A new multi-domain entropy signal classification method

Abstract

Multi-scale permutation entropy (MPE) and convolutional neural network (CNN) signal classification methods are effective techniques. However, it still has limitations such as inaccurate calculation results and the ability to extract only signal time-domain features. To solve these problems, a classification method based on refined composite MPE (RCMPE) and CNN is first proposed, and the computational accuracy of MPE and CNN is enhanced by setting different starting points to obtain multiple coarse-grained subsequences, and furthermore, the RCMPE and CNN are proposed to be used in the classification of signals, time–frequency domain variation method was introduced, and a multi-domain entropy signal classification method based on RCMPE and CNN is proposed, called the multi-domain entropy signal classification method, which extracts signal features from time-domain, frequency-domain, and multiple time scales to perform classification. Simulation experiments show that RCMPE has the best feature extraction effect when the length of a single sample is greater than or equal to 4096. Actual experiments have shown that the multi domain entropy classification method based on RCMPE has the highest signal recognition rate in discriminating between ship radiated noise signals and mechanical failure signals, especially in distinguishing mechanical fault signals, with a recognition rate of 100%.