In the process of steganography, the choice of carrier greatly affects the security of steganography system. Experiments showed that the security of steganography would be greatly reduced when the preprocessed images were used for steganographic embedding. When a large number of alternative images were given, multiple images could be selected for steganography to improve the security of the steganography system. In this paper, the deep neural network model was used to screen out the images that had been preprocessed in the alternative images, and the alternative images were preprocessed according to the output of the deep neural network, and the most suitable images for steganography were selected by combining the traditional embedding distortion algorithm. Compared with other multi-vector steganographic carrier selection schemes, the proposed carrier selection scheme had better performance in the case of preprocessed images in the alternative images, and could significantly improve the security of steganographic systems.