基于遗传神经网络的音频水印盲检测

Watermark Detection by Genetic Algorithm and Neural Network Indirect Computing Model with Indirect Formal Method

本文将神经网络和遗传算法结合应用于小波域音频水印盲检测.为弥补BP神经网络在训练过程中因初始权值选取不当而出现的收敛震荡、收敛速度较慢的现象,本文使用遗传算法对BP神经网络进行优化。在随机产生的初始权值种群中应用遗传算法得到最优解作为BP神经网络的初始权值,并对BP神经网络进行第二次训练。将优化后的BP神经网络应用到小波域音频水印盲检测中以提高水印系统的鲁棒性。实验证明遗传优化的BP神经网络在训练中呈现较好的收敛特性,在盲水印的提取中比传统BP网络具有更强的鲁棒性.

This paper presents a method for digital speech watermarking detection in wavlet domain which combines the advantages of Genetic Algorithm（GA） and BP neurual network. Use the GA to optimize the initial parameters ofth BP neural network ,then choose the optimal parameters as the weight matrix of the network for further training.Use the optimized BP neural network to detect the watermarking.The experimental results show that the wartermar detection does not need the original carrier and is more robust than it is by the traditional BP neural network.