Pulse condition mode recognition method of wearable pulse diagnosis instrument

Abstract

The invention discloses a pulse condition mode recognition method of a wearable pulse diagnosis instrument, which comprises the following steps: carrying out N > = 1 pairwise merging on pulse condition information data training sets with M classification requirements, and reducing M classifications into (M-N) classifications; forming a sub-data set Sn; dividing a training set and a verification set, performing classification model training on (M-N) classifications, and selecting a data prediction model X with the highest prediction rate; calculating a prediction result Y according to the prediction model X; the method comprises the following steps: carrying out random sorting on Sn data, generating i different sequences of Sn, dividing a training set and a verification set on each sorting result, carrying out classification model training on Sn (i), and selecting i / 2 models with high prediction rates; performing model result prediction on the i / 2 Sn (i), calculating the i / 2 prediction results by adopting a logical operation method, and obtaining a final Sn prediction result Ys by adopting a non-A, namely B dichotomy; and substituting the Ys prediction result into the previous Y prediction result, so that the Y prediction result reaches the requirements of M classifications again.

Description

Technical Field

[0001] The invention relates to the technical field of wearable pulse diagnosis instruments, and in particular to a pulse pattern recognition method of a wearable pulse diagnosis instrument. Background Art

[0002] The usual pulse classification data model uses SVM / LDA / Random Forest / Logistic Regression / Bayesian algorithms to train sample data sets to generate a classification model. When in use, the unknown samples are directly predicted and classified using a classification model. This model training method and recognition strategy can be applied to samples with high data differentiation, but for sample data with similar data and low differentiation, it is impossible to achieve a high accuracy prediction and cannot be put into practical use. Summary of the invention

[0003] The purpose of the present invention is to provide a pulse pattern recognition method for a wearable pulse diagnosis instrument, in order to solve the technical problems existing in the background technology.

[0004] In order to achieve the above object, the present invention adopts the following technical solutions:

[0005] A pulse pattern recognition method for a wearable pulse diagnosis instrument, comprising:

[0006] For the pulse information data training set with M classification requirements, N≥1 pairs of merges are performed, and M classifications are reduced to (MN) classifications; a sub-data set Sn is formed;

[0007] Divide the training set and the verification set, train the classification model for (MN) categories, and select the data prediction model X with the highest prediction rate;

[0008] Based on the prediction model X, calculate the prediction result Y;

[0009] Randomly sort the data of Sn to generate i different sequences of Sn. Based on each sorting result, divide the training set and the verification set, train the classification model for Sn(i), and select i / 2 models with high prediction rates.

[0010] The model results of i / 2 Sn(i) are predicted, and the i / 2 prediction results are calculated by the logical operation method, and the binary method of either A or B is used to obtain the final Sn prediction result Ys;

[0011] Substitute the prediction result of Ys into the previous prediction result of Y so that the prediction result of Y can meet the requirements of M classifications again.

[0012] In some embodiments, the model X is the projection of samples of the pulse information data training set on a straight line, the positions of the projections of samples of the same category on the straight line are close, and the positions of the projections of samples of different categories on the straight line are separated.

[0013] In some embodiments, the category of the unknown sample is determined according to the projection point position of the unknown sample.

[0014] The pulse pattern recognition method of a wearable pulse diagnosis instrument provided in this application has beneficial effects including but not limited to:

[0015] The present invention utilizes the characteristics of a wearable pulse diagnosis instrument with multiple sensor data sources, combines the acquisition time, generates multiple sub-model composite calculations, performs binary sub-model training for easily confused pulse types, and proposes a pulse classification model training method and recognition strategy, which effectively solves the problems existing in the background technology. The present invention can effectively improve the prediction accuracy of the model in various scenarios where data classification and recognition are required. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a usage status diagram of a wearable pulse monitor product made with an array pressure sensor.

[0017] Figure 2 This is a flow chart of the method of this application. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solutions and advantages of the present application clearer, the technical solutions in the embodiments of the present application will be described in more detail below in conjunction with the drawings in the preferred embodiments of the present application. In the drawings, the same or similar reference numerals throughout represent the same or similar parts or parts with the same or similar functions. The described embodiments are part of the embodiments of the present application, not all of the embodiments. The embodiments described below with reference to the drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limitations on the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present application.

[0019] The following will be combined Figure 1-2 It is worth noting that the following embodiments are only used to explain the present application and do not constitute a limitation on the present application.

[0020] Example 1

[0021] The present invention adopts array pressure sensors distributed on fingertips to design a wearable pulse diagnosis instrument, specifically including array pressure sensors corresponding to three different positions of Cun, Guan and Chi, wherein the number of pressure sensors on each fingertip is F≥1; in the pulse recognition process of the pulse diagnosis instrument, the pressure sensor on each fingertip collects pulse wave data for a certain period of time T≥1s, and the pulse wave information collected each time contains pulse information of T≥1s on F≥1 pressure sensors, and a pattern recognition method is used to identify the pulse type through the pulse information.

[0022] Generally, for the pulse recognition model, the more types need to be classified, the worse the model accuracy. The present invention first reduces the number of types that need to be classified by 1 / 2 by classifying them in pairs, and this method can be nested for use, so that when the number of classified types is small, the model recognition accuracy will be very high.

[0023] Then, separate classification model training is performed on the separate small data sets classified pairwise, especially tending to use linear model algorithms for training. Then, in order to improve the accuracy of the classification model of the small data sets classified pairwise, a training method of randomly distributing the data sets, randomly training, and generating multiple sub-models is adopted.

