Detection method and system of multimodal physiological indicators for pet health assessment

Abstract

The present application discloses a method and system for detecting multimodal physiological indicators for pet health assessment, the method comprising: collecting body temperature data, heart rate data, respiration data, activity data and sleep data to obtain a multimodal physiological indicator data set; using a random forest algorithm to perform state classification prediction on a feature data set to obtain a first evaluation category and a first confidence level; using a support vector machine to perform state classification prediction on a feature data set to obtain a second evaluation category and a second confidence level; based on the first confidence level and the second confidence level, performing a fusion calculation on the first evaluation category and the second evaluation category to obtain a comprehensive evaluation result. Through the scheme of the present application, the physiological health status of a pet can be more comprehensively evaluated.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence, and in particular to a detection method and system for multimodal physiological indicators for pet health assessment. Background Art

[0002] In recent years, with the improvement of people's living standards and the increase in the number of pets, the importance of pet health assessment has become increasingly prominent. The detection technology used for pet health assessment is an important tool to ensure the quality of life of pets, prevent diseases and provide timely treatment. Through effective health assessment, pet owners and veterinarians can better understand the health status of their pets, so as to develop scientific maintenance and treatment plans. Existing pet health assessment technologies mainly include physical examinations, laboratory tests and imaging examinations, which can provide health information of pets to a certain extent, but also have some limitations. Traditional pet health assessment technologies mainly rely on physical examinations and laboratory tests. Physical examinations include the evaluation of basic vital signs such as the pet's appearance, behavior, body temperature, heart rate and respiration. Laboratory tests involve the analysis of samples such as blood, urine and feces to detect possible disease indicators. The working principles of these methods are relatively simple. By observing and measuring the physiological parameters of pets and combining them with laboratory analysis results, the health status of pets can be judged.

[0003] However, there are some problems with these existing technologies. First, traditional physical examinations and laboratory tests usually require the cooperation of pets, and some pets may show tension or restlessness during the test, affecting the accuracy of the test results. Second, laboratory tests require invasive sampling, which may cause some discomfort to pets. In addition, although imaging examinations such as X-rays and ultrasounds can provide detailed images of internal organs, the equipment is expensive and complicated to operate, requiring professionals to operate and interpret the results. Some studies have explored the use of non-invasive methods to assess the health of pets. For example, by analyzing the pet's behavior patterns, eating habits, etc., combined with machine learning algorithms for intelligent identification and analysis. These methods avoid causing pain to pets and are more economical. But the disadvantage is that they are difficult to fully assess the physiological health of pets and lack sensitivity to some deep-seated pathological changes.

[0004] Therefore, there is an urgent need for a technical solution that can more comprehensively evaluate the physiological health status of pets. Summary of the invention

[0005] In order to solve the deficiencies of the prior art, the embodiments of the present application provide a method and system for detecting multimodal physiological indicators for pet health assessment. The present application solves the technical problems of the prior art such as the non-invasive detection and evaluation being too one-sided.

[0006] The embodiment of the present application provides a detection method for multimodal physiological indicators for pet health assessment, including: collecting body temperature data, heart rate data, respiratory data, activity data and sleep data to obtain a multimodal physiological indicator data set; performing data preprocessing on the multimodal physiological indicator data set to obtain a feature data set; using a random forest algorithm to perform state classification prediction on the feature data set to obtain a first evaluation category and a first confidence level; using a support vector machine to perform state classification prediction on the feature data set to obtain a second evaluation category and a second confidence level; based on the first confidence level and the second confidence level, performing a fusion calculation on the first evaluation category and the second evaluation category to obtain a comprehensive evaluation result.

[0007] In a possible implementation, the collecting of body temperature data, heart rate data, respiratory data, activity data and sleep data to obtain a multimodal physiological indicator data set includes: collecting time series data of the change in current body temperature based on the baseline body temperature to obtain body temperature data; collecting heart rate time series data to obtain heart rate data; collecting time series data of the average value of the number of respirations from the current time point to the target time point to obtain respiratory data; collecting activity time series data to obtain activity data; wherein the activity time series data is Among them, A(t) represents the activity data at time t, a x (t), a y (t) and a z (t) represents the acceleration value in three axes; collects time series data of sleep states including deep sleep, light sleep or wakefulness to obtain sleep data.

