首页>
外国专利>
ANIMAL SOUND IDENTIFICATION METHOD BASED ON DOUBLE SPECTROGRAM FEATURES
ANIMAL SOUND IDENTIFICATION METHOD BASED ON DOUBLE SPECTROGRAM FEATURES
展开▼
机译：基于双谱特征的动物声音识别方法
展开▼
页面导航
摘要
著录项
相似文献
摘要
An animal sound identification method based on double spectrogram features, comprising the following steps: establishing a sound sample library; acquiring a sound signal to be identified; converting a pre-stored sound sample and the sound signal to be identified into a spectrogram; standardizing the spectrogram, decomposing and projecting a feature value, and converting same into a projection feature XK; converting the spectrogram into an equivalent LBP value matrix u, and counting the variance of grey values of a corresponding pixel and surrounding pixels, so as to form a feature vector LBPV; combining the projection feature XK with the feature vector LBPV to form a double-layered feature Xk + LBPV; by using a double-layered feature set corresponding to the pre-stored sound sample in the sound sample library as a training sample set and using a double-layered feature corresponding to the sound signal to be identified as an input sample, obtaining a type, in the sound sample library, corresponding to the sound signal to be identified by means of random forest training; and outputting a result. The method improves the identification rate of various animal sounds with a low signal-to-noise ratio in different sound environments.
展开▼
机译：一种基于双谱图特征的动物声音识别方法，包括以下步骤：建立声音样本库;获取待识别的声音信号;将预先存储的声音样本和待识别的声音信号转换为声谱图;标准化频谱图，分解和投影特征值，并将其转换为投影特征X K ;将所述频谱图转换为等效的LBP值矩阵u，并对对应像素与周围像素的灰度值的方差进行计数，形成特征矢量LBPV;将投影特征X K 与特征向量LBPV结合，形成双层特征X k + LBPV;通过使用与声音样本库中预存储的声音样本相对应的双层特征集作为训练样本集，并将与待识别的声音信号相对应的双层特征作为输入样本，得到类型，在声音样本库中，对应于通过随机森林训练识别的声音信号;并输出结果。该方法提高了在不同声音环境中具有低信噪比的各种动物声音的识别率。
展开▼
