The dynamic balance of rotating wheels has a significant impact on vehicle handling and comfort. In order to detect the tire dynamic balance state timely and accurately, an attention mechanism based on feature learning structure is established on the traditional 1D-CNN model, and proposes an improved anti-noise 1D-CNN method for rotating wheel dynamic balance health state monitoring. Three kinds of dynamic unbalance states of rotating wheels are obtained by adding different mass balancing weights to the real vehicle, and then the vibration information of rotating wheels is obtained. Gaussian white noise is used as the noise input to process the vibration information of the measured rotating wheel in different dynamic equilibrium states to obtain the test sample data. The convolution operation mechanism and feature transformation are comprehensively used for t-SNE visual display. The results show that compared with the traditional 1D-CNN model (No attention mechanism), under the conditions of different signal-to-noise ratio, the improved dynamic balance state monitoring method of noise resistant 1D-CNN rotating wheel proposed in this paper shows better diagnostic accuracy, which can be as high as 99.95%.