[0008] In a possible implementation, the data preprocessing of the multimodal physiological indicator dataset to obtain a feature dataset includes: data cleaning of the multimodal physiological indicator dataset to remove sensor noise and outliers; filling missing data with linear interpolation; and standardizing the data to the same dimension.

[0009] In a possible implementation, the use of a random forest algorithm to perform state classification prediction on a feature data set to obtain a first evaluation category and a first confidence level includes: obtaining a prediction category output by each decision tree based on the feature data set; obtaining the most prediction categories by majority voting as the first evaluation category; and calculating the ratio of the number of decision trees whose prediction category is the first evaluation category to the total number of decision trees in the random forest to obtain a first confidence level.

[0010] In a possible implementation, the use of a support vector machine to perform state classification prediction on a feature data set to obtain a second evaluation category and a second confidence level includes: inputting the feature data set into three two-class classifiers, and selecting the category with the largest decision function value among the three two-class classifiers as the second evaluation category; and calculating the second confidence level of the second evaluation category based on the confidence values ​​of the three two-class classifiers.

[0011] In a possible implementation, based on the first confidence level and the second confidence level, the first evaluation category and the second evaluation category are fused and calculated to obtain a comprehensive evaluation result, including: calculating the posterior probability of the first evaluation category and the second evaluation category by combining the first confidence level and the second confidence level with the corresponding likelihood function; if the posterior probability of the first evaluation category is greater than the posterior probability of the second evaluation category, the comprehensive evaluation result is the first evaluation category, otherwise the comprehensive evaluation result is the second evaluation category.

[0012] In a possible implementation, the step of calculating the second confidence of the second evaluation category based on the confidence values ​​of the three two-category classifiers includes: k =∑ l≠k Confidence k，l , among which, Confidence k Represents the second confidence of the second evaluation category k, Confidence k , l Indicates the confidence level when the kth class is regarded as a positive example and the lth class is regarded as a negative example.

[0013] In a possible implementation, the calculating the posterior probabilities of the first evaluation category and the second evaluation category by combining the first confidence and the second confidence with the corresponding likelihood function comprises: Among them, P(C 1 |x) represents the posterior probability of the first evaluation category, P(C 2 |x) represents the posterior probability of the second evaluation category, p 1 represents the first confidence level, p 2 Represents the second confidence level, C 1 Indicates the first evaluation category, C 2 represents the second evaluation category, π 1 represents the posterior probability of random forest, π 2 represents the posterior probability of the support vector machine.

[0014] An embodiment of the present application also provides a detection system for multimodal physiological indicators for pet health assessment, including: an acquisition unit, a preprocessing unit and an evaluation unit; wherein the acquisition unit is used to collect body temperature data, heart rate data, respiratory data, activity data and sleep data to obtain a multimodal physiological indicator data set; the preprocessing unit is used to perform data preprocessing on the multimodal physiological indicator data set to obtain a feature data set; the evaluation unit is used to use a random forest algorithm to perform state classification prediction on the feature data set to obtain a first evaluation category and a first confidence level; use a support vector machine to perform state classification prediction on the feature data set to obtain a second evaluation category and a second confidence level; based on the first confidence level and the second confidence level, perform a fusion calculation on the first evaluation category and the second evaluation category to obtain a comprehensive evaluation result.

[0015] An embodiment of the present application also provides a detection device for multimodal physiological indicators for pet health assessment, including: a processor, a memory, and a system bus; wherein the processor and the memory are connected via the system bus; the memory is used to store one or more programs, and the one or more programs include instructions, which, when executed by the processor, enable the processor to execute the method described in the above embodiments.

[0016] In the detection method, system and device of multimodal physiological indicators for pet health assessment provided above, the embodiment of the present application can reduce interference and discomfort to pets through non-invasive multimodal detection, and improve the acceptability and popularity of detection. Comprehensive analysis of multiple physiological indicators can provide more comprehensive and accurate health assessment results, help veterinarians and pet owners better understand the health status of pets, timely discover and intervene in potential health problems, and with the help of artificial intelligence and big data technology, can achieve personalized health management and prevention, and improve the quality of life and health level of pets. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 A schematic diagram of a flow chart of a method for detecting multimodal physiological indicators for pet health assessment provided in an embodiment of the present application;

[0019] Figure 2 A schematic block diagram of a detection system for multimodal physiological indicators for pet health assessment provided in an embodiment of the present application. DETAILED DESCRIPTION

[0020] Various exemplary embodiments of the present application will now be described in detail with reference to the accompanying drawings. It should be noted that unless otherwise specifically stated, the relative arrangement of components and steps, numerical expressions and numerical values ​​set forth in these embodiments do not limit the scope of the present application.

