Environmental Noise Classification System Based On Convolutional Neural Network

At present,the use of neural networks to classify sounds is one of the research hotspots in the field of audio signal processing,and the common sound research focuses on the classification of music and speech.With the development of economy,there are more and more harmful noises in daily life.In order to realize the automatic analysis of environmental noise and related data,and then to solve the noise in a targeted manner,this dissertation takes environmental noise as the research object,introduces a classification system based on convolutional neural network.This system has the advantages of high recognition rate,low cost,and labor saving.First of all,in order to solve the problems of massive feature engineering and low classification accuracy of traditional sound classification algorithms,this dissertation studies sound signal processing,sound feature extraction,classification model recognition and other technologies,and focuses on the exploration of sound classification models.In this dissertation,the sound classification algorithms of convolutional neural network and circular neural network are respectively used and verified in the data set of Urban Sound8 K.The accuracy was 87.51 % and 78.92 % respectively.Secondly,in order to solve the problem of the low accuracy of the traditional classification algorithms of convolutional neural networks and recurrent neural networks,this dissertation proposes a GRU-FICN network.This network uses a one-dimensional multi-scale convolution model.The model connects the one-dimensional multiscale convolution with the improved long and short time memory network in parallel.And the algorithm is verified using the public noise dataset Urban Sound8 K,achieving an accuracy of 91.01 %.Compared with the classification results of convolutional neural network and cyclic convolutional neural network,the experiment shows that the algorithm has a higher classification accuracy.Finally,in order to verify the practicability of the algorithm proposed in this dissertation,the proposed GRU-FICN algorithm was applied to the environmental noise monitoring system,and after periodic experimental tests,it was confirmed that the environmental noise monitoring system embedded in the algorithm can not only identify the type of noise can also be monitored in real time and gives timely warning.It indicates that the algorithm has certain practical significance.In summary,this dissertation proposes an improved convolutional neural network classification algorithm and applies it to an environmental noise monitoring system.The comparison is performed under various experimental conditions,which verifies the reliability and effectiveness of the method in this dissertation.