Campus Garbage Image Classification Algorithm Based on New Attention Mechanism

Li Yan ； Xiaowei Wang ； Shoulin Yin

《International Journal of Electronics and Information Engineering》 13卷4期 (2021/12) Pp. 131-141

https://doi.org/10.6636/IJEIE.202112_13(4).01

摘要

The traditional campus garbage image classification can not meet the requirements of contemporary campus garbage classification. As a result, there is a low accuracy of garbage classification defects. Due to improving the efficiency of campus garbage classification, a campus garbage classification algorithm based on a new attention mechanism is proposed. The algorithm proposed in this paper is based on the ResNet-34 network and optimize in three aspects, including the multi-feature fusion, the attention mechanism, and loss function modification. Multi-feature fusion makes each feature more effectively fused and enhances the robustness of the extracted image features. The attention mechanism is constructed to complete the extraction of two local features and one global feature so that the model can obtain more effective feature information. In addition, the phenomenon of gradient disappearance is avoided. Finally, the improved loss function makes the feature information extracted from the model have better generalization ability and discrimination ability. The proposed three optimization methods are trained and tested on the Huawei Garbage sorting challenge public data sets respectively to ensure optimization feasibility. Finally, the proposed algorithm is verified by comparing the accuracy with other Garbage classification methods on the constructed Garbage data set. Thus, the low accuracy of traditional image classification is solved.

關鍵字

Attention Mechanism ； Campus Garbage Classification ； Feature Fusion ； Loss Function