[0021] Those skilled in the art will appreciate that the terms "first", "second" and the like in the embodiments of the present application are only used to distinguish different steps, devices or modules, etc., and do not represent any specific technical meaning, nor do they represent the necessary logical order between them. It should also be understood that in the embodiments of the present application, "multiple" may refer to two or more, and "at least one" may refer to one, two or more. It should also be understood that for any component, data or structure mentioned in the embodiments of the present application, in the absence of explicit limitation or contrary revelation given in the context, it can generally be understood as one or more. In addition, the term "and / or" in the present application is only a kind of association relationship describing the associated objects, indicating that there can be three kinds of relationships, for example, A and / or B, which can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in the present application generally indicates that the associated objects before and after are an "or" relationship. It should also be understood that the description of each embodiment in the present application emphasizes the differences between the embodiments, and the same or similar parts can refer to each other. For the sake of brevity, they will not be repeated one by one.

[0022] At the same time, it should be understood that, for ease of description, the sizes of the various parts shown in the drawings are not drawn according to the actual proportional relationship. The following description of at least one exemplary embodiment is actually only illustrative and is by no means intended to limit the present application and its application or use. The techniques, methods and devices known to ordinary technicians in the relevant fields may not be discussed in detail, but where appropriate, the techniques, methods and devices should be considered as part of the specification. It should be noted that similar numbers and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further discussed in subsequent drawings.

[0023] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0024] Figure 1The flowchart of the multimodal physiological index detection method for pet health assessment provided in the embodiment of the present application is as follows. The multimodal physiological index detection method combines multiple physiological data, including but not limited to body temperature, heart rate, respiratory rate, activity level and sleep pattern, etc., and realizes a comprehensive assessment of the health status of the pet through multiple sensors and data fusion technology. Figure 1 As shown, at step S101, body temperature data, heart rate data, breathing data, activity data and sleep data are collected to obtain a multimodal physiological indicator data set.

[0025] Specifically, first, collect the time series data of the change in current body temperature based on the baseline body temperature to obtain body temperature data. You can install an infrared temperature sensor on the pet's ear or armpit to collect data once a minute. You can also use non-contact infrared temperature measurement technology to obtain body temperature readings by measuring the thermal radiation on the pet's body surface: M = ∈σAT 4 , where M is the measured radiation energy, ∈ is the emissivity, σ is the Stefan-Boltzmann constant, A is the measurement area, and T is the absolute temperature. By pre-calibrating parameters such as emissivity, the target temperature can be inferred from the measured radiation energy. The advantage of this technology is that it does not require direct contact and avoids interference with pets. It also has the characteristics of high precision and fast response. A micro infrared sensor can be integrated into a pet collar or neck ring to achieve continuous monitoring of body temperature.

[0026] Then, collect the heart rate time series data to obtain the heart rate data. The photoplethysmography (PPG) principle can be used for non-invasive heart rate detection: Where LED_in and LED_out are the incident and transmitted light intensities, α is the extinction coefficient, and C is the concentration. By detecting the change in blood volume due to ventricular contraction, the pulse waveform can be extracted and the heart rate can be calculated. The photoplethysmography (PPG) sensor is installed on the neck or leg of the pet and data is collected once a second: Among them, HR(t) is the heart rate at time t, and RR(t) is the time interval between adjacent heartbeats. Micro LED light sources and photodiodes can be integrated on pet collars or wristbands to obtain light signals that penetrate limbs or auricles, extract pulse wave information from them, and realize real-time monitoring of heart rate.

[0027] Secondly, collect the average time series data of the number of respirations from the current time point to the target time point to obtain the respiration data. The accelerometer can be installed on the pet's chest to collect data once a second: where RR(t) is the respiratory rate at time t.

