Classification and Modeling of Acoustic Gunshot Signatures
Classification and Modeling of Acoustic Gunshot Signatures
复制标题
DOI：
10.1007/s13369-013-0655-5
复制
发表时间：
2013-09
影响因子：
2.9
通讯作者：
M. Djeddou;Tayeb Touhami
中科院分区：
文献类型：
--
作者：
M. Djeddou;Tayeb Touhami
本文研究了噪声环境下根据枪声特征对枪声进行分类的问题。对有用信号的准确选择是提取显著特征的关键。为此，我们建议使用一个预处理步骤来准确地选择与枪击特征相关的帧。此外，使用具有高斯混合模型(GMM)的整个特征集不能提供高的分类率。在此基础上，提出了一种层次分类方法。后者是基于GMM算法建模的倒谱特征，然后使用时间参数对两个子类进行分类。这在区分接近的签名类方面有很大的贡献。实验获得了高达96.29%的正确分类正确率。所使用的时间参数与炮口爆炸信号有关。为了推广这一特征的使用，推导了一个经验方程，并用实际数据集进行了验证。
In this paper, we deal with the problem of gunshot classification according to its acoustic signature in a noisy environment. The precise selection of useful signal is essential for a significant features’ extraction. For this purpose, we propose to use a preprocessing step to select accurately the gunshot signature related frames. In addition, the use of entire features set with the Gaussian mixture model (GMM) does not provide a high classification rate. Then, a hierarchical classification is proposed. This latter is based on cepstral features modeled by GMM algorithm and followed by a classification of two subclasses using temporal parameter. This contributes significantly in discriminating between close signature classes. Our experiments yielded a high correct classification rate up to 96.29 %. The used temporal parameter is related to muzzle blast signal. To generalize the use of this feature, an empirical equation is derived and validated with a real data set.