KR20220096278A - CNN-based exercise intensity classification system using pulse waves - Google Patents

Abstract

The present invention relates to a CNN-based exercise intensity classification system using pulse waves. The CNN-based exercise intensity classification system using pulse waves comprises: a PPG sensor which is a hardware terminal that measures a pulse wave signal; a monitor which is an output device; and an exercise intensity classification device which is a controller. The exercise intensity classification device comprises: a PPG signal input unit which inputs a signal from the PPG sensor; a data processing unit which extracts a P-peak interval and inputs the P-peak interval to a trained convolution neural network (CNN) unit; the CNN unit; an exercise intensity classification unit which classifies exercise intensity into exercise intensity classification classes according to heartbeats; an output unit which performs monitoring and displays graphs; a storage unit which stores data; and a control unit. Therefore, the present invention carries out preprocessing by measuring a pulse wave signal using a PPG sensor, extracts a P-peak interval, and inputs the extracted P-peak interval to a trained CNN artificial neural network to classify the exercise intensity into the exercise intensity classification classes according to heartbeats. Accordingly, the present invention can accurately calculate heartbeats and classification classes of various kinds of exercises and can be interlocked with an exercise application to be applied to a more objective exercise intensity feedback system.

Description

CNN-based exercise intensity classification system using pulse waves

The present invention measures a pulse wave signal through a PPG sensor using a pulse wave, performs a preprocessing process, extracts the interval of P-peak, and inputs it into the learned CNN artificial neural network to classify the exercise intensity into an exercise intensity classification class according to the heart rate Accordingly, it relates to a CNN-based exercise intensity classification system using pulse waves that accurately calculates the heart rate and classification class of various exercises and can be applied to a more objective exercise intensity feedback system by interworking with exercise applications.

Recently, due to the development of medical technology, as we enter an aging society, interest in health is increasing. According to the 2018 National Life Sports Participation Survey by the Ministry of Culture, Sports and Tourism, the participation rate in daily life sports increased from 45.5% to 62.2% for 6 years from 2013. In addition, many people prefer high-intensity exercise that has a high effect in a short time during exercise. High-intensity exercise consumes very high oxygen and calories during exercise, and it has a burning effect after consuming additional calories to supplement oxygen after exercise, so the exercise effect is very good. However, if you perform high-intensity exercise at the maximum level, a large strain on the heart may occur, which can lead to dangerous situations such as a heart attack or shock due to high blood pressure. Therefore, it is important to perform high-intensity exercise at an appropriate level for each individual. In general, heart rate or maximal oxygen uptake is used as a measure for estimating exercise intensity. In the case of maximal uptake, it is difficult to measure the maximal uptake in daily life because a gas analyzer must be used to measure the data, and an instrumentation mask and signal processing system are required. However, in the case of exercise intensity estimation using heart rate, since the heart rate sensor is diversified and miniaturized, it is easier to estimate the exercise intensity in daily life than the maximum oxygen intake. And as an example of the prior patent technology for estimating exercise intensity, registered Patent Publication No. 10-1777312 discloses a user terminal paid to a user, an exercise equipment terminal mounted on the exercise equipment, and an exercise equipment for integrated management of the exercise equipment terminal. In the exercise prescription system including the management server, the exercise equipment management server determines whether or not a disease is present from an individual survey, basic physical fitness information, and personal information, and designates one of a normal group category and a disease group category for each individual, One of the normal group category and the disease group category designated for each individual is stored, and the exercise equipment terminal readjusts the exercise intensity when the HR (heart rate) measured in real time during exercise is out of the preset target heart rate range, and the HRV It readjusts the exercise intensity according to the state of the LF/HF ratio obtained through PSD analysis, and has an exercise equipment calculation processing unit configured to readjust the exercise intensity when there is a sharp change compared to the QT interval of the electrocardiogram at rest. , when the exercise equipment terminal readjusts the exercise intensity according to the state of the LF/HF ratio obtained through PSD analysis of HRV, the LF/HF ratio is monitored in real time during exercise in units of 1 minute, and the An exercise prescription system, characterized in that it is automatically reset to a lower intensity than the prescribed exercise prescription, is disclosed when a state that is two times higher or lower than the stable range (1.5-2) for exercise occurs three or more times.

In addition, Patent Registration No. 10-1277177 discloses an exercise assisting device for assisting a user's exercise, comprising: an exercise program storage unit for storing an exercise program input from the user or developer;

When at least one exercise program is selected from among the stored exercise programs by the user, the selected at least one exercise program is input from the exercise program storage unit, and the at least one exercise program is used by the user to be used. Exercise program setting unit to set;

a movement and biometric information collection unit configured to collect the user's movement-related information and biometric information through at least one sensor;

an exercise pattern detection unit for detecting a specific motion corresponding to the set exercise program from among the collected user's motion related information, and outputting the biometric information together with result information on the detected specific motion;

Receives and stores result information and biometric information for the detected specific movement, analyzes the degree of improvement in the user's exercise ability using this information, stores and manages daily, weekly, monthly and yearly exercise results of the user, , an information analysis unit that stores and manages daily, weekly, monthly and yearly exercise capacity improvement levels;

There is disclosed an exercise assistance device comprising an input/output unit for outputting result information and biometric information on the detected specific movement or the analyzed degree of improvement in exercise ability of the user.

