Emotion recognition method and device based on ballistocardiogram signal

Abstract

The invention provides an emotion recognition method and device based on a ballistocardiogram signal, and can be applied to the technical field of physiological signal processing. The method comprises the following steps: acquiring ballistocardiogram signal data of a target object and respective ballistocardiogram signal data of a plurality of candidate objects; similarity processing is carried out on the ballistocardiogram signal data of the target object and the ballistocardiogram signal data of the multiple candidate objects, the candidate objects associated with the target object are determined from the multiple candidate objects, and at least one reference object is obtained; and processing the ballistocardiogram signal data of the reference object by using the emotion recognition model to obtain an emotion recognition result of the target object.

Description

Technical Field

[0001] The present disclosure relates to the technical field of physiological signal processing, and in particular to an emotion recognition method and device based on a cardiac impulse signal. Background Art

[0002] Emotion recognition is an important research direction in the field of affective computing and can be applied in many fields. Existing emotion recognition technology generally uses external behaviors or physiological signals such as facial expressions, voices, and postures to recognize emotions. Since external behaviors themselves can be controlled by humans, people can hide or disguise themselves as they wish. In addition, the physiological signals between individuals vary greatly. Therefore, the emotion recognition methods in related technologies have low recognition accuracy. Summary of the invention

[0003] In view of the above problems, the present disclosure provides an emotion recognition method and device based on heartbeat signals.

[0004] According to the first aspect of the present disclosure, there is provided an emotion recognition method based on cardiac impact signal, comprising: acquiring cardiac impact signal data of a target object and cardiac impact signal data of each of a plurality of candidate objects; performing similarity processing on the cardiac impact signal data of the target object and the cardiac impact signal data of each of a plurality of candidate objects, determining a candidate object associated with the target object from the plurality of candidate objects, and obtaining at least one reference object; and processing the cardiac impact signal data of the reference object using an emotion recognition model to obtain an emotion recognition result of the target object.

[0005] According to an embodiment of the present disclosure, the candidate objects include M, where M is a positive integer greater than 1, and the candidate objects are marked with emotion category labels; similarity processing is performed on the cardiac impact signal data of the target object and the cardiac impact signal data of each of the multiple candidate objects, and the candidate object associated with the target object is determined from the multiple candidate objects to obtain at least one reference object, including: for every two candidate objects among the M candidate objects, the first similarity between the two candidate objects is determined using the cardiac impact signal data and the emotion category labels of the two candidate objects to obtain a first similarity matrix of M×M dimensions; for each candidate object among the M candidate objects, the second similarity between the target object and the candidate object is determined using the cardiac impact signal data of the candidate object and the cardiac impact signal data of the target object to obtain a second similarity matrix of M×1 dimensions; according to the M×M first similarity matrix and the M×1 second similarity matrix, the M candidate objects and the target object are clustered to obtain multiple cluster clusters; from the multiple cluster clusters, the cluster cluster where the target object is located is determined to obtain a target cluster; the candidate objects included in the target cluster are determined as candidate objects associated with the target object to obtain at least one reference object.

[0006] According to an embodiment of the present disclosure, the cardiac impact signal data of the target object includes multiple groups of cardiac impact signal sub-data, M candidate objects commonly correspond to an emotion recognition sub-model, and each candidate object has a corresponding emotion recognition sub-model; using the cardiac impact signal data of the candidate object and the cardiac impact signal data of the target object to determine the second similarity between the target object and the candidate object includes: using the emotion recognition sub-model commonly corresponding to the M candidate objects to process the multiple groups of cardiac impact signal sub-data respectively, and obtain first emotion category prediction values ​​corresponding to each of the multiple groups of cardiac impact signal sub-data; for the mth candidate object among the M candidate objects, using the emotion recognition sub-model corresponding to the mth candidate object to process the target cardiac impact signal sub-data, and obtain a second emotion category prediction value, wherein the target cardiac impact signal sub-data is the cardiac impact signal sub-data corresponding to the target number of first emotion category prediction values ​​determined based on a preset sorting rule among the multiple first emotion category prediction values, 1 m M; determining a second similarity between the target object and the mth candidate object according to the target first emotion category prediction value and the second emotion category prediction value, wherein the target first emotion category prediction value is the largest first emotion category prediction value among the multiple first emotion category prediction values.

[0007] According to an embodiment of the present disclosure, determining the first similarity between the two candidate objects using the heartbeat signal data and the emotion category labels of the two candidate objects respectively includes: for the ath candidate object and the bth candidate object among the M candidate objects, using the emotion recognition sub-model corresponding to the ath candidate object to process the heartbeat signal data of the bth candidate object, to obtain a third emotion category prediction value of the bth candidate object, 1 M, 1 M; obtaining a first similarity between the ath candidate object and the bth candidate object according to the third emotion category prediction value of the bth candidate object and the emotion category label of the bth candidate object.

[0008] According to an embodiment of the present disclosure, the emotion recognition method based on the cardiac impact signal also includes: based on the balanced distribution adaptive optimization function, performing feature alignment processing on the cardiac impact signal data of the target object and the reference object to obtain the conversion signal data of the reference object; using the emotion recognition model to process the cardiac impact signal data of the reference object to obtain the emotion recognition result of the target object, including: using the emotion recognition model to process the conversion signal data of the reference object to obtain the emotion recognition result of the target object.

[0009] According to an embodiment of the present disclosure, based on a balanced distribution adaptive optimization function, feature alignment processing is performed on the cardiac impact signal data of the target object and the reference object to obtain the conversion signal data of the reference object, including: determining the target edge probability distribution and the target conditional probability distribution according to the cardiac impact signal data of the target object; determining the reference edge probability distribution and the reference conditional probability distribution according to the cardiac impact signal data of the reference object; using the balanced distribution adaptive optimization function to process the target edge probability distribution, the target conditional probability distribution, the reference edge probability distribution and the reference conditional probability distribution to obtain a conversion matrix; according to the conversion matrix and the cardiac impact signal data of the reference object, obtaining the conversion signal data of the reference object.

