Digital image steganalysis network strengthening framework based on evolutionary algorithm

Abstract

Abstract To solve the problem of increasing network parameters and increasing fitting fluctuation during training due to the improvement or expansion of image steganalysis network, a framework for strengthening steganalysis network was considered. The principle of searching space optimization by evolutionary algorithm was used to guide the adjustment of parameter direction during network training. Firstly, the diversity of the steganalysis network is initialized based on the rules of the evolutionary algorithm, and the coding mapping is carried out according to the network structure to achieve the purpose of unifying the subsequent network strengthening objects. Then, according to the training state of the network, the corresponding strengthening positioning strategy is developed, and the evaluation function of the network individual is designed. Finally, according to the characteristics of steganalysis network training, suitable selection, crossover and mutation strategies are designed, and the network is strengthened. Experimental results show that the network strengthening framework has better learning ability in the training process, improves the detection accuracy of Xu-Net and Yedroudj-Net by more than 1.1% and 1.3%, respectively, and improves the convergence speed.