[0028] Resistive respiration sensing can also be used to detect chest movement using the piezoresistive effect on a Brownian motion substrate: Among them, R 0 Baseline resistance, π l and π t are the piezoresistance coefficients of length and cross-sectional area, respectively, and l and A are the current length and cross-sectional area. When the thorax moves periodically, the length and cross-sectional area of ​​the sensor change, causing a corresponding change in the resistance value, from which the breathing frequency can be extracted. By integrating the flexible breathing sensor into pet clothing or collars, they will fit tightly to the pet's torso, capture the thorax movement signal, and achieve continuous monitoring of the breathing rate.

[0029] Next, the activity time series data is collected to obtain the activity data; wherein the activity time series data is Among them, A(t) represents the activity data at time t, a x (t), a y (t) and a z (t) represents the acceleration value in three axes. A three-axis accelerometer can be installed on the pet's collar to collect data once per second.

[0030] Finally, the time series data of the sleep state including deep sleep, light sleep or wakefulness is collected to obtain sleep data. A pressure sensor or an acceleration sensor can be installed in the pet's sleeping area to collect data once a minute: S(t) = sleepstate at time t, where S(t) represents the sleep state at time t (such as deep sleep, light sleep, wakefulness).

[0031] At step S102, the multimodal physiological indicator data set is preprocessed to obtain a feature data set. It should be understood that when preprocessing the multimodal physiological indicator data set, the quality and consistency of the data must be ensured to provide a reliable basis for subsequent analysis and classification. Specifically, data preprocessing includes the following key steps: data cleaning, missing data filling, and data standardization.

[0032] Furthermore, the multimodal physiological indicator data set is first cleaned to remove sensor noise and outliers. The purpose of data cleaning is to remove sensor noise and outliers to ensure the accuracy and integrity of the data. During the acquisition process, the sensor may be affected by various factors, such as electromagnetic interference, environmental noise, etc., resulting in certain noise and outliers in the data. If these noises and outliers are not processed, it may have an adverse effect on the subsequent analysis results. Data cleaning methods may include but are not limited to median filtering and sliding window technology.

[0033] Median filtering is an effective denoising method. Its basic principle is to replace the data point in the center of a window with the median of all data points in a window to remove sudden noise. Assuming the window size is 2k+1, the calculation formula of median filtering is: y[i]=median(x[ik],…,x[i+k]), where x[i] represents the original data point, y[i] represents the filtered data point, and median(·) represents the median operation.

[0034] The sliding window technique is to reduce the impact of noise by smoothing the data within a time window. Assuming the window size is 2k+1, the calculation formula for sliding window smoothing is: Among them, x[i] represents the original data point, and y[i] represents the smoothed data point.

[0035] Then, the missing data are filled by linear interpolation. For the missing data in the multimodal physiological index data set, appropriate methods need to be used to fill in the missing data to ensure the integrity of the data. Linear interpolation is a missing data filling method. Its basic idea is to use the linear relationship between known data points to estimate the missing data points. Let the time of the missing data point be t, and the time of the adjacent valid data points before and after it be t prev and t next , the corresponding data values ​​are D prev and D next , then the missing data point D missing The linear interpolation calculation formula is: This method is simple and computationally efficient and is suitable for situations where there is a strong linear relationship between the data points.

[0036] Secondly, standardize the data to the same dimension. Standardization methods can include mean-variance standardization and interval scaling. Mean-variance standardization is to obtain standardized data with a mean of 0 and a standard deviation of 1 by subtracting the mean from the data and dividing by the standard deviation. Let the data set be D = {d 1 , d 2 , …, d n}, its mean is μ, and its standard deviation is σ, then the standardized data D standard The calculation formula is: Interval scaling is to normalize the data to the interval [0, 1] by subtracting the minimum value from the data and dividing it by the range (maximum value minus minimum value). 1 , d 2 , …, d n}, its minimum value is D min , the maximum value is D max , then the standardized data Dscaled The calculation formula is:

[0037] Through these preprocessing steps, the quality of multimodal physiological indicator data can be significantly improved, ensuring that subsequent machine learning algorithms can be trained and predicted based on high-quality data, thereby improving the accuracy and reliability of pet health assessments.

[0038] At step S103, the random forest algorithm is used to perform state classification prediction on the feature data set to obtain a first evaluation category and a first confidence level. It should be understood that Random Forest is an ensemble learning algorithm based on decision trees, which constructs multiple decision trees and combines their prediction results to obtain better prediction performance. When constructing each decision tree, the algorithm randomly selects a subset from the training data as the training set and randomly selects a feature at each node for splitting. This randomness can reduce the risk of overfitting the model. The construction process of each decision tree is independent of each other, so multi-core parallel computing can be used to accelerate the training process. The random forest algorithm has good robustness to outliers and noise data, and usually does not require feature scaling.