[0010] According to an embodiment of the present disclosure, a balanced distribution adaptive optimization function is used to process a target edge probability distribution, a target conditional probability distribution, a reference edge probability distribution and a reference conditional probability distribution to obtain a transformation matrix, including: using a mean difference function to calculate the difference between the target edge probability distribution and the reference edge probability distribution to obtain an edge difference value; using a mean difference function to calculate the difference between the target conditional probability distribution and the reference conditional probability distribution to obtain a conditional difference value; using a balanced distribution adaptive optimization function to process the edge difference value and the conditional difference value to obtain a transformation matrix, wherein the balanced distribution adaptive optimization function is used to optimize the distribution difference between the heart attack signal data of the target object and the reference object respectively. According to an embodiment of the present disclosure, an emotion recognition model includes N binary classifiers, where N is a positive integer, and the number of binary classifiers is determined based on the number of sample emotion categories. The emotion recognition model is trained based on the following operations: obtaining training samples, the training samples include conversion signal data of a sample reference object and a sample emotion category label of a sample target object, and the sample reference object is an object associated with the sample target object; performing feature extraction on the conversion signal data of the sample reference object to obtain signal features; using N binary classifiers to process the signal features respectively to obtain sample emotion category prediction values ​​corresponding to each binary classifier; performing statistics on the N sample prediction values ​​based on a statistical function to obtain a prediction score corresponding to each sample emotion category; determining the emotion category corresponding to the maximum prediction score among multiple prediction scores as the sample emotion recognition result, wherein the sample emotion recognition result represents the emotion category prediction result of the sample target object; training an initial emotion recognition model based on the sample emotion recognition result and the sample emotion category label to obtain an emotion recognition model.

[0011] According to an embodiment of the present disclosure, the heartbeat signal data includes at least one of the following: heart rate variability data, respiratory change data, beat-by-beat heartbeat statistics data, and beat-by-beat heartbeat nonlinearity data.

[0012] According to a second aspect of the present disclosure, there is provided an emotion recognition device based on cardiac impact signals, comprising: an acquisition module, used to acquire cardiac impact signal data of a target object and respective cardiac impact signal data of a plurality of candidate objects; a screening module, used to perform similarity processing on the cardiac impact signal data of the target object and respective cardiac impact signal data of a plurality of candidate objects, determine a candidate object associated with the target object from the plurality of candidate objects, and obtain at least one reference object; and an identification module, used to process the cardiac impact signal data of the reference object using an emotion recognition model, and obtain an emotion recognition result of the target object.

[0013] According to the emotion recognition method and device based on the cardiac impact signal provided by the present disclosure, by performing similarity processing on the cardiac impact signal data of the target object and the cardiac impact signal data of each of the multiple candidate objects, a candidate object associated with the target object is determined from the multiple candidate objects to obtain at least one reference object; the cardiac impact signal data of the reference object is processed using the emotion recognition model to obtain the emotion recognition result of the target object. Since the cardiac impact signal data of the target object and the cardiac impact signal data of each of the multiple candidate objects are evaluated for similarity, a reference object with high similarity to the target object is determined from the multiple candidate objects, and then the emotion recognition result of the target object is predicted based on the cardiac impact signal data of the reference object, the influence of individual physiological differences on the prediction accuracy of the emotion recognition model is greatly reduced, and the introduction of knowledge that is irrelevant to or even conflicting with the target object is avoided, thereby overcoming the technical problem of low emotion recognition accuracy caused by physiological differences between individual objects. In addition, selecting a reference object with high similarity for emotion recognition reduces unnecessary computational overhead and helps improve the efficiency of emotion recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The above and other objects, features and advantages of the present disclosure will become more apparent through the following description of the embodiments of the present disclosure with reference to the accompanying drawings.

[0015] Figure 1 A flow chart of an emotion recognition method based on a cardiac impulse signal according to an embodiment of the present disclosure is shown.

[0016] Figure 2 An example schematic diagram of locating the J peak of cardiac ballistocardiac signal data according to an embodiment of the present disclosure is shown.

[0017] Figure 3 An example schematic diagram of determining a reference object according to an embodiment of the present disclosure is shown.

[0018] Figure 4 An example schematic diagram of obtaining emotion recognition results according to an embodiment of the present disclosure is shown;

[0019] Figure 5A structural block diagram of an emotion recognition device based on a cardiac impulse signal according to an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0020] Hereinafter, embodiments according to the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present disclosure. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0021] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise", "include", etc. used herein indicate the existence of features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0022] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0023] When using expressions such as "at least one of A, B, and C, etc.", they should generally be interpreted according to the meaning of the expression commonly understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0024] In the process of implementing the present invention, it was found that when collecting cardiac shock signals based on non-contact collection equipment, there are emotional differences between individuals due to different personality types among individuals. Due to the nonlinear time-varying characteristics of the cardiac shock signal itself, there are also individual differences. The emotional characteristics between individuals no longer conform to the independent and identical distribution assumed by existing machine learning, and the generalization of the model will be affected, resulting in a significant decrease in the accuracy of the model trained on the sample source individuals on the target individuals.

[0025] In view of this, according to an embodiment of the present disclosure, a method and device for emotion recognition based on a cardiac impact signal are provided. The method includes: obtaining cardiac impact signal data of a target object and cardiac impact signal data of each of a plurality of candidate objects; performing similarity processing on the cardiac impact signal data of the target object and the cardiac impact signal data of each of a plurality of candidate objects, determining a candidate object associated with the target object from the plurality of candidate objects, and obtaining at least one reference object; and using an emotion recognition model to process the cardiac impact signal data of the reference object, and obtaining an emotion recognition result of the target object.

[0026] In the technical solution of the present disclosure, the user information (including but not limited to user personal information, user image information, user device information, such as location information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved are all information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with relevant laws, regulations and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0027] It should be noted that the sequence numbers of the operations in the following method are only used as representations of the operations for the purpose of description, and should not be regarded as representing the execution order of the operations. Unless explicitly stated, the method does not need to be executed completely in the order shown.

[0028] Figure 1 A flow chart of an emotion recognition method based on a cardiac impulse signal according to an embodiment of the present disclosure is shown.

[0029] like Figure 1 As shown, the method includes operations S110 to S130.

[0030] In operation S110 , the ballistocardiographic signal data of the target subject and the ballistocardiographic signal data of each of the plurality of candidate subjects are acquired.

[0031] According to an embodiment of the present disclosure, the target object is an emotional object to be identified, and the candidate object may be an object having ballistocardiographic signal data obtained from a shared database platform.

[0032] According to an embodiment of the present disclosure, the ballistocardiogram data is a physiological signal (Ballistocardiogram, BCG) collected by a non-contact sensor. For example, the ballistocardiogram data is collected without contact by placing a sensor in a seat cushion, on the surface, inside or bottom of a mattress.

