Sound anomaly detection system based on deep learning

Abstract

The invention relates to a sound anomaly detection system based on deep learning, which comprises a sound feature extraction module, a depth learning classification module and a detection result post-processing module. The sound feature extraction module processes the obtained original sound data to obtain the corresponding audio features, and combines the speech features of a plurality of very short-time sound segments to obtain the short-time sound feature expression; the depth learning classification module is connected with the sound feature extraction module to label the original sound data; and the depth learning classification module trains the labeled sound data to obtain the depth learning anomaly detection model. After generating the abnormality detection model of depth learning,the short-time sound features generated by the sound feature extraction module are inputted into the abnormality detection model of depth learning for classification. The detection result post-processing module is connected with the depth learning classification module, and the detection result post-processing module encapsulates the result output by the depth learning classification module, andcombines the detection results of short-time sound to predict the abnormality detection results for a longer time.

Description

Technical field [0001] The invention relates to the field of sound abnormality detection, and in particular to a sound abnormality detection system based on deep learning. Background technique [0002] Various sounds will appear during the operation of the equipment in the substation, which may be normal sounds made in the running state, or abnormal sounds made by the equipment under fault conditions. After the power inspection personnel have been specially trained, they can judge the normal or abnormal condition of the equipment according to the different sounds emitted by the equipment. However, it is unrealistic to require power inspection personnel to monitor all substation sites 24 hours a day, and intermittent inspections cannot effectively detect abnormalities in substations in real time. At the same time, technological advancement has promoted the development of automation and intelligence in substation supervision. In response to this phenomenon, it is necessary to use...

AI Technical Summary

Problems solved by technology

However, it is unrealistic to require power inspection personnel to monitor all substations 24 hours a day, and intermittent inspections cannot detect abnormalities in substations in real time and effectively

Images