Noisy Vehicle Recognition Based On ResNet And GRU

With the acceleration of urbanization,the problem of traffic noise pollution has become more and more serious,and the "street bombing" behavior of noisy vehicles is even more difficult to control.The main reason is that the electronic control valve of such noisy vehicles makes the occurrence of "street bombing" controllable,which greatly increases the difficulty of manual supervision.Therefore,designing an "electronic police" that can be monitored throughout the day has become an urgent problem to be solved in the management of noisy vehicles.Based on the related research results of deep learning in the field of sound recognition,this paper designs a model structure based on Res Net and GRU to identify noisy vehicles in a complex noisy environment.Tests show that the recognition accuracy of this method is at More than 96%.Effectively promote the development of smart cities and contribute to solving traffic noise pollution.The main tasks of this article are as follows:(1)The unique voiceprint structure of the noisy vehicle is analyzed.Using the combination of the microphone array and the host computer,through the sound pressure level detection,the collected traffic noise samples are analyzed to find the unique voiceprint structure characteristics of the noisy vehicle below 1k on the slope.Compared with other acoustic features,this feature can better distinguish noisy vehicles from other noises.(2)Established a sample library of "street bombing" incidents of noisy vehicles,conducted shallow network learning and training on a small amount of manually labeled samples collected at the initial stage,and jointly judged the collection events with the trained model and the sound pressure level algorithm Samples,constantly iterate its own model,and enrich the sample library.It provides a feasible method for deep learning problems similar to the lack of sample libraries.(3)A new model architecture Res Net+GRU is proposed,which not only solves the problem of weak generalization ability of the BP model under complex interference,but also solves the problem of gradient disappearance and gradient explosion caused by the traditional RNN kernel,and adaptively obtains the characteristics Extracting the network level can extract excellent features that are more suitable for actual problems than the traditional CNN network,and finally get better results than other models in the recognition of noisy vehicles.Provide reference experience for other similar fields of work,such as images,etc.