基于新进化优化BP学习算法的心音识别方法

为提高人工智能辅助诊断心音识别的准确率, 根据心音信号的周期性特点, 提出以快速主成分分析算法对心音信号降维和提取特征, 同时基于单形进化算法, 优化BP神经网络学习算法的输出与期望的误差函数, 以改进BP神经网络的学习性能, 实现对心音信号高准确度的分类识别. 针对正常心音及8类异常心音信号进行性能分析与测试, 实验结果表明, 各类心音的平均识别率为95.96%, 改进算法比其他对比算法识别率分别提高了4.9%,3.9%,1.9%, 表明该算法能更有效地分类识别心音信号, 提高人工辅助诊断的识别率.

Heart Sound Recognition Method Based on New Evolutionary Optimization BP Learning Algorithm

In order to improve the accuracy of heart sound recognition of artificial intelligence assisted diagnosis, according to the periodicity of heart sound signal, we proposed a fast principal component analysis algorithm to reduce the dimension of heart sound signal and extract features. At the same time, based on the simplex evolution algorithm, the output of BP neural network learning algorithm and the expected error function were optimized to improve the learning performance of BP neural network and realize the classification and recognition of heart sound signal with higher accuracy. Aiming at the normal heart sound and eight kinds of abnormal heart sound signals, the performance was analyzed and tested. The experimental results show that the average recognition rate of all kinds of heart sounds is 95.96%. Compared with other algorithms, the improved algorithm improves the recognition rate by 4.9%, 3.9% and 1.9% respectively. It shows that the proposed algorithm can effectively classify and recognize heart sound signals and improve the recognition rate of artificial assisted diagnosis.