However, the prior art has disadvantages in that it is inconvenient to calculate the heart rate and classification class of various exercises, and the results are inaccurate, so that an objective exercise intensity feedback cannot be provided.

Therefore, the present invention was devised to solve the above problems, and implements a CNN (Convolutional Neural Network)-based exercise intensity classification system using pulse waves, and the implemented system measures pulse wave data using a PPG sensor, and then Estimate the exercise intensity by calculating the heart rate, extract the P-peak interval, and input it into the trained artificial neural network to classify three types of exercise intensity into exercise intensity classification classes according to heart rate. By inputting the three classified exercise intensities into the CNN, data on the classified exercise intensities are learned. Then, it is to provide a CNN-based exercise intensity classification system using pulse waves that accumulates exercise data of the user and provides feedback according to the exercise intensity to the user using the learned exercise intensity and the accumulated data.

The present invention relates to a CNN-based exercise intensity classification system using pulse waves, a PPG sensor measuring a pulse wave signal as a hardware terminal, a monitor as an output device; and

It consists of an exercise intensity classification device that is a controller;

The exercise intensity classification device includes a PPG signal input unit for inputting a signal from a PPG sensor, a data processing unit for inputting a learned convolutional neural network (CNN) unit by extracting the P-peak interval, and a convolutional artificial neural network. A (Convolution Neural Network, CNN) unit, an exercise intensity classification unit that classifies exercise intensity into exercise intensity classification classes according to heart rate, an output unit that displays monitoring and graphs, a storage unit that stores data, and a control unit; characterized.

Therefore, the present invention measures the pulse wave signal through the PPG sensor, performs a preprocessing process, extracts the interval of P-peak, and inputs it to the learned CNN artificial neural network to classify the exercise intensity into an exercise intensity classification class according to the heart rate, It has a remarkable effect that it can be applied to a more objective exercise intensity feedback system by accurately calculating the heart rate and classification class of various exercises and linking with the exercise application.

1 is an overall system configuration diagram of the present invention;
2 is a diagram showing the use of the biosignal measuring module of the present invention;
3 is a graph of an example of learning data of the present invention
4 is a configuration diagram of a CNN model of the present invention;
5 is a graph of training accuracy and loss rate of the present invention;

The present invention relates to a CNN-based exercise intensity classification system using pulse waves, a PPG sensor measuring a pulse wave signal as a hardware terminal, a monitor as an output device; and

It consists of an exercise intensity classification device that is a controller;

The exercise intensity classification device is a PPG signal input unit for inputting a signal from the PPG sensor,

A data processing unit that extracts the interval of P-peak and inputs it to the learned Convolution Neural Network (CNN) unit, a Convolution Neural Network (CNN) unit, and an exercise intensity classification class according to heart rate It is characterized in that it consists of; an exercise intensity classifying unit for classifying the exercise intensity, an output unit for monitoring and displaying a graph, a storage unit for storing data, and a control unit.

In addition, the exercise intensity is characterized in that it is classified into three types of exercise intensity.

In addition, the signal measured by the PPG sensor is characterized in that noise included in the original signal measured by the digital filter is removed.

The present invention will be described in detail with reference to the accompanying drawings as follows.

The deep learning-based exercise intensity classification system using pulse waves of the present invention measures the pulse waves in order to enable an effective exercise of the user and inputs them into the deep learning algorithm to classify the exercise intensity according to the pulse waves. According to the classified exercise intensity, the following exercise schedule and feedback are provided. 1 shows a system configuration diagram.

Specifically, the exercise intensity estimation and classification method of the present invention is described in detail, the present invention estimated the exercise intensity using the carbonene formula. The Karbornen formula can estimate exercise intensity using user information of age, exercise intensity, resting heart rate, and resting heart rate. Equation using this is shown in Equation (1).

(1)

(One)

In order to classify the exercise intensity, a total of three classes were classified according to the American Heart Association (AHA). If the exercise intensity was less than 50%, it was classified as low-intensity, if it was 50-70%, it was classified as medium-intensity, and if it was more than 70%, it was classified as high-intensity. Table 1 shows the exercise classification classes.

heart rate exercise intensity 1. Low strength (L) 125 or less 50% or less 2. Medium strength (M) 125 to 150 50 to 70% 3. High Intensity (V) 150 to 170 70 to 85%

Table 1. Exercise Intensity Classification Class

When explaining the data set configuration of the present invention, the present invention measured the pulse wave signal using a biosignal measurement module (P400, Physiolab Co.) to configure the learning data, and included in the original signal measured using a digital filter. Noise was removed. Figure 2 shows the measurement using the biosignal measurement (P400, Physiolab Co.) module.

