Traditional phase-only holographic imaging methods rely on high-intensity iteration, which is time-consuming, and the imaging quality is not high. To address this issue, a phase-only hologram generation algorithm based on depth learning and angular-spectrum layer-oriented, which can generate holograms quickly and improve the quality of hologram reconstruction, is proposed. The LeNet network structure predicts the complex amplitude information of three-dimensional objects, which reduces the amount of calculation. The accurate angular-spectrum algorithm creates a high-quality phase-only hologram of a three-dimensional object. The simulation results show that the algorithm is feasible and the quality of the reconstructed image is effectively improved.