基于深度学习声谱图分类的“听声识风”

“Identify Wind Force by Listening”Based on Deep Learning Spectrogram Classification

将深度学习与声谱图相结合,提出了一种新型的风级识别方法——“听声识风”.在实验室条件下模拟1~4级风并记录对应风声音频.通过傅里叶变换等方法将风声音频转换成声谱图,共得到2608幅二维声谱图像用作数据集.将声谱图数据集导入深度卷积神经网络GoogLeNet中进行风力等级识别,测试准确率达到了99.6%.为了进一步证明实验结果的可靠性,将声谱图数据集分别导入ResNet18、ShuffleNet中进行训练,均获得了99.2%的测试准确率,结果表明该方法可以有效地进行风级识别.“听声识风”研究首次通过深度学习声谱图分类实现了对风级的识别,这是一种智能的、快速的风级识别新方法.

Deep learning and spectrogram are innovatively combined to propose a new wind force level identification method,i.e.,“identifying wind force by listening”.Under laboratory conditions,the sound from 1~4 wind force was recorded as the original wind audio,which was converted into acoustic spectrograms with the Fourier transform and other methods,and a total of 2608 two-dimensional acoustic spectrograms were obtained as network input data.The data were imported into the deep convolutional neural network GoogLeNet for wind force recognition,which reached an accuracy of 99.6%.In order to further verify the reliability of the experimental results,the spectrogram data were imported into ResNet18 and ShuffleNet for training,and the accuracy rate in both networks was 99.2%.The results showed that this method can effectively carry out wind force identification.“Identifying wind force by listening”realized wind force identification through deep learning for the first time,which is a new intelligent and fast wind force identification method.