Trash Classification Network Based on Attention Mechanism

The classification and recycling of garbage can greatly improve the utilization of garbage resources. This paper proposes a new convolutional neural network that fuses a multi-branch Xception network with an attention mechanism module. The effective feature information is emphasized and the invalid information is suppressed to overcome the problem caused by the small data set. To verify the usefulness of this network structure in the field of garbage images, this paper uses a widely used data set in the field of garbage image classification. For any network without pre-trained weights, the network proposed in this paper outperforms all other methods by 94.4%.