应用目标检测网络自动检测ECG信号所含噪声

Automatic detection of noises in ECG signals by target detection network

现阶段的心电(ECG)信号去噪网络多通过滤波器滤除噪声,但是通常噪声滤除不彻底,从而造成ECG信号失真。基于此,本研究将目标检测网络用于ECG信号中的噪声检测,首先从MIT-BIT心律不齐数据库中筛选无噪声ECG记录,加入两种来自于MIT-BIH噪声压力测试数据库中的真实噪声,生成4个不同信噪比的含噪信号并构成训练及测试数据集。然后针对ECG信号的特殊性,对YOLOv3网络结构进行修改,设计YOLO-ECG目标检测网络,使用迁移学习策略训练目标检测网络。实验结果表明,本研究提出的网络在ECG信号中噪声检测时的F1值达0.9558,具有良好的检测效果。

At present,denoising network removes the noises in electrocardiogram(ECG)signals through the filter,but usually the noise filtering is not complete,which causes ECG signals to be distorted.Herein target detection network is used for the detection of noise in ECG signals.Firstly,two kinds of real noises from MIT-BIH noise stress test database are added in the noise-free ECG records which are screened from MIT-BIT arrhythmia database,thereby generating noisy signals of 4 different signal-to-noise ratios and forming a training set and a test set.Secondly,according to the particularity of ECG signals,the YOLOv3 network structure is modified for developing a YOLO-ECG target detection network,and the migration learning strategy is used to train the network.The experimental results show that the F1 value of the proposed network in the detection of noises in ECG signals is up to 0.9558,indicating that the proposed method has good performance in detecting noises in ECG signals.