基于小波降噪与ResNet分类网络的异常电子蜂鸣音识别系统设计

Design of Abnormal Electronic Beep Recognition System Based on Wavelet Denoising and ResNet Classification Network

设计一种蜂鸣音识别系统,以PC端声卡与收音麦克风采集蜂鸣音信号s_(1),利用小波包尺度系数比对法实现对采集的蜂鸣音信号降噪预处理得到信号s_(2),通过信号s_(2)的小波变换得到信号时频特征图p,再利用ResNet18网络模型分析特征图p,得到蜂鸣音检测结果,识别准确率达97.5%。该系统采用PyQt5设计人机交互界面,主要包含信号特征显示、信号采集参数设置、识别对象选择、数据通讯、数据持久化等功能。实验证明了该系统操作简单、运行稳定、扩展性强等特点。

A buzzer sound recognition system is designed,the buzzer signal s_(1) with the PC sound card and the radio microphone,and uses the wavelet packet scale coefficient comparison method to realize the noise reduction and preprocessing of the collected buzzer signal to obtain the signal s_(2),and obtain the signal time-frequency characteristics through the wavelet transform of the signal s_(2) figure p,then use the ResNet18 network model to analyze the feature map p to get the buzzer detection result,the recognition accuracy is 97.5%.The system uses PyQt5 to design a human-computer interaction interface,which mainly includes functions such as signal feature display,signal acquisition parameter setting,identification object selection,data communication,and data persistence.Experiments have proved that the system is simple to operate,stable in operation,and strong in scalability.