基于深度学习的小样本声目标识别方法

Small sample acoustic target recognition method based on deep learning

摘要:

声目标分类识别是声源识别领域的核心问题。然而，在应用深层神经网络进行声目标分类识别时，从少量样本中学习（样本复杂度较低）是一个具有挑战性的问题。针对此问题，提出了一种基于深度学习的小样本声目标识别方法，该方法将手工设计特征和对数梅尔声谱特征结合到一起，扩充了深度学习模型的可利用特征量，提高了声信号识别效率和精度。在实验验证中，该方法在测试集上实现了87.6％的识别精度；更进一步的，用较少量的训练样本对该方法和其它几种主流的深度学习模型的性能进行了比较验证，结果表明，该方法只需要更少量的数据即可实现同样的识别精度，在声源探测领域具有一定应用价值。

Abstract:

Acoustic target classification and recognition is the core problem in the field of sound source recognition. However, learning from a small number of samples (low sample complexity) is a challenging problem in the application of deep neural network to acoustic target classification and recognition. In order to solve this problem, a method of small sample acoustic target recognition based on deep learning is proposed. The method combines manual design features with logarithmic Mel spectrum features, which expands the available features of deep learning model and improves the efficiency and accuracy of acoustic signal recognition. In the experimental verification, the recognition accuracy of this method is 87.6% on the test set; furthermore, the performance of the method is compared with other mainstream deep learning models with a small number of training samples. The results show that the method can achieve the same recognition accuracy with only a small amount of data, which has certain application value in the field of sound source detection.