阅读说明：本技术 一种心电信号qrs波群定位方法及装置 (A kind of electrocardiosignal QRS wave group localization method and device ) 是由 罗伟 朱涛 李毅 于 2019-08-16 设计创作，主要内容包括：本发明涉及心电信号技术领域,公开了一种心电信号QRS波群定位方法及装置,其中方法包括以下步骤：分别以两组大小不同的数据段作为输入对神经网络进行训练,得到两个标签预测模型；以两个标签预测模型的预测结果作为输入对神经网络进行训练,得到融合预测模型；根据两个所述标签预测模型以及所述融合预测模型进行病例数据的QRS波群定位。本发明具有适用范围广、定位精度高的技术效果。(The present invention relates to electrocardiosignal technical fields, disclose a kind of electrocardiosignal QRS wave group localization method and device, wherein method the following steps are included: respectively using two groups of data segments of different sizes as input neural network is trained, obtain two Tag Estimation models；Neural network is trained using the prediction result of two Tag Estimation models as input, obtains fusion forecasting model；The QRS complex positioning of case data is carried out according to two Tag Estimation models and the fusion forecasting model.The present invention has technical effect applied widely, positioning accuracy is high.)

1. a kind of electrocardiosignal QRS wave group localization method, which comprises the following steps:

Neural network is trained using two groups of ecg signal data sections of different sizes as input respectively, obtains two labels Prediction model；

Neural network is trained using the prediction result of two Tag Estimation models as input, obtains fusion forecasting model；

The QRS complex positioning of case data is carried out according to two Tag Estimation models and the fusion forecasting model.

2. electrocardiosignal QRS wave group localization method according to claim 1, which is characterized in that respectively not with two groups of sizes Same data segment is trained neural network as input, obtains two Tag Estimation models, specifically:

A plurality of electrocardiosignal building sample data set is acquired, each sample data is demarcated；

Each sample data is cut with different cutting sizes, obtains two groups of data segments of different sizes；

Respectively the addition of each data segment embody its whether include QRS complex training label；

Neural network is trained using data segment described in two groups as input respectively, obtains two Tag Estimation models.

3. electrocardiosignal QRS wave group localization method according to claim 2, which is characterized in that with different cutting sizes Each sample data is cut, two groups of data segments of different sizes are obtained, specifically:

To set step-length and first be sized sample data described in the sizes cutting each such as at equal intervals, first group of number is obtained According to section；

To set step-length and second be sized sample data described in the sizes cutting each such as at equal intervals, second group of number is obtained According to section.

4. electrocardiosignal QRS wave group localization method according to claim 2, which is characterized in that calibration sample data, specifically Are as follows:

Demarcate the R crest location of the sample data.

5. electrocardiosignal QRS wave group localization method according to claim 2, which is characterized in that embodied for data segment addition Its whether include QRS complex training label, specifically:

Centered on the R crest location nearest apart from the data segment, it is cut into the sample data and the data segment The reference field of same size；

Calculate the IOU value of the data segment Yu its reference field；

It is that the data segment adds training label value according to the IOU value.

6. electrocardiosignal QRS wave group localization method according to claim 5, which is characterized in that according to the IOU value be institute It states data segment and adds training label value, specifically:

When the IOU value is less than the first given threshold, add for the data segment without QRS wave label value；

When the IOU value is greater than the second given threshold, QRS wave label value is added with for the data segment；

When the IOU value is greater than first given threshold and is less than second given threshold, added without label.

7. electrocardiosignal QRS wave group localization method according to claim 1, which is characterized in that with two Tag Estimation moulds The prediction result of type is trained neural network as input, obtains fusion forecasting model, specifically:

The prediction result of two Tag Estimation models is revised as the rear input neural network in the same size to be trained, Obtain the fusion forecasting model.

8. electrocardiosignal QRS wave group localization method according to claim 1, which is characterized in that two Tag Estimations The neural network that model uses is ResNet neural network, and the neural network that the fusion forecasting model uses is GRU nerve net Network.

9. electrocardiosignal QRS wave group localization method according to claim 3, which is characterized in that according to two labels Prediction model and the fusion forecasting model carry out the QRS complex positioning of case data, specifically:

The cutting case data are sized with described first and obtain first group of case data segment, are sized with described second It cuts the case data and obtains second group of case data segment, by first group of case data segment and second group of case data segment Corresponding Tag Estimation model is inputted respectively, obtains the first prediction label value and the second prediction label value；

The first prediction label value and the second prediction label value are inputted into the fusion forecasting model and obtain fusion forecasting label Value；

By the fusion forecasting label value compared with the trained label value, judge whether corresponding case data segment includes QRS wave Group, obtains the case data segment comprising QRS complex.

10. a kind of electrocardiosignal QRS wave group positioning device, which is characterized in that including processor and memory, the memory On be stored with computer program, when the computer program is executed by the processor, realize as described in claim 1-9 is any Electrocardiosignal QRS wave group localization method.

Technical field

The present invention relates to electrocardiosignal technical fields, and in particular to a kind of electrocardiosignal QRS wave group localization method and device.

Background technique

In cardiac diagnosis, the positioning of QRS complex is had very important effect.Neural network is as a kind of widely applied Mathematical model is also gradually applied in ECG's data compression.There is positioning accurate using the positioning that neural network carries out QRS complex Degree is high, the fast advantage of location efficiency.But since training sample is limited, determined at present by what neural network was trained When identifying to some special case data, i.e., the case data to differ greatly with training sample are identified bit model When, accuracy rate can decline to a great extent, and cause the applicable surface of location model relatively narrow, overall precision is pulled low.

Summary of the invention

It is an object of the invention to overcome above-mentioned technical deficiency, a kind of electrocardiosignal QRS wave group localization method and dress are provided It sets, it is narrow to the applicable surface of case data when solving to carry out QRS complex positioning by training neural network in the prior art, it is whole The low technical problem of precision.

To reach above-mentioned technical purpose, technical solution of the present invention provides a kind of electrocardiosignal QRS wave group localization method, packet Include following steps:

Neural network is trained using two groups of ecg signal data sections of different sizes as input respectively, obtains two Tag Estimation model；

Neural network is trained using the prediction result of two Tag Estimation models as input, obtains fusion forecasting mould Type；

The QRS complex for carrying out case data according to two Tag Estimation models and the fusion forecasting model is fixed Position.

The present invention also provides a kind of electrocardiosignal QRS wave group positioning device, including processor and memory, the storages It is stored with computer program on device, when the computer program is executed by the processor, realizes the electrocardiosignal QRS wave group Localization method.

Compared with prior art, the beneficial effect comprise that the present invention uses different size of data segment as defeated Enter to carry out the training of neural network, to obtain two different Tag Estimation models.The label that the training of larger data section obtains Prediction model is suitable for heart rate compared with the Tag Estimation model that slow, the wider case data of QRS wave, lesser data segment training obtain The case data very fast suitable for heart rate, QRS wave is relatively narrow.Using the prediction result of two Tag Estimation models as input again into The training of row neural network obtains fusion forecasting model.Since fusion forecasting model is based on two different Tag Estimation models It establishes, therefore fusion forecasting model is suitable for the prediction positioning of various different case data, applicable surface is wide, and for various differences The prediction positioning of case data can reach higher accuracy, to improve whole precision of prediction.

Detailed description of the invention

Fig. 1 is the flow chart of one embodiment of electrocardiosignal QRS wave group localization method provided by the invention；

Fig. 2 is the prediction result figure of one embodiment of electrocardiosignal QRS wave group localization method provided by the invention.

Specific embodiment

In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and It is not used in the restriction present invention.