基于特征融合和B-SVM的鸟鸣声识别算法

为了实现在野外通过低成本嵌入式系统识别鸟类，提出了基于特征融合和B-SVM的鸟鸣声识别方法。对鸟鸣声信号提取梅尔频率倒谱系数、翻转梅尔频率倒谱系数、短时能量和短时过零率组成特征参数，通过线性判别算法对特征参数进行特征融合。利用黑寡妇算法通过测试集对支持向量机模型的核参数和损失值进行优化得到B-SVM模型。利用Xeno-canto鸟鸣声数据集对本文算法进行了测试，结果表明该方法的识别准确率为93.23%。算法维度参数的大小和融合特征维度的高低是影响算法识别效果的重要因素。在相同条件下，文中所提的基于特征融合和B-SVM模型的鸟鸣声识别算法相较于其他特征参数和模型，识别的准确率更高，为野外鸟类识别提供了参考。

Bird sound recognition algorithm based on feature fusion and B-SVM

In order to identify birds in the wild through low-cost embedded systems, a bird sound recognition method based on feature fusion and B-SVM is proposed. The original feature parameters are composed of Mel frequency cepstrum coefficient(MFCC), inverted Mel frequency cepstrum coefficient, short-time energy and short-time zerocrossing rate extracted from birdsong signal, and the original feature parameters are fused by linear discriminant algorithm. By using the black widow algorithm to optimize the kernel parameters and loss values of the support vector machine model through a test set, the B-SVM model is obtained. In the Xeno-canto birdsong dataset, the recognition accuracy of this method is 93.23%. The size of the dimension parameters of the linear discriminant algorithm and the level of the fused feature dimension are important factors that affect the recognition performance of the algorithm.Under the same conditions, the bird sound recognition algorithm developed in this paper based on feature fusion and B-SVM model has a higher recognition accuracy compared to other feature parameters and models. It provides a reference for wild bird recognition.