Research on Underwater Target Feature Extraction Method Based on Improved MFCC

The extraction of feature parameters has an important influence on underwater target recognition. Aiming at the low recognition rate of the traditional Mel Frequency Cepstral Coefficient (MFCC) feature extraction method in underwater target recognition system, this paper proposes an improved MFCC feature extraction method. This method uses Short-Time Fourier Transform (STFT) instead of the original time-frequency conversion method and increases the number of Mel filters in the feature extraction process. Finally, a Convolutional Neural Network (CNN) is constructed as a classifier to compare the recognition effects of MFCC features before and after improvement of three types of underwater acoustic signals. Experimental results show that the improved MFCC feature extraction method achieves a higher recognition accuracy.