The Research On Detection Method Of Abnormal Sound In Elevator Based On Combined Classification Algorithm

With the increase of high-rise buildings in our country,elevators have gradually become a tool that people often use in daily life.Since the elevator is a closed space,some emergencies cannot be discovered and resolved in time,so people are paying more and more attention to the security problems in the elevator,and traditional video surveillance can no longer meet the public’s demand for safety.With the development of digital audio technology,the emergence of audio surveillance has made up for the problems of obstructions,blind areas,and inability to detect dangerous situations in time in video surveillance.In order to identify abnormal sounds effectively and accurately,this paper studies the detection of abnormal sounds in audio monitoring,and proposes an audio classification algorithm combining convolutional neural network and support vector machine,and applies it to the detection of abnormal sounds in elevators,which improves The security efficiency in the elevator has an important guiding role.The main work of this paper is as follows:(1)Preprocessing of abnormal audio signals in elevators.According to the sound classification and classification basis,analyze the characteristics of different types of abnormal sounds,determine the feasibility of their classification and realize the digitization of sound signals.The power spectrum quotient detection method is used to detect audio endpoints,and separate effective audio signals and ineffective noise signals.Through pre-emphasis,overlapping framing and adding Hamming window,the high frequency part of abnormal audio is improved,and the short-term stability of the audio signal is preserved.The preprocessing of the audio signal provides the basis for the feature extraction and classification of the subsequent audio signal.(2)Feature extraction of abnormal audio signals in elevators.Through experiments,the audio feature extraction effects of time domain features,frequency domain features,and different feature parameters in frequency domain features are compared.The experimental results show that Mel Frequency Cepstrum Coefficient has strong robustness.Great,the feature descriptions of the same type of sounds are more similar,and the feature descriptions of different types of sounds are more distinguishing.Extract 40-dimensional MFCC coefficients as the feature vector of abnormal audio.(3)Classification of abnormal audio signals in elevators.Design and establish the classification model,and classify it through Multilayer Perceptron,Support Vector Machines,Convolutional Neural Network and the combination of CNN+SVM in this article.The algorithm classifies the abnormal audio signals in the elevator,and compares the classification accuracy of the abnormal audio in the elevator with different classification algorithms and the classification accuracy between different audios.Experiments show that the CNN+SVM combined classification algorithm has a more accurate effect on the classification of abnormal sounds in the elevator,with an accuracy rate of 89.5%,and the classification accuracy of different audio is generally improved.The detection of abnormal audio has a guiding role in improving the security of the elevator.