[0033] In operation S120, similarity processing is performed on the ballistocardiographic signal data of the target object and the ballistocardiographic signal data of each of the plurality of candidate objects, and a candidate object associated with the target object is determined from the plurality of candidate objects to obtain at least one reference object.

[0034] According to an embodiment of the present disclosure, a similarity algorithm is used to process the cardiac signal data of the target object and the cardiac signal data of multiple candidate objects, and the similarity of the target object and the candidate objects in the direction of physiological characteristics is evaluated, and then a reference object with high similarity is selected from the multiple candidate objects.

[0035] According to an embodiment of the present disclosure, the ballistocardiographic signal data of the target object and the ballistocardiographic signal data of the candidate object have the same feature space and label space, but the data distribution is different.

[0036] According to an embodiment of the present disclosure, the reference object is an object having a significant correlation with the target object. For example, similarity evaluation is performed on the cardiac signal data of the target object and the candidate object, and then at least one reference object with a higher similarity value is selected from multiple candidate objects.

[0037] In operation S130, the ballistocardiographic signal data of the reference object is processed using the emotion recognition model to obtain an emotion recognition result of the target object.

[0038] According to an embodiment of the present disclosure, the emotion recognition model may be constructed based on a machine learning algorithm (Support Vector Machine (SVM) of a Radial Basis Function (RBF) kernel) for classification and regression tasks.

[0039] According to an embodiment of the present disclosure, the ballistocardial signal data of the reference object is input into the emotion recognition model, and the emotion recognition result is output. The emotion recognition result represents the emotion category prediction result of the target object.

[0040] For example, the emotion recognition result may be a positive result, a negative result, or a neutral result.

[0041] According to the embodiments of the present disclosure, the target object's cardiac impact signal data and the cardiac impact signal data of multiple candidate objects are similarly evaluated, and then a reference object with high similarity to the target object is determined from the multiple candidate objects, and then the target object's emotion recognition result is predicted based on the cardiac impact signal data of the reference object, thereby improving the generalization ability of the emotion recognition model, avoiding the introduction of knowledge that is irrelevant or even conflicting with the target object, thereby greatly reducing the impact of individual physiological differences on the prediction accuracy of the emotion recognition model, avoiding the introduction of knowledge that is irrelevant or even conflicting with the target object, thereby overcoming the technical problem of low emotion recognition accuracy caused by physiological differences between individual objects. In addition, selecting a reference object with high similarity for emotion recognition reduces unnecessary computational overhead and helps improve recognition efficiency.

[0042] According to an embodiment of the present disclosure, the heartbeat signal data includes at least one of the following: heart rate variability data, respiratory change data, beat-by-beat heartbeat statistics data, and beat-by-beat heartbeat nonlinearity data.

[0043] According to an embodiment of the present disclosure, the heart rate variability data characterizes the time domain, frequency domain, and nonlinear domain characteristics of the heartbeat interval, and the respiratory variation data characterizes the frequency and waveform characteristics of the respiratory rate.

[0044] According to an embodiment of the present disclosure, the beat-to-beat interval (IBI) is the time interval between two consecutive heartbeats. The beat-to-beat statistical data may be statistical data such as the mean value, standard deviation, maximum value, minimum value, etc. of the physiological data within this time interval. The beat-to-beat nonlinear data may be entropy characteristic data of the physiological data within this time interval, etc.

[0045] Figure 2 An example schematic diagram of locating the J peak of cardiac ballistocardiac signal data according to an embodiment of the present disclosure is shown.

[0046] like Figure 2 As shown, based on the template matching algorithm, the J peak of the cardiac impact signal data is located. The J peak is the main peak of the signal that occurs during ventricular contraction and blood flow impacting the descending aorta. The beat-by-beat heartbeat is extracted to obtain the beat-by-beat heartbeat (IBI) of the cardiac impact signal data. The extracted beat-by-beat interval sequence data is divided into subsequences each containing 100 IBIs, and then the heart rate variability data, respiratory change data, beat-by-beat heartbeat statistics data, beat-by-beat heartbeat nonlinear data, etc. are extracted from each subsequence.

[0047] According to an embodiment of the present disclosure, preprocessing of the cardiac impact signal data of the target object and the candidate object includes: obtaining the baseline cardiac impact signal data of the target object and the candidate object in a calm heartbeat state, and obtaining the mean of the heart rate variability characteristics, respiratory change characteristics, beat-to-beat statistical characteristics, and beat-to-beat nonlinear characteristics.

[0048] According to an embodiment of the present disclosure, with respect to the mean value of each feature, the feature data of the target object and the candidate objects are normalized to [-1, 1].

[0049] For example, the baseline heart rate variability data of the target object and the candidate object in a calm heartbeat state are obtained, the mean corresponding to the heart rate variability feature is obtained, and the heart rate variability data of the target object and the candidate object are standardized based on the mean to obtain the pre-processed heart rate variability data of the target object and the candidate object.

[0050] According to an embodiment of the present disclosure, the candidate objects include M, where M is a positive integer greater than 1, and the candidate objects are marked with emotion category labels; similarity processing is performed on the cardiac impact signal data of the target object and the cardiac impact signal data of each of the multiple candidate objects, and the candidate object associated with the target object is determined from the multiple candidate objects to obtain at least one reference object, including: for every two candidate objects among the M candidate objects, the first similarity between the two candidate objects is determined using the cardiac impact signal data and the emotion category labels of the two candidate objects to obtain a first similarity matrix of M×M dimensions; for each candidate object among the M candidate objects, the second similarity between the target object and the candidate object is determined using the cardiac impact signal data of the candidate object and the cardiac impact signal data of the target object to obtain a second similarity matrix of M×1 dimensions; according to the M×M first similarity matrix and the M×1 second similarity matrix, the M candidate objects and the target object are clustered to obtain multiple cluster clusters; from the multiple cluster clusters, the cluster cluster where the target object is located is determined to obtain a target cluster; the candidate objects included in the target cluster are determined as candidate objects associated with the target object to obtain at least one reference object.

[0051] According to an embodiment of the present disclosure, the target object has no emotion category label, and each candidate object is marked with an emotion category label. For example, the first candidate object is marked with emotion category label 0, which means that candidate object 1 is a negative category.

[0052] According to an embodiment of the present disclosure, the heartbeat signal data and emotion category labels of the two candidate objects are processed by using a label similarity algorithm to obtain a first similarity between the two candidate objects.

