3D light field display technology has been paid attention to by research scholars because of its large viewing angle and dense viewing viewpoint. Resolution is an important parameter of 3D light field display technology, and the method to improve the resolution is complicated, so the research scholars began to focus on the visual resolution. To improve the visual resolution of the light field displays, an approach based on convolutional neural network to acquire pre-processing elemental image (PEI) is proposed. In the imaging procedure of light field display, lens aberrations diffuse the pixels on the imaging plane. The aliasing areas of the diffuse pixels can be regarded as new visual pixels and used as extra information carriers. A visual resolution-enhanced convolutional neural network (CNN) is employed to obtain the pre-processing elemental image array (PEIA) from a high-resolution elemental image array (HEIA). The PEIA is loaded onto the LCD, which is optically transformed by the lens array and diffused by the diffuser to render a 3D light field display image with visual resolution enhancement. In the experiment, by using the LEIA, lens array and diffuser, a light field display with a 70° viewing angle and improved visual resolution is demonstrated.