Abstract: Aiming at the issues of image feature insensitivity, high requirement of expert experience and low recognition accuracy of traditional machine vision methods in train wheelset bearing damage detection, based on the framework of convolutional and transformer fusion networks, this paper proposes a method for identifying damage to train wheelset bearings. First, an image preprocessing method named image enhancement category reorganization is used to improve the quality of the acquired image dataset and to eliminate the effects of imbalance dataset. Second, Convolutional Neural Network (CNN) has high model construction and training efficiency due to the adoption of local sensing field and weight sharing strategy, which can only sense local neighborhoods but has limited ability to capture global feature information.Transformer is a network model based on self-attention mechanism. With strong parallel computing ability, it is able to learn the remote dependencies between image pixels in the global scope, and has more powerful global information extraction ability. Therefore, based on the idea of fusion of convolution and self-attention, VGG and Transformer parallel fusion network (VTPF-Net) is designed in this paper, which integrates the global contour features and local details of the image. Furthermore, the multiscale dilation spatial pyramid convolution (MDSPC) module is constructed to fully mine the multiscale semantic features in the feature map using multiscale dilation convolution progressive fusion. Finally, experimental analyses were carried out based on the NEU-DET image defect dataset and the self-constructed train wheelset bearing image dataset. The experimental results demonstrate that the proposed model has an accuracy of 99.44% and 98.96% for the recognition of 6 types of defects and 4 types of images of wheelset bearings in NEU-DET data, respectively. Compared to existing CNN models, ViT model with self-attention mechanism, and CNN-Transformer fusion model, the proposed method shows significantly better evaluation metrics and accurately identifies different types of image samples.