[0053] According to an embodiment of the present disclosure, each candidate object needs to calculate a first similarity with itself and with the remaining M-1 candidate objects respectively, so as to obtain an M×1-dimensional first similarity matrix corresponding to each candidate object. Therefore, an M×M-dimensional first similarity matrix is ​​obtained for the M candidate objects.

[0054] According to an embodiment of the present disclosure, a label similarity algorithm is used to calculate the second similarity between the target object and each candidate object, thereby obtaining a second similarity matrix of M×1 dimensions.

[0055] According to an embodiment of the present disclosure, a first similarity matrix of M×M dimensions and a second similarity matrix of M×1 dimensions constitute a similarity matrix of M×(M+1) dimensions. The similarity matrix of M×(M+1) dimensions is used as a feature of topic clustering, and a clustering algorithm (Kmeans) is used to perform clustering processing on M candidate objects and target objects to obtain multiple clustering results.

[0056] For example, multiple clustering results may include a clustering result corresponding to a cluster number of 1, a clustering result corresponding to a cluster number of 2, and a clustering result corresponding to a cluster number of 3. The variance ratio values ​​of the multiple clustering results are sorted, and the clustering result corresponding to the maximum variance ratio value is determined as the optimal clustering result. The optimal clustering result includes multiple cluster clusters.

[0057] In one embodiment, the clustering algorithm Kmeans is shown in formula (1):

[0058] (1);

[0059] in, Characterizes the clustering result corresponding to the number of clusters being 1, Characterize the clustering results corresponding to the number of clusters K, Represents a similarity matrix of M×(M+1) dimensions.

[0060] In one embodiment, the variance ratio value is as shown in formula (2):

[0061] (2);

[0062] in, Characterizes the variance ratio value corresponding to the clustering result corresponding to the number of clusters K, Characterize the trajectory of the matrix, Characterizes the covariance matrix between multiple clusters in the clustering results, The covariance matrix that represents the data in the clustering results, Characterizes the sum of the number of target objects and candidate objects.

[0063] According to an embodiment of the present disclosure, the optimal clustering result includes multiple clustering clusters, and the clustering cluster where the target object is located is determined as the target cluster; the candidate objects included in the target cluster are determined as candidate objects associated with the target object, and the candidate objects associated with the target object are determined as reference objects.

[0064] According to the embodiments of the present disclosure, by defining the similarity between candidate objects and between candidate objects and target objects, all objects are clustered and then sorted according to the similarity matrix, and at least one reference object that is similar to the target object with high confidence is adaptively selected to be included in the final set of objects used for identification, thereby greatly reducing the impact of individual physiological differences on the prediction accuracy of the emotion recognition model.

[0065] According to an embodiment of the present disclosure, the cardiac impact signal data of the target object includes multiple groups of cardiac impact signal sub-data, and M candidate objects correspond to an emotion recognition sub-model in common, and each candidate object has a corresponding emotion recognition sub-model; using the cardiac impact signal data of the candidate object and the cardiac impact signal data of the target object to determine the second similarity between the target object and the candidate object may include: using a label similarity algorithm to respectively calculate the second similarity between the target object and each candidate object, and obtain an M×1-dimensional second similarity matrix. Specifically, using the emotion recognition sub-model corresponding to the M candidate objects to respectively process the multiple groups of cardiac impact signal sub-data, and obtain the first emotion category prediction value corresponding to each of the multiple groups of cardiac impact signal sub-data; for the mth candidate object among the M candidate objects, using the emotion recognition sub-model corresponding to the mth candidate object to process the target cardiac impact signal sub-data, and obtain the second emotion category prediction value, wherein the target cardiac impact signal sub-data is the cardiac impact signal sub-data corresponding to the target number of first emotion category prediction values ​​determined based on a preset sorting rule among the multiple first emotion category prediction values, 1 m M; determining a second similarity between the target object and the mth candidate object according to the target first emotion category prediction value and the second emotion category prediction value, wherein the target first emotion category prediction value is the largest first emotion category prediction value among multiple first emotion category prediction values.

[0066] According to an embodiment of the present disclosure, the plurality of first emotion category prediction values ​​are probability values ​​for predicting that each group of heart ballistogalacti signal sub-data is of a certain emotion category.

[0067] For example, if the number of targets is p and there are 3 emotion categories, the first emotion category prediction values ​​corresponding to each emotion category are sorted in ascending order, and the first p / 3 larger first emotion category prediction values ​​are determined from the multiple first emotion category prediction values ​​corresponding to each emotion category, thereby obtaining p larger first emotion category prediction values, and the heart impact signal sub-data corresponding to each of the p larger first emotion category prediction values ​​are determined as the target heart impact signal sub-data.

[0068] According to an embodiment of the present disclosure, the physiological signals of the target object are sampled multiple times to obtain multiple groups of cardiac signal sub-data. The emotion recognition sub-model may be a classifier model.

[0069] According to an embodiment of the present disclosure, the target object does not have a real emotion category label. Since the credibility of the label obtained by predicting the target object based on the emotion recognition sub-model trained on the candidate object is not high, a high-confidence pseudo-label selection strategy is designed to select a high-confidence pseudo-label corresponding to the target object.

[0070] According to an embodiment of the present disclosure, an emotion recognition sub-model is jointly trained using the cardiac impact signal data and emotion category labels of M candidate objects, and each group of cardiac impact signal sub-data of the target object is processed based on the common emotion recognition sub-model to obtain a first emotion category prediction value. The first emotion category prediction value is a pseudo emotion category label for the target object, representing the probability that a single group of cardiac impact signal sub-data belongs to a certain emotion category.

[0071] According to an embodiment of the present disclosure, a maximum first emotion category prediction value is selected from multiple first emotion category prediction values, the maximum first emotion category prediction value is determined as a target first emotion category prediction value, and the category corresponding to the target first emotion category prediction value is a pseudo emotion category label.

[0072] According to an embodiment of the present disclosure, the second similarity between the target object and the mth candidate object is calculated: first, the target cardiac arrest signal sub-data is processed using the emotion recognition sub-model corresponding to the mth candidate object to obtain a second emotion category prediction value, and then the second similarity between the target object and the mth candidate object is determined based on the target first emotion category prediction value and the second emotion category prediction value.