[0039] In an embodiment of the present application, the health status of a pet is evaluated based on its body temperature, heart rate, respiratory rate, activity level and sleep status, and is divided into three categories: "healthy", "sub-healthy" and "sick". Through the method of the above embodiment, a certain amount of labeled data is collected, and data preprocessing is performed to obtain a feature data set. The feature data set includes the following features: body temperature change (the difference between the body temperature data and the baseline body temperature), heart rate, respiratory rate, activity level, sleep status (deep sleep, light sleep or wakefulness, represented by 0, 1, 2). This feature data set is trained using a random forest algorithm to obtain a classification model for predicting the health status of new pet data.

[0040] The random forest algorithm achieves prediction by constructing multiple decision trees. The decision tree construction process is as follows: First, a subset is randomly selected from the training data with replacement as the training set of the current decision tree. Then, at each node, a feature subset is randomly selected from all features, and then the optimal feature is selected from this subset for splitting. At the same time, this step is repeated until the stopping condition is met (such as the maximum depth, the number of node samples is less than the threshold, etc.). For the sample set D of the current node, it is necessary to select an optimal feature j and a threshold t to divide D into two child nodes D left and D right :D left ={x|x[j]≤t}; D right ={xx[j]>t}.

[0041] The Gini coefficient (Gini Impurity) or information gain (Information Gain) can be used as an indicator for selecting the optimal feature. Taking the Gini coefficient as an example, it is defined as follows: Gini (D) = 1-∑ (p k ) 2 , where p k It indicates the proportion of samples belonging to category k in the dataset D. The value range of the Gini coefficient is [0, 1]. The smaller the value, the higher the purity of the dataset.

[0042] For feature j and threshold t, the Gini coefficient after splitting can be calculated: Gini(j, t) = (|D l eft| / |D|)*Gini(D l eft)+(|D r ight| / |D|)*Gini(D r ight), select the feature j and threshold t that minimize Gini(j, t) as the splitting condition of the current node.

[0043] Next, repeat the above steps until enough decision trees are built. For a new sample x to be predicted, each decision tree will output a prediction category, that is, the prediction category output by each decision tree is obtained based on the feature data set. Then, the most prediction categories are obtained by majority voting as the first evaluation category. The random forest algorithm combines the prediction results of all decision trees by voting or averaging to obtain the final prediction result. For classification problems, majority voting is used: where y i Represents the prediction result of the i-th decision tree, N trees Represents the total number of decision trees.

[0044] In addition, after obtaining the first evaluation category, the ratio of the number of decision trees whose prediction category is the first evaluation category to the total number of decision trees in the random forest is calculated to obtain the first confidence. Specifically, for classification problems, the confidence can be defined as: confidence = max(p k ), where p k Represents the voting ratio of the kth category. The higher the confidence, the more confident the model is in the current prediction result.

[0045] At step S104, a support vector machine is used to perform state classification prediction on the feature data set to obtain a second evaluation category and a second confidence level. It should be understood that support vector machine (SVM) is a supervised machine learning algorithm that is commonly used to solve classification and regression problems. Its basic idea is to separate data samples of different categories by constructing a hyperplane, and maximize the distance between the hyperplane and the nearest data sample. These nearest data sample points are called support vectors. The support vector machine has good generalization performance and can effectively avoid overfitting problems. It has good applicability for high-dimensional data. For nonlinear problems, data can be mapped to high-dimensional space through kernel techniques, thereby constructing a linear classifier in high-dimensional space.

[0046] Based on the feature data set in the above-mentioned embodiment, including features such as temperature change, heart rate, respiratory rate, activity level and sleep status, the health status of the pet needs to be divided into three categories: "healthy", "sub-healthy" and "sick".

[0047] The previous embodiment has used the random forest algorithm to obtain the first evaluation category and the first confidence level. Now, the support vector machine algorithm will be used to perform classification prediction on the same feature data set to obtain the second evaluation category and the second confidence level.

