Abstract:
In order to solve the problem of partial blockage detection of urban drainage pipeline and the difficulty in identifying the degree of blocking,a method based on multi-feature extraction,combined with random forest for blockage recognition is proposed in this paper.Firstly,the acoustic response signals of the drainage pipeline are divided into frames as feature set.Then the characteristics of A weighted total sound pressure level,energy entropy and fractal box dimension of the feature components are extracted respectively,so the classification feature sets can be constructed.In addition,the distance separability criterion is introduced into feature selection to remove redundancy and enhance the discrimination of the classification feature sets.The weighted fusion of the three features is achieved by using the trace of the scatter matrix in the class as the weight.Finally,random forest identification model is established from the classification feature sets based on the weighted fusion.The results demonstrate that multi-feature fusion has a higher recognition rate based on the distance separability criterion of the classification feature set.Meanwhile,random forest identification model has higher accuracy and faster recognition rate compared with SVM recognition model.Furthermore,the method can not only effectively identify different degrees of pipe blockage and multiple blockage,but also can eliminate the impact of the pipe accessories such as lateral connection for fault identification.