Online Monitoring System Based on Convolutional Neural Networks for Urban Environmental Noise

Noise monitoring is becoming more and more important with the development of the city. In this paper, an online monitoring system for urban environmental noise based on a convolutional neural network is designed and developed which mainly includes a backstage management system and noise acquisition and transmission unit. The central control system uses the FCU1104 IoT gateway which is responsible for data acquisition and the distributed remote modules are used for the data transmission and control signal sending. Modbus RTU, Modbus TCP and SFTP communication protocols are used for data communication for security and stability. A convolutional neural network is used to deal with Fbank features of the audio for acoustic source recognition. And then, the number of CONV layers and the Pooling layer is determined. At last, a one-month data acquisition rate test and accuracy of noise recognition test are carried out. The test results show that the data acquisition rate is 98.585% and the accuracy of noise recognition test is 81.4%. It prove that this system has stable rate of data acquisition and high accuracy of acoustic source recognition.