AI physique diagnosis software analysis system

Abstract

The invention relates to the technical field of medical diagnosis systems, and discloses an AI physique diagnosis software analysis system which comprises an MCU module used for sending commands to execution modules and receiving and processing data signals of the execution modules to achieve diagnosis operation of the analysis system; the data acquisition module is used for collecting facial images and pulse information physiological indexes of the patient through an acquisition medium; the image processing module is used for analyzing facial features of the patient, the facial features comprise complexion, tongue coating and tongue quality, and corresponding evaluation data are generated through the obtained facial feature information. According to the AI physique diagnosis software analysis system, the accuracy and efficiency of traditional Chinese medicine physique identification are remarkably improved, the subjectivity and professional threshold are reduced, remarkable social and economic benefits are generated, and the system can more accurately evaluate the physique and display experimental data by using a deep learning algorithm and multi-modal data analysis.

Description

Technical Field

[0001] The present invention relates to the technical field of medical diagnosis systems, and in particular to an AI physical diagnosis software analysis system. Background Art

[0002] With the continuous emergence of emerging technologies, such as big data, artificial intelligence, the Internet of Things, virtual reality, and augmented reality, the research on constitution identification methods also needs to keep up with the development trend of the information society, and constantly improve and innovate, moving towards objectivity, standardization, convenience, and intelligence, so as to better meet the needs of the times. Tongue diagnosis is an important part of TCM inspection and an important embodiment of the TCM principle of external diagnosis and internal diagnosis. TCM manual identification of constitution mainly involves four examinations, including inspection, auscultation, questioning, and palpation. The specific steps are as follows:

[0003] (1) Observation: observe the patient's complexion, shape, tongue coating and tongue quality;

[0004] (2) Olfaction: judging the body's condition by listening to sounds and smells;

[0005] (3) Interview: Ask about symptoms, medical history, lifestyle, eating habits, etc.;

[0006] (4) Palpation: Evaluate the characteristics of the pulse and judge the condition of Qi and blood through pulse diagnosis;

[0007] However, there are the following problems in TCM's artificial identification of human body constitution:

[0008] A. Highly subjective: TCM diagnosis relies on the physician’s experience and intuition, which may lead to inconsistent judgments.

[0009] B. High time cost: The traditional identification process is cumbersome and time-consuming.

[0010] C. High professional knowledge requirements: It requires extensive knowledge of traditional Chinese medicine and clinical experience, which is difficult for beginners to master;

[0011] Up to now, there are some TCM constitution identification documents based on artificial intelligence for tongue characteristics, such as publication numbers CN118570539A and CN118645250A. However, the above-mentioned methods for constitution identification based on artificial intelligence are single, either through a separate tongue analysis or a separate pulse analysis, resulting in large errors in the analysis results, and there are no relevant reports on a system that simultaneously collects and combines the two for analysis. Summary of the invention

[0012] The purpose of the present invention is to provide an AI constitution diagnosis software analysis system. By using deep learning algorithms and multimodal data analysis, the system can more accurately evaluate constitution. It aims to solve the problem that the existing methods for realizing constitution identification based on artificial intelligence are single, either through a separate tongue analysis or a separate pulse analysis, resulting in large errors in the analysis results.

[0013] The present invention is implemented in this way: an AI physical constitution diagnosis software analysis system comprises:

[0014] The MCU module is used to implement the diagnostic operation of the analysis system by sending commands to each execution module and receiving and processing data signals from each execution module;

[0015] Data acquisition module: used to collect facial images, pulse information and physiological indicators of patients through acquisition media;

[0016] Image processing module: used to analyze the facial features of the patient, including complexion, tongue coating and tongue quality, and generate corresponding evaluation data through the acquired facial feature information;

[0017] Pulse analysis module: used to extract features from the pulse information physiological index data collected by the data acquisition module, and analyze the attribute feature evaluation data of the pulse;

[0018] Data analysis module: used to generate corresponding evaluation data from the acquired facial feature information and the pulse attribute feature evaluation data for combined analysis, classify the acquired physical data, and output the physical evaluation results.

[0019] Furthermore, the data acquisition module collects facial features of the patient for completing tongue image modeling detection and analysis, including:

[0020] The acquisition medium includes a camera, through which the facial features of the patient are collected;

[0021] Use a median filter or a Gaussian filter to remove image noise and enhance contrast. The Gaussian filter formula is:

[0022]

[0023] Segment the tongue surface of the facial image, automatically calculate the optimal threshold, and segment the tongue area. The segmentation formula is:

[0024]

[0025] After the tongue area is segmented, Canny edge detection is used to extract the tongue edge and determine the tongue shape;

[0026] Perform feature extraction on the segmented tongue area. The extracted features include color features and texture features. The color feature is to extract the RGB and HSV color values ​​of the tongue. The texture feature is to calculate the gray level co-occurrence matrix and extract the contrast, correlation, and uniformity features.