[0048] First, the feature data set is input into three two-class classifiers, and the category with the largest decision function value among the three two-class classifiers is selected as the second evaluation category. For multi-class problems, the strategy of "one-vs-Rest" or "one-vs-One" can be adopted to decompose it into multiple two-class problems. In the embodiment of the present application, taking "one-vs-many" as an example, for K categories, K two-class classifiers need to be trained, and the k-th classifier uses the k-th class as a positive example and other classes as negative examples. When predicting, the category with the largest decision function value is selected as the prediction result.

[0049] Then, the second confidence of the second evaluation category is calculated based on the confidence values ​​of the three two-class classifiers. The confidence of the support vector machine can be measured by the size of the decision function value. For the two-class problem, the confidence can be defined as: in is the decision function value of the support vector machine. In this application, for multi-class problems, a "one-to-one" strategy can be adopted to train a two-class classifier for each pair of categories. Assuming there are K categories, then we need to train For the kth class, calculate the sum of its confidence with each other class: k =∑ l≠k Confidence k，l, among which, Confidence k Represents the second confidence of the second evaluation category k, Confidence k，l Indicates the confidence level when the kth class is regarded as a positive example and the lth class is regarded as a negative example.

[0050] At step S105, based on the first confidence and the second confidence, the first evaluation category and the second evaluation category are fused and calculated to obtain a comprehensive evaluation result. In the above method, the feature data set of the pet's health status is classified and predicted using the random forest algorithm and the support vector machine algorithm, respectively, to obtain the first evaluation category and the first confidence, the second evaluation category and the second confidence. Now, it is necessary to fuse these two evaluation results to obtain the final comprehensive evaluation result. In practical applications, it is difficult for a single classifier to meet the needs of all scenarios because different classifiers have different preferences and defects. By fusing the results of multiple classifiers, the accuracy and robustness of the classification can be effectively improved. Multi-classifier fusion can reduce the bias and variance of a single classifier and improve generalization ability. Different classifiers may have different classification results for the same sample, and more reliable results can be obtained through fusion. Use the advantages of different classifiers to make up for the defects of a single classifier.

[0051] Specifically, let the first evaluation category be C 1 , the first confidence level is p 1 ; The second assessment category is C 2 , the second confidence level is p 2 . Need to be based on C 1 , C 2 、p 1 and p 2 The comprehensive evaluation category C and comprehensive confidence p are calculated.

[0052] The first confidence level and the second confidence level are combined with the corresponding likelihood function to calculate the posterior probability of the first evaluation category and the second evaluation category. Specifically, the posterior probability of each model is calculated based on the likelihood function of each model on the training data, and their prediction results are weighted averaged.

[0053] Suppose there are K classifiers For a given input feature vector x, the kth classifier predicts that the probability of it belonging to category C is According to Bayes' theorem, the posterior probability of each model when x is observed can be calculated: in is the likelihood function of the kth model when x is observed, which can be calculated using the training data; is the prior probability of the kth model. It can usually be assumed that the prior probabilities of all models are equal. P(x) is a normalizing constant.

[0054] Then, the fused comprehensive probability can be calculated: In other words, the probability predicted by each model is multiplied by the posterior probability of the model, and then the weighted average is taken to obtain the integrated probability after fusion.

[0055] Finally, the category with the largest comprehensive probability is selected as the final comprehensive evaluation category: At the same time, the probability value corresponding to this category is the comprehensive confidence.

[0056] In the pet health status assessment system, assume that the random forest model predicts the first assessment category C 1 The probability is p 1 , the support vector machine model predicts the second evaluation category C 2 The probability is p 2 , and their posterior probabilities are π 1 and π 2 , then the comprehensive evaluation category C and comprehensive confidence p can be calculated as follows:

[0057] Among them, P(C 1 |x) represents the posterior probability of the first evaluation category, P(C 2 |x) represents the posterior probability of the second evaluation category, p 1 represents the first confidence level, p 2 Represents the second confidence level, C 1 Indicates the first evaluation category, C 2 represents the second evaluation category, π 1 represents the posterior probability of random forest, π 2 represents the posterior probability of the support vector machine.

[0058] If P(C 1 |x)>P(C 2 |x), then C=C 1 , and p=P(C 1 |x); otherwise C=C 2 , and p=P(C 2 |x), that is, if the posterior probability of the first evaluation category is greater than the posterior probability of the second evaluation category, the comprehensive evaluation result is the first evaluation category, otherwise the comprehensive evaluation result is the second evaluation category.

