Atmospheric scattering and turbulence during the transmission of ultraviolet beams can cause severe intersymbol interference and transmission attenuation in a nonline-of-sight wireless ultraviolet optical communication system. To prevent these problems, a wireless ultraviolet scattering channel estimation method based on deep learning is proposed. In the training stage of deep learning model, the deep neural network (DNN) is optimized using differential evolution algorithm to accurately estimate the channel characteristics based on the optimal network output. Then, channel attenuation is compensated at the receiver. Simulation results show that compared with least-squares estimation, the mean-square error of the proposed method is increased by one order of magnitude, and the bit error rate is increased by two orders of magnitude. Compared with minimum mean-square error estimation, the mean-square error of the proposed method is increased by 38%, and the bit error rate is increased by 78%. In addition, differential evolution algorithm during DNN training accelerates learning convergence and promotes global optimization. Furthermore, the proposed method maintains stability in different environments with varying turbulence intensities in the channel model.