Based on the approximate dual relationship between the orbital angular momentum modal variable and the target azimuth variable, vortex electromagnetic wave radar can achieve two-dimensional highresolution imaging of stationary targets, but the Bessel function term in the target echo is able to seriously affect the azimuth focusing performance. The existing Bessel function compensation method based on inverse projection algorithm is computationally intensive and difficult to apply in practice. In view of the above-mentioned problems, this paper proposes a method by utilizing U-Net convolutional neural network for suppressing the influence of Bessel function and realizing high-resolution imaging of vortex electromagnetic wave radar. Firstly, the U-Net network is improved according to the sparse characteristics of radar targets in the observation space, and on the basis of this, the target defocus image is obtained by two-dimensional fast Fourier transform preprocessing of the target echo signal, and the target defocus image is further used as an input of the improved U-Net network, and the target ideal electromagnetic scattering model is used as the network output to supervise and train the network. Finally, based on the unknown target echo signal, the preprocessed target defocus image is input to the well-trained network model, and the well-focused high-resolution imaging results can be obtained. The simulation experiments show that the proposed method can effectively improve the focusing performance of target imaging, and the network model still is good in generalization ability in the presence of noise.