The heart rate was calculated from the measured data as the number of P-peaks, and the exercise intensity was estimated using the Carbonen formula. When exercising at high intensity, the heartbeat interval is short, and when exercising at low intensity, the interval becomes longer. The interval of the P-peak was calculated using the indexes of both maximum values of the pulse wave waveform with the estimated exercise intensity, and the length of the training data was equally configured by the zero-padding technique. 3 shows an example of the training data.

And the present invention used a convolutional artificial neural network (CNN) model. CNN is an artificial neural network that extracts and classifies the same pattern of image or time series data through convolution operation. RNN (Recurrent Neural Network) that can reflect past learning to current learning using a cyclic structure specialized for repetitive and sequential data learning as a classification algorithm, creates a classification label by itself, distorts the space, and repeats the process of classifying data There are DNNs, etc. that derive the optimal curve line. CNN can learn the characteristics of data in the form of an array by maintaining the relationship between the arrays. Data loss can be prevented because the size can be reduced by capturing the characteristics of the filter, generating data through the process of convolution, and applying a pooling operation that uses only the filter. The implemented CNN model consists of a total of 15 layers: a convolution layer that extracts feature maps of data, a maxpooling layer and dropout layer to prevent overfitting, and a density layer that connects input and output.

Experiments and results of the present invention are described as follows.

For the performance evaluation of the implemented model of the present invention, the accuracy was evaluated using the classification performance evaluation index. To measure training and prediction performance, an error matrix was used to compare predicted values with actual values.

Predicted Values Actual Values True Positive False Negative false positive True Negative

Table 2. Error matrix

Actual Values of the error matrix represent actual values, and Predicted Values represent predicted values. Using this, the accuracy indicating the percentage of correct answers out of the total number of predictions, the precision of those classified as true by the model as being true, and the reproducibility of the model as being classified as true among the true values are expressed in Equation (2). indicated.

(2)

(2)

And in the present invention, a CNN-based exercise intensity classification system using pulse waves is implemented, and the implemented system measures a pulse wave signal through a PPG sensor, performs a preprocessing process, and extracts the interval of P-peak to learn CNN artificial Three types of exercise intensity were classified into exercise intensity classification classes according to heart rate by input into the neural network, and the training was performed by repeating the CNN model 50 times. The performance evaluation of the implemented system was performed and shown in FIG. 5 .

The validation evaluation of the CNN artificial neural network model implemented in the present invention was performed. To evaluate the performance of the implemented system, 50 data for each exercise intensity were measured, and a total of 150 data were input to the artificial neural network to analyze the accuracy. As a result of the experiment, the accuracy of L was 100%, that of M was 96%, and that of V was 100%, confirming a total accuracy of 98.7%. It is considered that some errors occurred because the input data was close to the boundary between the two types of exercise intensity. Table 3 shows the experimental results for each classification class.

Actual L M V Predictions L 50 2 0 M 0 48 0 V 0 0 50 Avg. 100 96 100 98.7

Table 3. Experimental results of classification classes

And the present invention can be classified by adjusting the exercise intensity by multiplying the weight by the data input by input means such as a touch key or a touch pad. , In particular, as the pulse wave is highly related to temperature, such as when the temperature drops, the heart rate drops and the heart takes more pressure to maintain body temperature. Multiply it to give the exercise intensity. In particular, by using the public weather data of the Korea Meteorological Administration, it is possible to receive the temperature according to the region in real time with the API and multiply it by a weighting index to present the appropriate exercise intensity. Similar to temperature, indoor humidity is weighted using a humidity sensor or public data from the Korea Meteorological Administration.

Therefore, in the present invention, a CNN-based exercise intensity classification system using pulse waves was implemented. The implemented system measures the pulse wave signal through the PPG sensor and performs a pre-processing process. After that, the interval of P-peak is extracted and input into the trained CNN artificial neural network to classify three types of exercise intensity into exercise intensity classification classes according to heart rate. As a result of the experiment of the implemented system, the accuracy of 98.7% was confirmed. In addition, it is possible to apply to a more objective exercise intensity feedback system by adding the heart rate and classification class of various exercises and linking with the exercise application.

Claims (3)

A PPG sensor that measures a pulse wave signal, which is a hardware terminal, and a monitor that is an output device; and
It consists of an exercise intensity classification device that is a controller;
The exercise intensity classification device is a PPG signal input unit for inputting a signal from the PPG sensor,
A data processing unit that extracts the interval of P-peak and inputs it to the learned Convolution Neural Network (CNN) unit;
Convolution Neural Network (CNN) unit,
Exercise intensity classification unit that classifies exercise intensity into exercise intensity classification classes according to heart rate;
Output unit to display monitoring, graph,
a storage unit for storing data;
And a control unit; CNN-based exercise intensity classification system using pulse waves, characterized in that consisting of The CNN-based exercise intensity classification system using pulse waves according to claim 1, wherein the exercise intensity is classified into three types of exercise intensities. [Claim 3] The CNN-based exercise intensity classification system using pulse waves according to claim 2, wherein noise included in the original signal measured by a digital filter is removed from the signal measured by the PPG sensor.

Images