[0073] In one embodiment, the second similarity As shown in formula (3):

[0074] (3);

[0075] in, Represents the mth candidate object, Characterize the target object, Represents the predicted value of the first emotion category of the target, Represents the predicted value of the second emotion category, Characterize the target heart attack signal sub-data, The emotion recognition sub-model representing the mth candidate object, Characterize the label similarity evaluation algorithm.

[0076] According to an embodiment of the present disclosure, a high-confidence pseudo-label selection strategy is designed to select target cardiac arrest signal sub-data corresponding to credible pseudo-labels, and then the target object and candidate objects are evaluated based on label similarity, providing high-confidence support for subsequent screening of associated reference objects.

[0077] According to an embodiment of the present disclosure, determining the first similarity between the two candidate objects using the heartbeat signal data and the emotion category labels of the two candidate objects may include: using a label similarity algorithm to process the heartbeat signal data and the emotion category labels of the two candidate objects to obtain the first similarity between the two candidate objects. Specifically, for the a-th candidate object and the b-th candidate object among the M candidate objects, using the emotion recognition sub-model corresponding to the a-th candidate object to process the heartbeat signal data of the b-th candidate object to obtain the third emotion category prediction value of the b-th candidate object, 1 M, 1 M; obtaining a first similarity between the ath candidate object and the bth candidate object according to the third emotion category prediction value of the bth candidate object and the emotion category label of the bth candidate object.

[0078] According to an embodiment of the present disclosure, the emotion category label of the bth candidate object is the true label emotion category of the bth candidate object.

[0079] In one embodiment, the first similarity As shown in formula (4):

[0080] (4);

[0081] in, Characterize the a-th candidate object, Characterize the bth candidate object, Characterize the emotion category label of the b-th candidate object, The predicted value of the third emotion category representing the b-th candidate object, The heartbeat signal data representing the b-th candidate object, The emotion recognition sub-model that represents the a-th candidate object.

[0082] Figure 3 An example schematic diagram of determining a reference object according to an embodiment of the present disclosure is shown.

[0083] like Figure 3 As shown, the candidate object set includes M candidate objects (candidate object 1, candidate object 2, ..., candidate object M), and similarity evaluation is performed based on the label similarity evaluation algorithm. Each candidate corresponds to the first similarity matrix 1 of M×1 dimension ( ); The first similarity matrix 2 of M×1 dimension corresponding to candidate object 2 and each candidate ( ); The candidate object M corresponds to the first similarity matrix M of M×1 dimension for each candidate ( ); The target object corresponds to the second similarity matrix of M×1 dimensions ( ), therefore, the M candidate objects obtain the first similarity matrix of M×M dimension, and the similarity matrix of M×(M+1) dimension is constructed according to the first similarity matrix of M×M dimension and the second similarity matrix of M×1 dimension . The M×(M+1)-dimensional similarity matrix As the feature of topic clustering, clustering algorithm is used to cluster M candidate objects and target objects, and the clustering results obtained with different cluster numbers are obtained ( ), the variance ratio values ​​of different clustering results are used to select q candidate objects that are significantly correlated with the target object from the candidate object set to form a reference object set.

[0084] According to an embodiment of the present disclosure, the emotion recognition method based on the cardiac impact signal also includes: based on the balanced distribution adaptive optimization function, performing feature alignment processing on the cardiac impact signal data of the target object and the reference object to obtain the conversion signal data of the reference object; using the emotion recognition model to process the cardiac impact signal data of the reference object to obtain the emotion recognition result of the target object, including: using the emotion recognition model to process the conversion signal data of the reference object to obtain the emotion recognition result of the target object.

[0085] According to an embodiment of the present disclosure, the cardiac ballistic signal data of the target object and the cardiac ballistic signal data of the reference object have the same feature space and label space, but the data distribution is different. It is necessary to use a balanced distribution adaptive optimization function (Balanced Distribution Adaptation, BDA) to align the respective features of the cardiac ballistic signal data of the target object and the reference object.

[0086] According to an embodiment of the present disclosure, the conversion signal data of the reference object is the heart attack signal data after feature alignment processing, and the conversion signal data of the reference object is processed using the emotion recognition model to obtain the emotion recognition result of the target object.

[0087] According to an embodiment of the present disclosure, based on a balanced distribution adaptive optimization function, feature alignment processing is performed on the cardiac impact signal data of the target object and the reference object to obtain the conversion signal data of the reference object, including: determining the target edge probability distribution and the target conditional probability distribution according to the cardiac impact signal data of the target object; determining the reference edge probability distribution and the reference conditional probability distribution according to the cardiac impact signal data of the reference object; using the balanced distribution adaptive optimization function to process the target edge probability distribution, the target conditional probability distribution, the reference edge probability distribution and the reference conditional probability distribution to obtain a conversion matrix; according to the conversion matrix and the cardiac impact signal data of the reference object, obtaining the conversion signal data of the reference object.

[0088] According to an embodiment of the present disclosure, the target object and the reference object respectively correspond to multiple groups of cardiac impact signal sub-data, the target marginal probability distribution is the overall probability distribution of the multiple groups of cardiac impact signal sub-data without considering the emotion category labels, and the target conditional probability distribution is the probability distribution of the multiple groups of cardiac impact signal sub-data under the condition of a certain emotion category label.

[0089] According to an embodiment of the present disclosure, the optimization goal of the balanced distribution adaptive optimization function is to simultaneously reduce the marginal probability distribution and the conditional probability distribution between the target object and the reference object according to certain weights to minimize the distribution difference.

[0090] According to an embodiment of the present disclosure, a conversion matrix is ​​obtained by solving in the process of realizing the optimization target of the balanced distribution adaptive optimization function; and then the conversion signal data of the reference object is obtained by multiplying the transposed matrix of the conversion matrix with the heart ballistic signal data of the reference object.

[0091] In one embodiment, the conversion signal data of the reference object As shown in formula (5):

[0092] (5);

[0093] in, represents the transpose of the transformation matrix, Ballistocardiographic signal data characterizing a set of reference subjects.

[0094] According to an embodiment of the present disclosure, the transposed matrix of the conversion matrix and the ballistocardiac signal data of the target object can be used to obtain the Perform multiplication calculation to obtain the conversion signal data of the target object , the conversion signal data of the target object Convert signal data to reference object The two probability distributions are aligned.

[0095] According to the embodiments of the present disclosure, the feature alignment process is divided into two parts: one is to align the feature distribution of the target object and the reference object under a given emotion category label; the other is to align the overall feature distribution of the target object and the reference object without considering the emotion category label, and perform emotion recognition based on the aligned features, thereby improving the generalization of the emotion recognition model.