[0059] Furthermore, it is important to understand that P(C 1 |x) and P(C 2 |x) respectively indicate that when the input feature vector x is given, the sample belongs to category C 1 and C 2The posterior probability of .

[0060] Specifically, P(C 1 |x) means that given the input feature x, the sample belongs to category C 1 The probability of P(C 2 |x) means that given the input feature x, the sample belongs to category C 2 These two probabilities are calculated by Bayes’ theorem and are the comprehensive probabilities obtained by weighted averaging the prediction results of different models for the same sample.

[0061] According to Bayes' theorem: in, The first model (such as random forest) predicts that sample x belongs to category C 1 The probability of The first model The posterior probability when the sample x is observed, and the other The second model (such as support vector machine). P(C 2 |x) is calculated in a similar way, except that C 1 Replace with C 2 That's it.

[0062] By comparing P(C 1 |x) and P(C 2 |x) can determine the final category of the comprehensive evaluation, and the larger probability value is the comprehensive confidence of the category. 1 |x)=P(C 2 |x), that is, the posterior probabilities of the two categories are equal, then there are several ways to deal with it: you can randomly select C 1 or C 2 As a comprehensive evaluation category. This method is simple, but has a certain degree of randomness. Comparing the two models and The original prediction confidence of the corresponding category and The one with higher confidence is selected as the comprehensive evaluation category. If the prior probabilities of the two categories are different, the category with a higher prior probability can be selected as the final result. The prior probability can be estimated from the training data. In general, when the probabilities of the two categories are equal, it is necessary to select an appropriate strategy based on the requirements of the specific task and the characteristics of the data.

[0063] Figure 2A schematic block diagram of a detection system for multimodal physiological indicators for pet health assessment provided in an embodiment of the present application. It should be understood that the system shown in the figure is exemplary and not restrictive. This means that the system architecture involved is not limited to a specific form or design, but is presented as an example. In other words, the architecture shown in the figure can be regarded as a way of expression to clearly describe related concepts and relationships, and does not exclude other forms of architecture. Therefore, when interpreting the architecture in the picture, it should be understood that the model is flexible and diverse, and its purpose is to provide an exemplary description rather than a restrictive provision on a specific form.

[0064] Specifically, the detection system of multimodal physiological indicators for pet health assessment includes: an acquisition unit 201, a preprocessing unit 202 and an evaluation unit 203. The acquisition unit 201 is used to collect body temperature data, heart rate data, respiratory data, activity data and sleep data to obtain a multimodal physiological indicator data set; the preprocessing unit 202 is used to perform data preprocessing on the multimodal physiological indicator data set to obtain a feature data set; the evaluation unit 203 is used to use a random forest algorithm to perform state classification prediction on the feature data set to obtain a first evaluation category and a first confidence level; use a support vector machine to perform state classification prediction on the feature data set to obtain a second evaluation category and a second confidence level; based on the first confidence level and the second confidence level, the first evaluation category and the second evaluation category are fused and calculated to obtain a comprehensive evaluation result.

[0065] Furthermore, an embodiment of the present application also provides a detection device for multimodal physiological indicators for pet health assessment, including: a processor, a memory, and a system bus; the processor and the memory are connected via the system bus; the memory is used to store one or more programs, and the one or more programs include instructions, which, when executed by the processor, enable the processor to execute any of the above methods.

[0066] Furthermore, an embodiment of the present application also provides a computer program product, which, when executed on a terminal device, enables the terminal device to execute any of the above methods.

[0067] It can be known from the description of the above implementation mode that those skilled in the art can clearly understand that all or part of the steps in the above-mentioned embodiment method can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product can be stored in a storage medium such as ROM / RAM, a disk, an optical disk, etc., including several instructions for enabling a computer device (which can be a personal computer, a server, or a network communication device such as a media gateway, etc.) to execute the methods described in the various embodiments of the present application or certain parts of the embodiments.

[0068] It should be noted that the various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same or similar parts between the various embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description.