[0024] Then, multiple sub-models are sorted according to model accuracy, and sub-models with high model accuracy are selected. With only two classification types, either A or B, multiple sub-models are superimposed with logical operations to obtain all A prediction results, and the remaining are B prediction results, which can effectively improve the recognition accuracy of the classification model of small data sets classified in pairs. Finally, the classification results of the small data sets classified in pairs are iterated step by step to the classification results of the large model at the previous level, and finally the overall classification prediction results are obtained.

[0025] In order to realize the above technical solution, the present invention comprises the following steps:

[0026] 1. For the pulse information data training set with M classification requirements, N ≥ 1 pairs of merges are performed, and M classifications are reduced to (MN) classifications;

[0027] 2. Appropriately divide the training set and the verification set, train the classification model for (MN) classifications, and select the data prediction model X with the highest prediction rate. Here, the model can be the projection of the samples of the pulse information data training set on a straight line. The positions of the projections of the same type of samples on the straight line are as close as possible, and the positions of the projections of the samples of different categories on the straight line are as separated as possible. Then determine the category of the unknown sample according to the position of the projection point of the unknown sample; it can also be other classification models, which will not be repeated here.

[0028] 3. Based on the prediction model X, calculate the prediction result Y;

[0029] 3. For the sub-data sets Sn that are merged two by two, randomly sort the data of Sn to generate i different sequences of Sn. Based on each sorting result, appropriately divide the training set and the verification set, train the classification model for Sn(i), and select i models with higher prediction rates. Here, the linear classification model is preferred;

[0030] 4. Predict the model results for i Sn(i), calculate the i prediction results using the logical operation method, and use the binary method of either A or B to obtain the final Sn prediction result Ys;

[0031] 5. Substitute the prediction result of Ys into the previous prediction result of Y so that the prediction result of Y meets the requirements of M classifications again.

[0032] When training the classification model, the identified categories are merged in pairs to reduce the number of categories and improve the overall prediction accuracy. When training a data set with only two categories, the data set is randomly sorted to divide the training subset and the verification subset to generate multiple different data sets, and then multiple different prediction models are generated. The results of the prediction models are logically operated to obtain the final prediction results of the two-classification model.

[0033] The present invention is further described in detail with examples given below.

[0034] As attached Figure 1 , the wearable pulse monitor product 400 made of array pressure sensor, in accordance with the standard requirements of traditional Chinese medicine pulse diagnosis, collects pulse information at three different positions of the brachial artery at 401 cun, 402 guan, and 403 chi, with 3 pressure sensors placed at each position, and 9 pulse signals can be obtained at a time, the acquisition time is 60s, the pulse frequency is 60 times / minute, and 540 pulse cycle signals at different positions can be obtained. In order to train an accurate and reliable pulse recognition model, pulse information of about 10,000 different samples is collected, and according to the requirements of 8 types of pulse recognition, floating pulse, sinking pulse, weak pulse, strong pulse, slow pulse, rapid pulse, slippery pulse, and string pulse need to be identified respectively.

[0035] For this purpose, as attached Figure 2 Do the following:

[0036] 1. For the pulse data training set with 8 pulse classification requirements, 4 pairs of merges are performed, and the 8 classifications are reduced to 4 classifications;

[0037] 2. Appropriately divide the training set and the verification set, train the classification model for the four categories, and select the data prediction model X with the highest prediction rate;

[0038] 3. Based on the prediction model X, calculate the prediction result Y

[0039] 3. For the two-by-two merged sub-data sets S1-S4, randomly sort the data sets of S1-S4, generate 100 different sequence combinations in the data sets of S1-S4, and appropriately divide the training set and the verification set in each sorting combination, train the classification model for S1(i)-S4(i), and select 50 models with higher prediction rates. Here, the linear classification model is preferred;

[0040] 4. In the sub-data sets S1-S4, the model results are predicted for the 50 selected Sn(i) respectively, and the 50 prediction results are calculated using the OR operation, and the binary method of either A or B is used to obtain the final Sn prediction result Yn;

[0041] 5. Substitute the prediction result of Yn into the previous prediction result of Y so that the prediction result of Y can meet the requirements of 8 categories again.

[0042] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A pulse pattern recognition method for a wearable pulse diagnosis instrument, characterized in that: include: For the pulse information data training set with M classification requirements, N≥1 pairs of merges are performed, and M classifications are reduced to (MN) classifications; a sub-data set Sn is formed; Divide the training set and the verification set, train the classification model for (MN) categories, and select the data prediction model X with the highest prediction rate; Based on the prediction model X, calculate the prediction result Y; Randomly sort the data of Sn to generate i different sequences of Sn. Based on each sorting result, divide the training set and the verification set, train the classification model for Sn(i), and select i / 2 models with high prediction rates. The model results of i / 2 Sn(i) are predicted, and the i / 2 prediction results are calculated by the logical operation method, and the binary method of either A or B is used to obtain the final Sn prediction result Ys; Substitute the prediction result of Ys into the previous prediction result of Y so that the prediction result of Y can meet the requirements of M classifications again.

2. The pulse pattern recognition method of a wearable pulse diagnosis instrument according to claim 1, characterized in that: The model X is the projection of the samples of the pulse information data training set on a straight line. The positions of the projections of the same type of samples on the straight line are close, and the positions of the projections of the samples of different types on the straight line are separated.

3. The pulse pattern recognition method of a wearable pulse diagnosis instrument according to claim 2, characterized in that: The category of the unknown sample is determined according to the projection point position of the unknown sample.

Images