[0096] According to an embodiment of the present disclosure, a balanced distribution adaptive optimization function is used to process a target edge probability distribution, a target conditional probability distribution, a reference edge probability distribution and a reference conditional probability distribution to obtain a transformation matrix, including: using a mean difference function to calculate the difference between the target edge probability distribution and the reference edge probability distribution to obtain an edge difference value; using a mean difference function to calculate the difference between the target conditional probability distribution and the reference conditional probability distribution to obtain a conditional difference value; using a balanced distribution adaptive optimization function to process the edge difference value and the conditional difference value to obtain a transformation matrix, wherein the balanced distribution adaptive optimization function is used to optimize the distribution difference between the heart attack signal data of the target object and the reference object respectively.

[0097] According to an embodiment of the present disclosure, the mean difference function may be a maximum mean difference function (Maximum Mean Discrepancy, MMD).

[0098] In one embodiment, the balanced distribution adaptive optimization function As shown in formula (6):

[0099] (6);

[0100] in, Characterize the balance factor, Characterize the target edge probability distribution, Characterize the reference marginal probability distribution, Characterize the target conditional probability distribution, Characterizes the reference conditional probability distribution, Represents the reference object set obtained by screening from the candidate object set, Characterize the target object, D Characterize the maximum mean difference function, D Characterize the edge difference value, D Characterize the condition difference value, A single set of heartbeat signal sub-data representing a reference object, A single set of cardiac ballistometry signal sub-data representing the target object.

[0101] According to the embodiment of the present disclosure, since the target object has no real emotion category label, It cannot be calculated, so the class conditional probability is used Approximately replace the .

[0102] According to an embodiment of the present disclosure, ,when When , the main role of reducing the marginal probability distribution in domain adaptation is When , reducing the conditional probability distribution plays a major role in domain adaptation.

[0103] According to an embodiment of the present disclosure, the balanced distribution adaptive optimization function such as formula (6) is expanded to obtain formula (7):

[0104] (7);

[0105] Among them, ns represents the total number of groups of reference objects, that is, the total number of sampling times, nt represents the total number of groups of target objects, and C represents the emotion category. The i-th group of cardiac ballistic signal sub-data representing the reference object, The j-th group of heart attack signal sub-data representing the target object, The number of groups of the heartbeat signal sub-data representing the reference object set belonging to the cth emotion category label, The number of groups of the heartbeat signal sub-data representing the target object belonging to the cth emotion category label, Represents the sample set that belongs to the cth emotion category label in the reference object set, Represents the sample set of the target object that belongs to the cth emotion category label.

[0106] According to an embodiment of the present disclosure, after performing matrix transformation and regularization transformation on formula (7), formula (8) is obtained:

[0107] (8);

[0108] Among them, X represents , The matrix composed of represents the regularization parameter, Characterize the transformation matrix, represents the identity matrix, Characterizes the matrix norm (Frobenius norm), , 1 is a matrix with 1 elements, , are all maximum mean difference matrices, where , , Characterize the i-th group of heart ballistic signal sub-data, Characterize the j-th group of cardiac ballistogalactigraphic signal sub-data.

[0109] According to an embodiment of the present disclosure, formula (8) is transformed into a Lagrangian function to obtain formula (9):

[0110] (9);

[0111] in, Characterize the Lagrangian operator, .

[0112] According to an embodiment of the present disclosure, formula (9) is solved to obtain the optimal conversion matrix A.

[0113] (10);

[0114] The optimal transformation matrix A is the d smallest eigenvectors of formula (10).

[0115] Figure 4 An example schematic diagram of obtaining emotion recognition results according to an embodiment of the present disclosure is shown.

[0116] like Figure 4 As shown, the cardiac impact signal data of the target object and the cardiac impact signal data of each of the multiple candidate objects are preprocessed respectively, and the preprocessing includes feature extraction and normalization processing. Then, based on the label similarity evaluation algorithm, similarity processing is performed on the cardiac impact signal data of the target object and the cardiac impact signal data of each of the multiple candidate objects. Then, based on the clustering selection algorithm, the candidate object associated with the target object is determined from the multiple candidate objects to obtain at least one reference object; based on the balanced distribution adaptive optimization function, feature alignment processing is performed on the cardiac impact signal data of the target object and the reference object to obtain the conversion signal data of the reference object; the conversion signal data of the reference object is processed using the emotion recognition model to obtain the emotion recognition result of the target object.

[0117] According to an embodiment of the present disclosure, an emotion recognition model includes N binary classifiers, where N is a positive integer, and the number of binary classifiers is determined based on the number of sample emotion categories. The emotion recognition model is trained based on the following operations: obtaining training samples, the training samples include conversion signal data of a sample reference object and a sample emotion category label of a sample target object, and the sample reference object is an object associated with the sample target object; performing feature extraction on the conversion signal data of the sample reference object to obtain signal features; using N binary classifiers to process the signal features respectively to obtain sample emotion category prediction values ​​corresponding to each binary classifier; performing statistics on the N sample prediction values ​​based on a statistical function to obtain a prediction score corresponding to each sample emotion category; determining the emotion category corresponding to the maximum prediction score among multiple prediction scores as the sample emotion recognition result, wherein the sample emotion recognition result represents the emotion category prediction result of the sample target object; training an initial emotion recognition model based on the sample emotion recognition result and the sample emotion category label to obtain an emotion recognition model.

[0118] According to an embodiment of the present disclosure, the number of binary classifiers is determined based on the permutations and combinations of the number of sample emotion categories. For example, if the sample emotion categories include negative, positive, and neutral categories, three binary classifiers need to be constructed. The first binary classifier is constructed based on the negative and positive categories, the second binary classifier is constructed based on the negative and neutral categories, and the third binary classifier is constructed based on the neutral and positive categories.

[0119] According to the embodiments of the present disclosure, the number of sample reference objects needs to be selected according to the training sample situation. In this solution, the highest accuracy is obtained by setting the number of sample reference objects to 9.

[0120] According to an embodiment of the present disclosure, the converted signal data of the sample reference object is a ballistocardiographic signal after feature alignment processing is performed on the sample reference object and the sample target object.