[0069] It should also be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0070] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for detecting multimodal physiological indicators for pet health assessment, characterized in that: include: Collect body temperature data, heart rate data, breathing data, activity data and sleep data to obtain a multimodal physiological indicator data set; Preprocess the multimodal physiological index data set to obtain a feature data set; Using a random forest algorithm to perform state classification prediction on the feature data set to obtain a first evaluation category and a first confidence level; Using a support vector machine to perform state classification prediction on the feature data set to obtain a second evaluation category and a second confidence level; Based on the first confidence level and the second confidence level, the first evaluation category and the second evaluation category are fused and calculated to obtain a comprehensive evaluation result; wherein, the method includes: calculating the posterior probability of the first evaluation category and the second evaluation category by combining the first confidence level and the second confidence level with the corresponding likelihood function; if the posterior probability of the first evaluation category is greater than the posterior probability of the second evaluation category, the comprehensive evaluation result is the first evaluation category, otherwise the comprehensive evaluation result is the second evaluation category; The step of calculating the posterior probabilities of the first evaluation category and the second evaluation category by combining the first confidence level and the second confidence level with the corresponding likelihood function includes: ,in, represents the posterior probability of the first evaluation category, represents the posterior probability of the second evaluation category, represents the first confidence level, represents the second confidence level, represents the first evaluation category, represents the second evaluation category, represents the posterior probability of the random forest, represents the posterior probability of the support vector machine.

2. The detection method according to claim 1, characterized in that: in, The body temperature data, heart rate data, breathing data, activity data and sleep data are collected to obtain a multimodal physiological indicator data set, including: Collecting the time series data of the change of the current body temperature based on the baseline body temperature to obtain body temperature data; Collecting heart rate time series data to obtain heart rate data; Collect the average time series data of the number of respirations from the current time point to the target time point to obtain respiration data; Collect the activity time series data to obtain activity data; The activity time series data is: in, Indicates time Activity data at the time, Indicates the acceleration values ​​in three axes; The time series data of the sleep state including deep sleep, light sleep or wakefulness is collected to obtain sleep data.

3. The detection method according to claim 1, characterized in that: in, The data preprocessing of the multimodal physiological indicator data set to obtain a feature data set includes: Data cleaning of multimodal physiological indicator datasets to remove sensor noise and outliers; Missing data were filled using linear interpolation; Normalize the data to the same dimension.

4. The detection method according to claim 1, characterized in that: in, The using of the random forest algorithm to perform state classification prediction on the feature data set to obtain a first evaluation category and a first confidence level includes: Get the predicted category output by each decision tree based on the feature data set; The majority voting method is used to obtain the most predicted categories as the first evaluation category; The first confidence level is calculated by calculating the ratio of the number of decision trees whose predicted category is the first evaluation category to the total number of decision trees in the random forest.

5. The detection method according to claim 1, characterized in that: in, The method of using a support vector machine to perform state classification prediction on the feature data set to obtain a second evaluation category and a second confidence level includes: Input the feature data set into three two-class classifiers, and select the category with the largest decision function value among the three two-class classifiers as the second evaluation category; A second confidence for the second evaluation category is calculated based on the confidence values ​​of the three binary classifiers.

6. The detection method according to claim 5, characterized in that: in, The calculating of the second confidence of the second evaluation category based on the confidence values ​​of the three two-category classifiers includes: ,in, Indicates the second evaluation category The second confidence level, Indicates that the Class as a positive example, The confidence level when a class is a negative example.

7. A multimodal physiological index detection system for pet health assessment, applied to the detection method according to any one of claims 1 to 6, characterized in that: include: Acquisition unit, pre-processing unit and evaluation unit; wherein, The acquisition unit is used to collect body temperature data, heart rate data, breathing data, activity data and sleep data to obtain a multimodal physiological indicator data set; The preprocessing unit is used to perform data preprocessing on the multimodal physiological indicator data set to obtain a feature data set; The evaluation unit is used to use a random forest algorithm to perform state classification prediction on a feature data set to obtain a first evaluation category and a first confidence level; use a support vector machine to perform state classification prediction on the feature data set to obtain a second evaluation category and a second confidence level; and based on the first confidence level and the second confidence level, perform a fusion calculation on the first evaluation category and the second evaluation category to obtain a comprehensive evaluation result.

8. A multimodal physiological index detection device for pet health assessment, characterized in that: include: A processor, a memory, and a system bus; wherein the processor and the memory are connected via the system bus; The memory is used to store one or more programs, wherein the one or more programs include instructions, and when the instructions are executed by the processor, the processor executes the method according to any one of claims 1 to 6.

Images