Localization is an important issue in wireless sensor networks (WSNs). Aimed at the shortcomings of low localization accuracy of the existing 3D localization algorithms, in this paper, we develop a three-dimensional localization scheme of RF emitters which combines collaborative spectrum sensing with deep learning. We propose a semantic segmentation approach to identify the coverage range of the RF emitters which converts the three-dimensional sensing data into a series of two-dimensional image slices. Then, we design a weighted localization algorithm to accurately locate the RF emitters. The simulation results show that the proposed method is accurate in positioning under various parameter configurations.