Feature extraction and classification of mine microseism and blast based on EMD-SVD

To solve the difficult problem in identifying rock mass microseism and blasting vibration signals, a method for feature extraction and classification is proposed based on the empirical mode decomposition (EMD) and singular value decomposition (SVD). Firstly, the mine signals are decomposed by EMD, and the IMF1 to IMF6 selected by correlation coefficients and variance contribution ratios are the main intrinsic mode functions (IMFs). Then the SVD is used to obtain singular valuesσ_i(_i=1,2···,6) of feature vector matrix constructed of the main IMFs. Furthermore, the support vector machine (SVM) is adopted to train, classify and recognize the signals of Yongshaba mine. The results show that there are large differences of singular valuesσ₁,σ₂ and σ₃ between microseisms and blasts, and the best pattern recognition is obtained whenσ₁ is 7.5 with an accuracy rate of 88.25%. In addition, the SVM method with an accuracy rate of 93% is better than the BP neural network method, Bayes method and boundary value method. In conclusion, the proposed method provides a new way for the feature extraction and classification of mine microseism and blast.