 26 March 2024
  • Accept Date: 24 June 2024

Share

How to cite

Statistics