Traffic scene sound classification method and system based on deep learning

Abstract

The invention discloses a traffic scene sound classification method and system based on a deep learning technology. The method comprises the steps: extracting a signal time-frequency feature from a sound signal after obtaining and screening the sound signal in real time through a sound collection device, and carrying out the analysis through a trained deep learning model, and obtaining a prediction classification matrix; and converting to obtain a confidence coefficient matrix comprising all sound event types appearing in the sound signal and corresponding time intervals of the sound event types. According to the invention, the sound signals are analyzed by using the deep learning training model, so that the types of the sound events and the occurrence time of the sound events can be accurately obtained for analysis, the accuracy and the capture rate of a monitoring snapshot system related to sound discrimination are remarkably improved, and the method has higher accuracy in classifying the sound of the traffic scene, has the capability of classifying multiple types of sound, and effectively reduces the false alarm of a monitoring and recognition system.

Description

technical field [0001] The present invention relates to a technology in the field of traffic sound recognition processing, in particular to a method and system for classifying traffic scene sounds based on deep learning. Background technique [0002] Existing sound classification technologies use calculated sound pressure levels, spectral features, MFCC features, etc. as the basis for sound classification. Usually, the requirements for sound signals are high, and the accuracy and recall of classification results are greatly affected by the usage scenarios. With the development of intelligent transportation technology and the continuous improvement of people's demand for traffic management, surveillance systems with video as the core are widely used in many traffic scenarios, and play a key role in traffic management, security and other applications. [0003] However, most of the existing monitoring systems are still based on video monitoring, and only some monitoring systems...

AI Technical Summary

Problems solved by technology

For the current illegal whistle capture system of motor vehicles, although the system combines sound monitoring and video monitoring, it still uses traditional methods in sound classification, and usually only deals with one type of sound, and does not have sound classification in sound processing. In actual use, there are many misjudgments in the judgment of the sound, which directly leads to the wrong capture and false alarm of the capture system, which is not conducive to the enforcement of traffic management personnel.

Images