[0027] The extracted color features and texture features are trained, divided into training sets and test sets, and the model performance is evaluated using cross-validation to generate a feature model. The generated feature model formula is:

[0028] f(x)=w T φ(x)+b;

[0029] The generated feature models are classified and evaluated, and feature selection and model parameters are adjusted according to the evaluation results.

[0030] Furthermore, the color feature is to extract the RGB and HSV color values ​​of the tongue, and the calculated average color value is:

[0031]

[0032] The texture feature is to calculate the gray level co-occurrence matrix, extract the contrast, correlation, and uniformity features, and the gray level co-occurrence matrix formula is:

[0033]

[0034] Furthermore, the data acquisition module collects the patient's pulse information physiological indicators for completing pulse condition modeling detection and analysis, including:

[0035] The collection medium also includes a pulse sensor, through which the pulse information physiological indexes are collected;

[0036] Use bandpass filters to remove low-frequency noise and high-frequency interference and normalize;

[0037] Extracting features of pulse information physiological indicators, the extracted features include time domain features and frequency domain features;

[0038] Model the feature extraction content to generate training, and divide the data set for training and verification;

[0039] The model performance was evaluated using confusion matrix and F1 score, and feature extraction and model parameters were optimized based on the evaluation results.

[0040] Furthermore, a bandpass filter is used to remove low-frequency noise and high-frequency interference and normalization processing is performed, including:

[0041] The frequency response function of the bandpass filter is:

[0042]

[0043] Where, s is the signal frequency, ωc is the cutoff frequency;

[0044] The normalization processing formula is:

[0045]

[0046] Among them, Xnorm is the normalized value, X is the original value, Xmin is the minimum value, and Xmax is the maximum value.

[0047] Furthermore, feature extraction is performed on the facial image and the pulse information physiological index, including:

[0048] Calculate the period, peak value and waveform width of the pulse wave. The period is calculated as follows:

[0049]

[0050] Where T is the pulse wave period and f represents the pulse wave frequency;

[0051] The formula for calculating the spectral characteristics of the pulse wave using FFT is:

[0052]

[0053] where x n is a discrete signal, and e is a rotation factor.

[0054] Furthermore, the feature extraction content is modeled and trained, and the data set is divided for training and verification, including:

[0055] Select a machine learning algorithm and combine the time domain features with the frequency domain features for training. First, model the joint probability distribution p(x|y), then calculate the conditional probability p(y|x) that the sample belongs to each class, that is, the class posterior probability. Use this value to complete the classification. The model training formula is:

[0056]

[0057] Among them, p(x|y) is the joint probability density function, and p(x) is the marginal density function of the sample input vector x.

[0058] Furthermore, the data analysis module generates corresponding evaluation data from the acquired facial feature information and combines and analyzes the attribute feature evaluation data of the pulse, classifies the acquired physical fitness data, and outputs physical fitness evaluation results, including:

[0059] Complete tongue image model output and pulse image model output;

[0060] The feature vectors of tongue image and pulse image are concatenated to form a new feature vector Vcombined, where:

[0061]

[0062] The outputs of the two models are combined, and the performance of the combined model is evaluated using indicators such as accuracy, F1 score, and ROC curve. Finally, the model parameters and fusion strategy are continuously adjusted based on patient feedback and new data to optimize the diagnostic results and complete the output.

[0063] Furthermore, the way of combining the outputs of the two models includes weighted average, voting method or stacking model, and the weighted average formula is:

[0064] C final =w1·C tongue +w2·C pulse

[0065] Among them, w1 and w2 are weight coefficients, which can be determined by cross-validation.

[0066] Furthermore, the camera and pulse sensor of the acquisition medium are respectively a 1080p high-resolution camera and a photoplethysmography sensor, and the sampling frequency of the photoplethysmography sensor needs to be greater than 500 Hz to ensure the accuracy and integrity of the signal.

