Foreign Object Detection for Electric Energy Meters Based on Multi-feature Fusion and Random Forest

Foreign object detection is an important measure to guarantee accurate measurement of electric energy meters. In view of the inefficient manual detection of foreign objects for electric energy meters, an automatic detection method of foreign objects for electric energy meters based on multi-feature fusion and random forest is proposed. Firstly, wavelet de-noising is carried out for the sound signal produced by electric energy meters under detection in the presence of background noise. Then, the time and frequency feature parameters are extracted to form the mixed feature matrix, which is input into the random forest composed of decision trees for classification. By analyzing the recognition rate of different feature fusion methods, an optimal feature fusion method was obtained. Experimental results show that the fusion mode of short-time energy, spectral entropy, LPC and MFCC exhibits the best performance, and its foreign object detection accuracy is higher than 93%.