基于LDA和小波分解的肺音特征提取方法

针对用小波分解提取肺音特征后特征向量维数较高的问题，提出了一种结合线性判别分析和小波分解的肺音特征提取方法。在该方法中，首先对肺音信号进行小波分解，然后将小波分解得到的小波系数转化成小波能量特征向量，接着使用线性判别分析法对该特征向量进行降维处理，得到新的低维特征向量，最后用SVM对低维特征进行识别。在实验中，选取了三种肺音信号：正常肺音、爆裂音、哮鸣音，用所提出的方法将8维的小波能量特征降为2维特征，在2维特征上进行了分类识别，并和降维之前的结果进行了比较，实验结果表明利用线性判别分析对小波能量特征降维后极大地提高了识别精度。同时，和其他几种典型的肺音特征提取方法进行了比较，实验结果表明结合线性判别分析与小波分解的特征提取方法得到了更高的识别精度。

Feature extraction method of lung sound based on LDA and wavelet decomposition

The feature vectors, which are extracted from lung sounds with the wavelet decomposition, have higher dimension. To solve the problem, a method to extract the feature of lung sounds is proposed, which combines the linear discriminant analysis and the wavelet decomposition. In the method, the lung sounds are firstly executed the wavelet decomposition. Then the wavelet coefficients from the wavelet decomposition are transformed into the wavelet energy feature vectors. Next the dimension of the feature vectors is reduced with the linear discriminant analysis. Finally, the lung sounds are recognized with SVM according to the low dimensional feature vector. In experiments, three kinds of lung sound signals are used：normal, crackle and wheeze. The wavelet energy feature vectors with 8 dimension are reduced to 2 dimension with the presented method. The lung sounds are classified with SVM according to the two-dimensional feature vectors. The results are compared with the results from the original data. The results demonstrate that the recognition accuracy is greatly improved through reducing the dimension of the wavelet energy feature vector with the linear discriminant analysis. At the same time, the proposed method is compared with other lung sound feature extraction method. The results also show that the recognition accuracy is higher by combining the linear discriminant analysis and the wavelet decomposition.