[0067] Compared with the prior art, the AI ​​physical fitness diagnosis software analysis system provided by the present invention has the following beneficial effects:

[0068] 1. By providing an AI constitution diagnosis software analysis system, the accuracy and efficiency of TCM constitution identification have been significantly improved, the subjectivity and professional threshold have been reduced, and significant social and economic benefits have been generated. By using deep learning algorithms and multimodal data analysis, the system can more accurately evaluate constitution. Experimental data show that compared with the judgment of traditional Chinese medicine practitioners, the constitution identification accuracy of the AI ​​system has increased by about 20%, reaching more than 95%; the AI ​​system eliminates the influence of human factors, making the constitution identification results more objective and consistent. Compared with traditional methods, the consistency of evaluation between doctors has increased from the original 70% to 90%; and the system can continuously improve its identification ability through continuous data accumulation and machine learning optimization algorithms. Through the analysis of past data, the accuracy of the system can be improved by about 5%-10% each year.

[0069] 2. By improving the efficiency and accuracy of TCM constitution identification, promoting the modernization and popularization of traditional Chinese medicine, more patients can receive accurate TCM health management. It is estimated that by shortening the diagnosis time and improving the reception capacity, each doctor can increase the income of the hospital or clinic by about 20% each year. At the same time, the improvement of patient satisfaction and treatment effect will also promote the patient's continued medical treatment. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 This is a schematic diagram of the structure of the AI ​​physical constitution diagnosis software analysis system proposed by the present invention;

[0071] Figure 2 This is a data processing logic flow chart of the head portrait processing module in the AI ​​physical diagnosis software analysis system proposed by the present invention;

[0072] Figure 3 This is a data processing logic flow chart of the pulse analysis module in the AI ​​physical constitution diagnosis software analysis system proposed by the present invention;

[0073] Figure 4 This is a data processing logic flow chart of the data analysis module in the AI ​​physical constitution diagnosis software analysis system proposed in the present invention. DETAILED DESCRIPTION

[0074] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0075] The implementation of the present invention is described in detail below in conjunction with specific embodiments.

[0076] The same or similar numbers in the drawings of this embodiment correspond to the same or similar parts; in the description of the present invention, it should be understood that if the terms "upper", "lower", "left", "right" and the like indicate directions or positional relationships based on the directions or positional relationships shown in the drawings, it is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction. Therefore, the terms describing the positional relationship in the drawings are only used for illustrative purposes and cannot be understood as limitations on the present invention. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.

[0077] Reference Figure 1As shown, the AI ​​constitution diagnosis software analysis system includes: an MCU module, which is used to realize the diagnostic operation of the analysis system by sending commands to each execution module and receiving and processing the data signals of each execution module; a data acquisition module: used to collect the patient's facial image and pulse information physiological indicators through the acquisition medium; an image processing module: used to analyze the patient's facial features, which include complexion, tongue coating and tongue quality, and generate corresponding evaluation data through the acquired facial feature information; a pulse analysis module: used to extract features from the pulse information physiological indicator data collected by the data acquisition module, and analyze the attribute feature evaluation data of the pulse; a data analysis module: used to combine and analyze the corresponding evaluation data generated by the acquired facial feature information and the attribute feature evaluation data of the pulse, classify the acquired constitution data, and output the constitution evaluation result. The advantages of artificial intelligence technology in identifying the medical constitution of the human body in this technical solution are:

[0078] Objectivity: AI can provide more objective assessments through image processing and data analysis, reducing human error.

[0079] Efficiency: AI technology can quickly analyze large amounts of data, saving time.

[0080] Repeatability: AI algorithms can maintain consistency under the same conditions, reducing differences in judgment between different doctors.

[0081] Data accumulation and learning: AI can continuously learn and optimize, combined with big data analysis, to provide more precise recommendations for individualized treatment.

[0082] Reference Figure 2 As shown, the data acquisition module collects the patient's facial features for tongue image modeling, detection and analysis, including:

[0083] The acquisition medium includes a camera, through which the facial features of the patient are collected;

[0084] Use a median filter or a Gaussian filter to remove image noise and enhance contrast. The Gaussian filter formula is:

[0085]

[0086] Segment the tongue surface of the facial image, automatically calculate the optimal threshold, and segment the tongue area. The segmentation formula is:

[0087]

[0088] After the tongue area is segmented, Canny edge detection is used to extract the tongue edge and determine the tongue shape;

[0089] Feature extraction is performed on the segmented tongue area. The extracted features include color features and texture features. The color feature is to extract the RGB and HSV color values ​​of the tongue. The texture feature is to calculate the gray-level co-occurrence matrix and extract the contrast, correlation, and uniformity features.