[0121] In one embodiment, a classifier is constructed for the uth emotion category and the vth emotion category. As shown in (11):

[0122] (11);

[0123] in, Represents the predicted value of the sample emotion category, Characterize the signal characteristics, A value of 1 represents the prediction score of the uth emotion category plus 1. A value of -1 means that the prediction score of the vth emotion category is increased by 1. Both represent classifier parameters, .

[0124] In one embodiment, the prediction score corresponding to the vth sample emotion category is As shown in (12):

[0125] (12);

[0126] in, Characterize the indicator function.

[0127] According to an embodiment of the present disclosure, the emotion category corresponding to the maximum prediction score among multiple prediction scores is determined as the sample emotion recognition result. The loss function can be used to calculate the loss value between the sample emotion recognition result and the sample emotion category label, and the parameters of the initial emotion recognition model can be adjusted according to the loss value to obtain the emotion recognition model.

[0128] According to the embodiment of the present disclosure, the recognition effect of the emotion recognition model is verified by the leave-one-subject-out (LOSO) method, that is, only one subject's data is left as the test set, and the rest of the subjects' data is used as the training set until all subjects have completed the test set. The evaluation indicators can be accuracy, precision, recall, etc.

[0129] According to an embodiment of the present disclosure, an initial emotion recognition model is trained based on conversion signal data of a sample reference object, and the initial emotion recognition model is adjusted according to the sample emotion category labels and sample emotion recognition results of a sample target object to obtain an optimized emotion recognition model, thereby improving the accuracy and generalization of the model in recognizing the emotions of the target object, avoiding the introduction of knowledge that is irrelevant or even conflicting with the target object, and reducing the risk of negative transfer.

[0130] Based on the above-mentioned emotion recognition method based on cardiac impact signal, the present disclosure also provides an emotion recognition device based on cardiac impact signal. Figure 5 The device is described in detail.

[0131] Figure 5 A structural block diagram of an emotion recognition device based on a cardiac impulse signal according to an embodiment of the present disclosure is shown.

[0132] like Figure 5 As shown, the emotion recognition device 500 based on the cardiac impulse signal of this embodiment includes an enhancement module 510 , an extraction module 520 and a recognition module 530 .

[0133] The acquisition module 510 is used to acquire the ballistocardiographic signal data of the target object and the ballistocardiographic signal data of each of the multiple candidate objects. In one embodiment, the acquisition module 510 can be used to perform the operation S110 described above, which will not be described in detail here.

[0134] The screening module 520 is used to perform similarity processing on the cardiac perturbation signal data of the target object and the cardiac perturbation signal data of each of the multiple candidate objects, determine the candidate object associated with the target object from the multiple candidate objects, and obtain at least one reference object. In one embodiment, the screening module 520 can be used to perform the operation S120 described above, which will not be repeated here.

[0135] The recognition module 530 is used to process the ballistocardiographic signal data of the reference object using the emotion recognition model to obtain the emotion recognition result of the target object. In one embodiment, the recognition module 530 can be used to perform the operation S130 described above, which will not be repeated here.

[0136] According to an embodiment of the present disclosure, the screening module 520 includes a first screening submodule, a second screening submodule, a third screening submodule, a fourth screening submodule and a fifth screening submodule.

[0137] The first screening submodule is used to determine the first similarity between every two candidate objects among the M candidate objects by using the heart ball signal data and emotion category labels of the two candidate objects respectively, and obtain a first similarity matrix of M×M dimensions.

[0138] The second screening submodule is used to determine the second similarity between the target object and each candidate object among the M candidate objects by using the cardiac ballistic signal data of the candidate object and the cardiac ballistic signal data of the target object to obtain a second similarity matrix of M×1 dimensions.

[0139] The third screening submodule is used to perform clustering processing on the M candidate objects and the target object according to the first similarity matrix of M×M dimensions and the second similarity matrix of M×1 dimensions to obtain a plurality of clusters.

[0140] The fourth screening submodule is used to determine the cluster where the target object is located from the multiple clusters to obtain the target cluster.

[0141] The fifth screening submodule is configured to determine the candidate objects included in the target cluster as candidate objects associated with the target object, and obtain at least one reference object.

[0142] According to an embodiment of the present disclosure, the second screening submodule includes a first screening unit, a second screening unit and a third screening unit.

[0143] The first screening unit is used to process the multiple groups of cardiac impact signal sub-data respectively by using the emotion recognition sub-model corresponding to the M candidate objects, so as to obtain the first emotion category prediction value corresponding to each of the multiple groups of cardiac impact signal sub-data.

[0144] The second screening unit is used for processing the target cardiac impact signal sub-data by using the emotion recognition sub-model corresponding to the m-th candidate object among the M candidate objects to obtain a second emotion category prediction value, wherein the target cardiac impact signal sub-data is the cardiac impact signal sub-data corresponding to the target number of first emotion category prediction values ​​determined based on a preset sorting rule among the multiple first emotion category prediction values, 1 m M.

[0145] The third screening unit is used to determine the second similarity between the target object and the mth candidate object according to the target first emotion category prediction value and the second emotion category prediction value, wherein the target first emotion category prediction value is the largest first emotion category prediction value among multiple first emotion category prediction values.

[0146] According to an embodiment of the present disclosure, the first screening submodule includes a fourth screening unit and a fifth screening unit.

[0147] The fourth screening unit is used to process the heart attack signal data of the bth candidate object using the emotion recognition sub-model corresponding to the ath candidate object and the bth candidate object among the M candidate objects, so as to obtain a third emotion category prediction value of the bth candidate object.

[0148] The fifth screening unit is used to obtain the first similarity between the ath candidate object and the bth candidate object according to the third emotion category prediction value of the bth candidate object and the emotion category label of the bth candidate object.

[0149] According to an embodiment of the present disclosure, the emotion recognition device 500 based on the ballistocardiographic signal further includes an optimization module.

[0150] The optimization module is used to perform feature alignment processing on the heart ballistic signal data of the target object and the reference object based on the balanced distribution adaptive optimization function to obtain the conversion signal data of the reference object.

[0151] According to an embodiment of the present disclosure, the optimization module includes a first optimization submodule, a second optimization submodule, a third optimization submodule and a fourth optimization submodule.

[0152] The first optimization submodule is used to determine the target edge probability distribution and the target conditional probability distribution according to the ballistocardiographic signal data of the target object.

[0153] The second optimization submodule is used to determine a reference edge probability distribution and a reference conditional probability distribution according to the ballistocardiographic signal data of a reference object.

