To address the problem of low accuracy of tire laser scattergram recognition, this paper proposes a new classification network for tire laser scattergram (CA-ResNet50). Firstly, ResNet50-based residual network is selected to change the residual block structure in the traditional ResNet50 network model to maximize the role of batch normalization. Then, a lightweight convolutional attention module is introduced to enhance the feature extraction ability of the network model for tire defects. Next, LeakyRelu activation function is used instead of the Relu activation function to solve the neuronal deactivation problems.Finally, the training data set is extended to overcome the problems of insufficient data volume and overfitting of the network model in training. The CA-ResNet50 proposed in this paper is compared with the current commonly used classification network models on the same dataset, and the experimental results prove that the testing accuracy of the proposed network model in this paper is higher than other networks for tire laser scatter maps, and the recognition accuracy can reach 99.7%.