[0090] The extracted color features and texture features are trained, divided into training sets and test sets, and the model performance is evaluated using cross-validation to generate a feature model. The generated feature model formula is:

[0091] f(x)=w T φ(x)+b;

[0092] The generated feature models are classified and evaluated, and feature selection and model parameters are adjusted according to the evaluation results.

[0093] The color feature is to extract the RGB and HSV color values ​​of the tongue, and the calculated average color value is:

[0094]

[0095] The texture feature is to calculate the gray level co-occurrence matrix, extract the contrast, correlation, and uniformity features, and the formula for calculating the gray level co-occurrence matrix is:

[0096]

[0097] Reference Figure 3 As shown, the data acquisition module collects the patient's pulse information and physiological indicators for pulse modeling detection and analysis, including:

[0098] The collection medium also includes a pulse sensor, through which the pulse information and physiological indicators are collected;

[0099] Use bandpass filters to remove low-frequency noise and high-frequency interference and normalize;

[0100] Extracting features of pulse information physiological indicators, the extracted features include time domain features and frequency domain features;

[0101] Model the feature extraction content to generate training, and divide the data set for training and verification;

[0102] The model performance was evaluated using confusion matrix and F1 score, and feature extraction and model parameters were optimized based on the evaluation results.

[0103] Use bandpass filters to remove low-frequency noise and high-frequency interference and normalize, including:

[0104] The frequency response function of the bandpass filter is:

[0105]

[0106] Where, s is the signal frequency, ωc is the cutoff frequency;

[0107] The normalization formula is:

[0108]

[0109] Among them, Xnorm is the normalized value, X is the original value, Xmin is the minimum value, and Xmax is the maximum value.

[0110] Feature extraction of facial images, pulse information and physiological indicators, including:

[0111] Calculate the period, peak value and waveform width of the pulse wave. The period is calculated as follows:

[0112]

[0113] Where T is the pulse wave period and f represents the pulse wave frequency;

[0114] The formula for calculating the spectral characteristics of the pulse wave using FFT is:

[0115]

[0116] where x n is a discrete signal, and e is a rotation factor.

[0117] Model the feature extraction content and generate training, and divide the data set for training and verification, including:

[0118] Select a machine learning algorithm and combine the time domain features with the frequency domain features for training. First, model the joint probability distribution p(x|y), then calculate the conditional probability p(y|x) that the sample belongs to each class, that is, the class posterior probability. Use this value to complete the classification. The model training formula is:

[0119]

[0120] Among them, p(x|y) is the joint probability density function, and p(x) is the marginal density function of the sample input vector x.

[0121] Reference Figure 4 As shown, the data analysis module generates corresponding evaluation data from the acquired facial feature information and combines and analyzes the attribute feature evaluation data of the pulse, classifies the acquired physical data, and outputs the physical evaluation results, including:

[0122] Complete tongue image model output and pulse image model output;

[0123] The feature vectors of tongue image and pulse image are concatenated to form a new feature vector Vcombined, where:

[0124]

[0125] The outputs of the two models are combined, and the performance of the combined model is evaluated using indicators such as accuracy, F1 score, and ROC curve. Finally, the model parameters and fusion strategy are continuously adjusted based on patient feedback and new data to optimize the diagnostic results and complete the output.

[0126] Ways to combine the outputs of two models include weighted average, voting method or stacking model. The weighted average formula is:

[0127] C final =w1·C tongue +w2·C pulse

[0128] Among them, w1 and w2 are weight coefficients, which can be determined by cross-validation.

[0129] The camera and pulse sensor of the acquisition medium are respectively a 1080p high-resolution camera and a photoelectric volumetric pulse wave sensor. The sampling frequency of the photoelectric volumetric pulse wave sensor needs to be greater than 500 Hz to ensure the accuracy and integrity of the signal.

[0130] The present invention significantly improves the accuracy and efficiency of TCM constitution identification by introducing artificial intelligence technology, reduces subjectivity and professional threshold, and produces significant social and economic benefits. These advantages make the present invention have important application prospects and market value in the field of TCM, and specifically have the following beneficial effects:

[0131] 1. Improve recognition accuracy

[0132] By using deep learning algorithms and multimodal data analysis, the system can more accurately assess physical constitution. Experimental data show that compared with the judgment of traditional Chinese medicine practitioners, the AI ​​system's physical constitution identification accuracy has increased by about 20%, reaching more than 95%.

