Integrating Residual Network and Channel Attention Mechanism for Tire Pattern Image Retrieval

Abstract

Tire pattern image retrieval is important in obtaining useful clues for criminal case solving and traffic accident control. For complex deep learning models, the increase of the number of hidden layers can improve the model performance to a certain extent, but complex network architecture may cause the difficulty in network training, resulting in low retrieval efficiency. The use of residual blocks in residual networks can relieve this problem. To further improve the accuracy of tire pattern image retrieval, this paper combines a modified residual network with attention mechanism and proposes a novel tire pattern image retrieval model. A modified residual network is used for feature extraction, and a channel attention mechanism is added in the four convolution blocks (conv2-conv5) for feature selection. This method not only reduces the training parameters and make the model simpler, but also facilitates the extraction of more accurate tire pattern image features. Simulation results demonstrate the effectiveness of the proposed algorithm.