S1 and S2 Heart Sound Recognition using Optimized BP Neural Network

For the problems of Back Propagation(BP) neural network relying on initial weights, slowing convergence and easily falling into local extremum, the development ability of standard Artificial Bees Colony algorithm is weak, local search ability is poor, etc, propose an improved artificial bees colony algorithm to optimize BP neural network for fundamental heart sound(FHS) recognition. A novel improving following bees global search and probability selection algorithm, applying the optimized BP neural network to the FHS recognition is proposed. For the problems of heart sound contain noisy and Mel Frequency Cepstrum Coefficient(MFCC) feature parameters of heart sound signal are not effective under the condition of low signal-to-noise ratio(SNR). Propose an improved method to extract MFCC parameters, experimental results show that heart sound improved Mel Frequency Cepstrum Coefficient(IMFCC) feature is superior to MFCC and homomorphic envelope(Homo-Env) feature in the same case of classifier. In the same feature parameters, the improved Artificial Bees Colony algorithm optimization of BP neural network recognition accuracy has a greater degree of improvement, comparing with the classical BP, Random forest, support vector machine, k-Nearest Neighbor algorithm.