[0133] 2. Reduce subjectivity

[0134] The AI ​​system eliminates the influence of human factors, making the results of constitution identification more objective and consistent. Compared with traditional methods, the consistency of evaluation between doctors has increased from 70% to 90%.

[0135] 3. Shorten the diagnosis time

[0136] The traditional Chinese medicine identification process usually takes 5 to 10 minutes, while the AI-based system can complete the physical constitution assessment within 3 minutes, significantly improving the efficiency of diagnosis. This efficiency improvement allows doctors to receive more patients at the same time.

[0137] 4. Lower professional threshold

[0138] The system provides an intuitive and easy-to-use constitution identification tool for beginners and non-professionals, reducing their reliance on traditional Chinese medicine knowledge. Preliminary surveys show that non-professionals using the system can achieve identification capabilities similar to those of professional physicians after a brief training.

[0139] 5. Data accumulation and optimization

[0140] The system can continuously improve its recognition ability through continuous data accumulation and machine learning optimization algorithms. Through the analysis of past data, the accuracy of the system can be improved by about 5%-10% each year.

[0141] 6. Social Benefits

[0142] By improving the efficiency and accuracy of TCM constitution identification, we will promote the modernization and popularization of traditional Chinese medicine, and enable more patients to receive accurate TCM health management.

[0143] 7. Economic Benefits

[0144] It is expected that by shortening diagnosis time and improving reception capacity, each doctor can increase the annual revenue of the hospital or clinic by about 20%. At the same time, the improvement of patient satisfaction and treatment effect will also promote the continuous medical treatment of patients.

[0145] The present invention uses artificial intelligence technology, combined with pulse analysis and tongue analysis, to achieve automatic identification of TCM constitutions; solve the problems of subjectivity, high time cost and professional knowledge threshold in traditional TCM constitution identification; and also solve the problem that the existing manual constitution identification methods are single, and lack accuracy and reliability. The technical solution mainly includes the following parts:

[0146] 1. System architecture

[0147] Data acquisition module: collects patients' facial images, pulse information and other physiological indicators through cameras and sensors.

[0148] Image processing module: Use image recognition technology to analyze facial features, including complexion, tongue coating and tongue quality, and generate corresponding evaluation data.

[0149] Pulse analysis module: obtains pulse data through the pulse sensor and uses algorithms to analyze the characteristics of the pulse (intensity, frequency, etc.).

[0150] Data analysis module: Combine the collected data, use machine learning algorithms to classify physical constitutions, and output physical constitution assessment results.

[0151] 2. Technical features

[0152] Deep learning algorithm: Convolutional neural network (CNN) is used for image analysis to improve the accuracy of facial diagnosis.

[0153] Multimodal data fusion: combines image data and pulse data to enhance the comprehensiveness of physical constitution identification.

[0154] Real-time feedback mechanism: Provides instant physical assessment report and recommends corresponding conditioning plans.

[0155] 3. Invention

[0156] Objectivity and efficiency: AI technology is used to reduce human errors and achieve fast and accurate physical identification.

[0157] Repeatability: The consistency of the AI ​​algorithm ensures that different patients can obtain similar evaluation results under the same conditions.

[0158] Learning and optimization: The system can continuously absorb new data, optimize algorithms, and improve recognition accuracy.

[0159] In this embodiment, the entire operation process can be controlled by a computer, and with the addition of a PLC, etc., automated operation control can be achieved. In each operation link, sensors can be set up to provide signal feedback to achieve sequential execution of steps. These are all common knowledge of current automated control and will not be described in detail in this embodiment.

[0160] 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. AI physical constitution diagnosis software analysis system, characterized by: include: The MCU module is used to implement the diagnostic operation of the analysis system by sending commands to each execution module and receiving and processing data signals from each execution module; Data acquisition module: used to collect facial images, pulse information and physiological indicators of patients through acquisition media; Image processing module: used to analyze the facial features of the patient, including complexion, tongue coating and tongue quality, and generate corresponding evaluation data through the acquired facial feature information; Pulse analysis module: used to extract features from the pulse information physiological index data collected by the data acquisition module, and analyze the attribute feature evaluation data of the pulse; Data analysis module: used to generate corresponding evaluation data from the acquired facial feature information and the pulse attribute feature evaluation data for combined analysis, classify the acquired physical data, and output the physical evaluation results.

