计算机工程与应用 ›› 2022, Vol. 58 ›› Issue (3): 151-158.DOI: 10.3778/j.issn.1002-8331.2008-0075

• 网络、通信与安全 • 上一篇    下一篇

遗传算法优化时间卷积网络的手机来源识别

武钦芳,吴张倩,苏兆品,张国富   

  1. 1.合肥工业大学 计算机与信息学院,合肥 230601 
    2.智能互联系统安徽省实验室(合肥工业大学),合肥 230009
    3.工业安全与应急技术安徽省重点实验室(合肥工业大学),合肥 230601
    4.安全关键工业测控技术教育部工程研究中心,合肥 230601
  • 出版日期:2022-02-01 发布日期:2022-01-28

Source Cell-Phone Identification Using Genetic Algorithm Optimized Temporal Convolutional Network

WU Qinfang, WU Zhangqian, SU Zhaopin, ZHANG Guofu   

  1. 1.School of Computer Science and Information Engineering, Hefei University of Technology, Hefei 230601, China
    2.Intelligent Interconnected Systems Laboratory of Anhui Province(Hefei University of Technology), Hefei 230009, China
    3.Anhui Province Key Laboratory of Industry Safety and Emergency Technology(Hefei University of Technology), Hefei 230601, China
    4.Engineering Research Center of Safety Critical Industrial Measurement and Control Technology, Ministry of Education, Hefei 230601, China
  • Online:2022-02-01 Published:2022-01-28

摘要: 基于语音的手机来源识别已成为近年来多媒体取证领域中的一个研究热点。已有研究鲜有考虑环境背景噪声,难以满足司法领域实际应用场景的需求。提出一种遗传算法优化时间卷积网络的手机来源识别方法。基于对数域的Mel滤波器组系数特征,利用时间卷积网络进行深度语音特征学习,并利用线性判别分析提取低维深度特征,将低维深度特征输入到支持向量机中进行训练和识别。特别的,为了提高整体的识别性能,引入遗传算法,通过设计编码方式、适应度函数和遗传操作对时间卷积网络结构进行智能优化。对比实验结果表明,所提方法可对时间卷积网络结构进行自动设计,尽可能地发挥网络性能,从而进一步提升了识别准确率。

关键词: 手机来源识别, 时间卷积网络, 网络结构, 遗传算法, 智能优化

Abstract: The speech based source cell-phone identification has been a hot topic in the field of multimedia forensics in recent years. Existing studies rarely consider environmental background noises, which is difficult to meet the needs of practical application scenarios in the judicial field. To this end, this paper proposes a genetic algorithm optimized temporal convolutional network(TCN) for source cell-phone identification. Firstly, the logarithmic Mel-filter bank coefficients are input to the TCN for learning deep speech features, based on which the low-dimensional deep feature(LDDF) is extracted by using the linear discriminant analysis. Then, the LDDF is input into a support vector machine for training and recognition. Particularly, to improve the whole recognition performance, the genetic algorithm is introduced and encoding scheme, fitness function, and genetic operations are successively developed to intelligently optimize the architecture of TCN. Finally, comparative experimental results show that the proposed method can automatically design the TCN architecture to maximize its performance as much as possible and further improves the recognition accuracy.

Key words: source cell-phone identification, temporal convolutional network, network architecture, genetic algorithm, intelligent optimization