摘要
针对目前冲击声识别系统稳健性较差问题,提出一种基于贝叶斯压缩感知(BCS)的冲击声识别方法。该方法基于BCS模型,根据训练样本构造传感矩阵;基于稀疏系数的分布特性设计分类算法完成目标识别;将提出的方法与传统分类算法的识别结果做了对比和分析。实验结果表明:提出的方法能够对识别相似冲击声进行精确分类,识别精度和抗噪性均优于SVM和GMM分类算法。
For the poor robustness of impact sounds recognition system available, an impulsive sound recognition method based on Bayesian compressive sensing （BCS） is proposed. The method is based on the BCS model. The sensing matrix is con-structed by the training samples. The target recognition is accomplished by designing sorting algorithm on the basis of distribu- tion character of sparse coefficient. The recognition results from the new method and traditional classification algorithm are com- pared and analyzed. The experimental results show that the method can accurately distinguish sounds with very similar properties and outperforms the SVM and GMM methods in recognition precision and noise immunity.
出处
《现代电子技术》
2013年第17期52-54,共3页
Modern Electronics Technique
基金
国家自然科学基金项目(11074202)
关键词
冲击声
压缩感知
观测矩阵
特征提取
impact sound
compressive sensing
measurement matrix
feature extraction