2. The AI ​​physical constitution diagnosis software analysis system according to claim 1, characterized in that: The data acquisition module collects the patient's facial features for tongue image modeling detection and analysis, including: The acquisition medium includes a camera, through which the facial features of the patient are collected; Use a median filter or a Gaussian filter to remove image noise and enhance contrast. The Gaussian filter formula is: Segment the tongue surface of the facial image, automatically calculate the optimal threshold, and segment the tongue area. The segmentation formula is: After the tongue area is segmented, Canny edge detection is used to extract the tongue edge and determine the tongue shape; Perform feature extraction on the segmented tongue area. The extracted features include color features and texture features. The color feature is to extract the RGB and HSV color values ​​of the tongue. The texture feature is to calculate the gray level co-occurrence matrix and extract the contrast, correlation, and uniformity features. The extracted color features and texture features are trained, divided into training sets and test sets, and the model performance is evaluated using cross-validation to generate a feature model. The generated feature model formula is: f(xx))=wφ(x)+b The generated feature models are classified and evaluated, and feature selection and model parameters are adjusted according to the evaluation results.

3. The AI ​​physical diagnosis software analysis system according to claim 2, characterized in that: The color feature is to extract the RGB and HSV color values ​​of the tongue, and the calculated average color value is: The texture feature is to calculate the gray level co-occurrence matrix, extract the contrast, correlation, and uniformity features, and the gray level co-occurrence matrix formula is:

4. The AI ​​physical diagnosis software analysis system according to claim 3, characterized in that: The data acquisition module collects the patient's pulse information physiological indicators for completing pulse modeling detection and analysis, including: The collection medium also includes a pulse sensor, through which the pulse information physiological indexes are collected; Use bandpass filters to remove low-frequency noise and high-frequency interference and normalize; Extracting features of pulse information physiological indicators, the extracted features include time domain features and frequency domain features; Model the feature extraction content to generate training, and divide the data set for training and verification; The model performance was evaluated using confusion matrix and F1 score, and feature extraction and model parameters were optimized based on the evaluation results.

5. The AI ​​physical constitution diagnosis software analysis system according to claim 4, characterized in that: Use bandpass filters to remove low-frequency noise and high-frequency interference and normalize, including: The frequency response function of the bandpass filter is: Where, s is the signal frequency, ωc is the cutoff frequency; The normalization processing formula is: Among them, Xnorm is the normalized value, X is the original value, Xmin is the minimum value, and Xmax is the maximum value.

6. The AI ​​physical diagnosis software analysis system according to claim 5, characterized in that: Feature extraction of facial images, pulse information and physiological indicators, including: Calculate the period, peak value and waveform width of the pulse wave. The period is calculated as follows: Where T is the pulse wave period and f represents the pulse wave frequency; The formula for calculating the spectral characteristics of the pulse wave using FFT is: where x n is a discrete signal, and e is a rotation factor.

7. The AI ​​physical diagnosis software analysis system according to claim 6, characterized in that: Model the feature extraction content and generate training, and divide the data set for training and verification, including: Select a machine learning algorithm and combine the time domain features with the frequency domain features for training. First, model the joint probability distribution p(x|y), then calculate the conditional probability p(y|x) that the sample belongs to each class, that is, the class posterior probability. Use this value to complete the classification. The model training formula is: Among them, p(x|y) is the joint probability density function, and p(x) is the marginal density function of the sample input vector x.

8. The AI ​​physical diagnosis software analysis system according to claim 1, characterized in that: The data analysis module generates corresponding evaluation data from the acquired facial feature information and combines and analyzes the attribute feature evaluation data of the pulse, classifies the acquired physical data, and outputs physical evaluation results, including: Complete tongue image model output and pulse image model output; The feature vectors of tongue image and pulse image are concatenated to form a new feature vector Vcombined, where: The outputs of the two models are combined, and the performance of the combined model is evaluated using indicators such as accuracy, F1 score, and ROC curve. Finally, the model parameters and fusion strategy are continuously adjusted based on patient feedback and new data to optimize the diagnostic results and complete the output.

9. The AI ​​physical diagnosis software analysis system according to claim 8, characterized in that: Ways to combine the outputs of the two models include weighted average, voting method or stacking model, and the weighted average formula is: Cfinal=w1Ctongue+w2·Cpulse Among them, w1 and w2 are weight coefficients, which can be determined by cross-validation.

10. The AI ​​physical constitution diagnosis software analysis system according to claim 4, characterized in that: The camera and pulse sensor of the acquisition medium are respectively a 1080p high-resolution camera and a photoelectric volumetric pulse wave sensor. The sampling frequency of the photoelectric volumetric pulse wave sensor needs to be greater than 500 Hz to ensure the accuracy and integrity of the signal.

Images