A Novel Insect Sound Recognition Algorithm Based on MFCC and CNN
Mei Zhang, Lina Yan, Guilan Luo, Gang Li, Wenzhi Liu, Lina Zhang
Abstract
It is very necessary to recognize the sound of insects by using the differences sound between the different kinds of insects when they are moveing, feeding, and calling. Therefore, the insect sound recognition algorithm based on Mel Cepstrum Coefficient (MFCC) and Convolutional Neural Network (CNN) is proposed in this paper. After the sound samples are preprocessed, Mel Cepstral Coefficients (MFCCs) are extracted from the sound samples as feature parameters, and then the sound is converted into a feature map. Finally, the feature map is input into the constructed convolutional neural network (CNN) for training. Through the setting of hyperparameters and the selection of optimization algorithms, the trained network is used to recognition insect sound. Its average recognition rate reached 92.56%, which greatly improved the accuracy of insect voice recognition. The test results prove that MFCC and CNN can be used to effectively identify insects.