Research On Detection Of Abnormal Sound Based On Smart Lighting Monitoring System

Strengthening the security construction of smart cities,ensuring the safety of people's daily lives,and creating a harmonious society have become issues of concern to governments of all countries.With the rapid development of 5G,Internet of Things technology,and intelligent technology,street lights everywhere in towns have also evolved from traditional lighting functions to smart street lights with functions such as video surveillance,broadcasting,and video.The monitoring of smart street lights has the advantage of low cost,but there are blind spots in video surveillance,which cannot completely cover the entire surveillance area.The combination of sound and video has great significance for improving the intelligence level of the existing monitoring system,maintaining social harmony,and building a smart city.Aiming at the problem of monitoring blind spots in existing monitoring systems,this study mainly uses abnormal sounds such as crying and screaming that occur when an abnormal event occurs,and realizes sampling through wireless sensor network nodes deployed in smart lighting systems.Based on the analysis of the characteristics of the abnormal sound,the convolutional neural network is used to realize the abnormal sound recognition.A compression algorithm for convolutional networks based on clipping is proposed,which solves the problem that the convolution model is too large to be applied to edge devices.According to the installation location of smart lighting equipment,an abnormal sound localization strategy is designed to finally realize the identification and localization of abnormal sound.The main work of the paper is as follows:(1)Aiming at the problem of large convolutional neural network models,a model-based compression algorithm based on clipping is proposed.First,analyze the relevant theories of convolutional neural networks and model cutting algorithms.For model compression algorithms,analyze the key features of model compression,delete redundant weights in convolution,and reach the standard of model compression.Finally,prove the algorithm through experiments.Effectiveness.(2)Aiming at the current lack of sound source localization methods based on smart lighting application scenarios,this paper designs anomalous sound localization strategies based on smart lighting topology based on the characteristics of smart lighting equipment and the installation location of the lighting equipment,using time delay and energy localization methods to achieve The final localization of the sound source.Finally,the effectiveness of this localization strategy is proved by experimental simulation.(3)Based on the above research,this paper designs a prototype of the intelligent lighting system.The system includes two parts: software and hardware.The client can realize the real-time monitoring of the entire system.When an abnormal event occurs,the camera angle is automatically adjusted and the remote real-time alarm is provided.