A waste classification method based on a capsule network

Garbage recycling and automatic sorting are efficient ways to address the paradox of rising municipal waste. Although traditional image classification methods can solve the rubbish image classification problem, they ignore the spatial relationship between features, which can easily lead to misclassification of the same object. In this paper, we propose the ResMsCapsule network, which is a trash picture categorization model based on the capsule network. By combining the residual network and multi-scale module, the ResMsCapsule network can improve the performance of the basic capsule network greatly. Extensive experiments using the publicly available dataset TrashNet show that the ResMsCapsule method has a simpler network structure and higher garbage classification accuracy. The classification accuracy of the ResMsCapsule network is 91.41%, and the number of parameters is only 40% of that of ResNet18, which is better than other image classification algorithms.

Automated summary

Garbage recycling and automatic sorting are efficient ways to address the paradox of rising municipal waste. In this paper, we propose a trash picture categorization model called ResMsCapsule network, which combines the residual network and multi-scale module to greatly improve the performance of the basic capsule network. Extensive experiments using the TrashNet dataset show that the ResMsCapsule method has a simpler network structure and higher garbage classification accuracy than other image classification algorithms. The ResMsCapsule network achieves a classification accuracy of 91.41% with only 40% of the parameters of ResNet18.