GARBAGE IMAGE CLASSIFICATION BASED ON RESIDUAL ATTENTION MECHANISM

To solve the problem of garbage image automatic recognition, a classification algorithm based on residual attention mechanism and dual loss function is proposed. The method was based on the pre-training model. The model focused on local feature information and global feature information at the same time by constructing residual attention module. A dual loss function combining label smoothing and Focal Loss was added, which effectively alleviated the problem of over-fitting caused by category imbalance. The experimental results show that the proposed method can effectively improve the feature extraction capability and classification accuracy of the model. The classification accuracy on the TrashNet, Huawei datasets and extended datasets is 98.32%, 96.92% and 96.66%, respectively, which is better than the current best performance method.