Domestic waste image classification algorithm based on improved EfficientNetV2
Lou Li, Nan Gou · 2023
Aiming at the problem of low efficiency and poor real-time performance of domestic waste classification, this paper proposes a domestic waste classification model based on improved EfficientNetV2. The model was trained by introducing transfer learning, and ECA (Efficient Channel Attention) attention mechanism was introduced to enhance the feature extraction of the model. LeakyReLU activation function and Adam optimization algorithm were selected to improve the generalization ability of the model. Thus, a high-performance and lightweight garbage classification model ECAE-Net (Efficient Channel Attention EfficientNetV2) is obtained. Experimental results show that the accuracy of the proposed model reaches 96.8%, which is 1.3%higher than that of the initial model EfficientNetV2-S. At the same time, the number of parameters and floating point calculations are reduced by 57.3%and 52.3%respectively. The improved model has achieved excellent performance in the domestic waste classification task.