Acoustic Signal Recognition Of Excavation Equipments Based On Multi-feature Fusion

With the continuous advancement of urbanization in China,all kinds of urban construction happen,and the frequent underground pipeline accidents have caused great risk and loss to the society.Through analysis of the causes of underground pipe network accidents,the main reason is the damage caused by external forces.Therefore,how to protect the underground pipeline from violent construction of blind construction damage has become a problem which needs to be solved by electricity departments.Aimed at above problems,this paper analyzes four types of excavation equipment that would be used in the early stage of urban construction based on the recognition of sound signals,and proposed an acoustic signal recognition algorithm of Excavation Equipment based on Multi-Feature Fusion.The algorithm could effectively protect the underground pipelines.It can identify the sound source through acoustic signal of excavation equipment,gives early warning when detecting sound signals that would threaten underground pipelines and finally protect the pipelines.Based on the actual situation of the construction site,this paper analyzes the mechanism of acoustic generation in the excavation equipment and collects the acoustic data.In terms of feature selection,this paper has analyzed the classical LPCC feature extraction algorithm and MFCC feature extraction algorithm,with the characteristics of excavation equipment.Combining the methods of PCA and ELM-SAE respectively,this paper proposes two feature fusion algorithms,PCA-based feature fusion acoustic recognition algorithm and ELM-SAE based feature fusion acoustic recognition algorithm.This paper adopts the characteristic learning mode in machine learning to study the two algorithms above.To verify the performance of the proposed algorithms,this study has trained and tested the two models and used the conventional single-feature learning method as the experimental control group.The result shows that compared with the single-feature learning method,the PCA-based feature fusion acoustic recognition algorithm has a certain promotion on recognition performance and generalization performance of the model.While compare with feature fusion acoustic recognition algorithm based on PCA,the algorithm based on ELM-SAE has not only obviously improved on its recognition effects,but also has a relative greater advantage on its generalization ability.And this study has done experimental analysis point at the influence of the order of the single feature on model recognition,in the ELM-SAE-based feature fusion voice recognition algorithm.