Research on garbage recognition and classification based on ConvNeXt
K. Song, Fei Wang, W. Wang · IET conference proceedings. · 2023
This study proposes a garbage classification methodology that employs Convolutional Neural Networks (CNNs) to tackle the prevalent issues observed in China's urban household waste categorization system, as per the conventions and vocabulary of academic discourse. Using ConvNext_tiny as the backbone network, add channel attention and spatial attention mechanism modules to enhance the model's ability to extract garbage features; To improve the classification accuracy of the model, PolyLoss is used instead of the cross-entropy loss function; Introduce the learning rate attenuation strategy of cosine annealing using warm-up, and cooperate with the batch_size to achieve adaptive adjustment of the learning rate; Optimize model parameters with transfer learning to further improve model accuracy. The experimental findings indicate that the model's performance on the test dataset is exceptional, manifesting an impressive accuracy rate of 97.36%, a precision rate of 97.51%, and a recall rate of 95.51%, which is of great practical and reference significance for solving the problem of urban domestic waste classification, improving resource utilization efficiency, and reducing environmental pollution.