摘要
为解决由于噪声的干扰导致心拍识别性能仍不理想的问题,提出一种基于深度自编码器的心拍识别方法。该方法使用收缩自编码器和稀疏自编码器相融合,从噪声心拍信号中生成具有稀疏性的有效心拍特征,并使用余弦距离度量输入样本和生成特征之间的相似度。基于生成的心拍特征,使用面向不同患者的卷积神经网络模型进行心拍识别。在MIT-BIH数据库上对该方法进行了实验验证,并与经典心拍识别方法对比。实验结果显示该方法大大提高了心拍识别的总体准确率和F1值,且具有较高的识别性能。
In order to solve the problem that the recognition performance of heart beat is still not ideal due to the interference of noise, an heart beat recognition method is proposed based on depth auto-encoder. The method used the contractive auto-encoder and sparse auto-encoder to generate the effective heart beat feature with sparse characteristics from the noise heart beat signal, and used the cosine distance to measure the similarity between the input sample and the generated feature. Based on the generated beat features, a convolution neural network model for different patients was used to recognize the beat. This method was verified by experiments on MIT-BIH database, and compared with the classical beat recognition method. Experimental results show that this method greatly improves the overall accuracy and F1 value of heart beat recognition, and has high recognition performance.
作者
白淑雯
游大涛
武相军
Bai Shuwen;You Datao;Wu Xiangjun(School of Computer and Information Engineering,Henan University,Kaifeng 475004,Henan,China;School of Software,Henan University,Kaifeng 475004,Henan,China)
出处
《计算机应用与软件》
北大核心
2022年第12期159-166,共8页
Computer Applications and Software
基金
国家自然科学基金项目(61872125)
河南省科技攻关项目(182102410051)。
关键词
心拍识别
深度自编码器
心拍特征生成
不同患者
卷积神经网络
Heart beat recognition
Depth auto-encode
Beat feature generation
Different patients
Convolutional neural network(CNN)