[0154] The third optimization submodule is used to process the target edge probability distribution, the target conditional probability distribution, the reference edge probability distribution and the reference conditional probability distribution by using the balanced distribution adaptive optimization function to obtain a conversion matrix.

[0155] The fourth optimization submodule is used to obtain the conversion signal data of the reference object according to the conversion matrix and the ballistocardi signal data of the reference object.

[0156] According to an embodiment of the present disclosure, the third optimization submodule includes a first optimization unit, a second optimization unit and a third optimization unit.

[0157] The first optimization unit is used to calculate the difference between the target edge probability distribution and the reference edge probability distribution by using a mean difference function to obtain an edge difference value.

[0158] The second optimization unit is used to calculate the difference between the target conditional probability distribution and the reference conditional probability distribution by using the mean difference function to obtain a conditional difference value.

[0159] The third optimization unit is used to process the edge difference value and the conditional difference value using a balanced distribution adaptive optimization function to obtain a conversion matrix, wherein the balanced distribution adaptive optimization function is used to optimize the distribution difference between the heart ballistic signal data of the target object and the reference object respectively.

[0160] According to an embodiment of the present disclosure, any multiple modules among the modules, submodules, units, and subunits can be combined into one module for implementation, or any one of the modules can be split into multiple modules. Alternatively, at least part of the functions of one or more of these modules can be combined with at least part of the functions of other modules and implemented in one module. According to an embodiment of the present disclosure, at least one of the modules, submodules, units, and subunits can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application-specific integrated circuit (ASIC), or can be implemented by hardware or firmware such as any other reasonable way of integrating or packaging the circuit, or implemented in any one of the three implementation methods of software, hardware, and firmware, or in any appropriate combination of any of them. Alternatively, at least one of the modules, submodules, units, and subunits can be at least partially implemented as a computer program module, and when the computer program module is run, the corresponding function can be executed.

[0161] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram may represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of the code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box may also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of boxes in the block diagram or flowchart, can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions. It can be understood by those skilled in the art that the features recorded in the various embodiments of the present disclosure can be combined and / or combined in a variety of ways, even if such a combination or combination is not explicitly recorded in the present disclosure. In particular, without departing from the spirit and teaching of the present disclosure, the features described in the various embodiments of the present disclosure may be combined and / or combined in a variety of ways. All of these combinations and / or combinations fall within the scope of the present disclosure.

[0162] The embodiments according to the present disclosure are described above. However, these embodiments are only for the purpose of illustration, and are not intended to limit the scope of the present disclosure. Although the embodiments are described above, this does not mean that the measures in the various embodiments cannot be used in combination to advantage. Without departing from the scope of the present disclosure, those skilled in the art may make a variety of substitutions and modifications, which should all fall within the scope of the present disclosure.

Claims

1. An emotion recognition method based on cardiac impulse signal, characterized in that: The method comprises: Acquire the ballistocardiographic signal data of the target object and the ballistocardiographic signal data of each of the plurality of candidate objects; Performing similarity processing on the cardiac ballistic signal data of the target object and the cardiac ballistic signal data of each of the plurality of candidate objects, determining a candidate object associated with the target object from the plurality of candidate objects, and obtaining at least one reference object; and The heartbeat signal data of the reference object is processed using the emotion recognition model to obtain the emotion recognition result of the target object.

2. The method according to claim 1, characterized in that The candidate objects include M, where M is a positive integer greater than 1, and the candidate objects are marked with emotion category labels; Performing similarity processing on the cardiac signal data of the target object and the cardiac signal data of each of the plurality of candidate objects, determining a candidate object associated with the target object from the plurality of candidate objects, and obtaining at least one reference object comprises: For every two candidate objects among the M candidate objects, determine a first similarity between the two candidate objects using the heartbeat signal data and the emotion category labels of the two candidate objects respectively, to obtain a first similarity matrix of M×M dimensions; For each candidate object among the M candidate objects, determine a second similarity between the target object and the candidate object using the cardiac ballistic signal data of the candidate object and the cardiac ballistic signal data of the target object to obtain a second similarity matrix of M×1 dimensions; According to the first similarity matrix of M×M dimensions and the second similarity matrix of M×1 dimensions, clustering is performed on the M candidate objects and the target object to obtain a plurality of clusters; Determine the cluster where the target object is located from multiple clusters to obtain a target cluster; The candidate objects included in the target cluster are determined as candidate objects associated with the target object, and at least one reference object is obtained.

3. The method according to claim 2, characterized in that The heartbeat signal data of the target object includes multiple groups of heartbeat signal sub-data, the M candidate objects collectively correspond to an emotion recognition sub-model, and each candidate object has a corresponding emotion recognition sub-model; The determining the second similarity between the target object and the candidate object by using the cardiac ballistic signal data of the candidate object and the cardiac ballistic signal data of the target object comprises: Using the emotion recognition sub-models corresponding to the M candidate objects, respectively processing the multiple groups of cardiac impact signal sub-data, to obtain the first emotion category prediction values ​​corresponding to the multiple groups of cardiac impact signal sub-data; For the mth candidate object among the M candidate objects, the target cardiac impact signal sub-data is processed by using the emotion recognition sub-model corresponding to the mth candidate object to obtain a second emotion category prediction value, wherein the target cardiac impact signal sub-data is the cardiac impact signal sub-data corresponding to the target number of first emotion category prediction values ​​determined based on a preset sorting rule among the plurality of first emotion category prediction values, 1 m M; A second similarity between the target object and the mth candidate object is determined according to a target first emotion category prediction value and the second emotion category prediction value, wherein the target first emotion category prediction value is the largest first emotion category prediction value among multiple first emotion category prediction values.

4. The method according to claim 2, characterized in that: Determining the first similarity between the two candidate objects by using the respective ballistocardiographic signal data and emotion category labels of the two candidate objects comprises: For the ath candidate object and the bth candidate object among the M candidate objects, the heartbeat signal data of the bth candidate object is processed by using the emotion recognition sub-model corresponding to the ath candidate object to obtain the third emotion category prediction value of the bth candidate object, 1 M, 1 M; A first similarity between the ath candidate object and the bth candidate object is obtained according to the third emotion category prediction value of the bth candidate object and the emotion category label of the bth candidate object.

5. The method according to claim 1, characterized in that: The method further comprises: Based on the balanced distribution adaptive optimization function, feature alignment processing is performed on the ballistocardiographic signal data of the target object and the reference object to obtain the conversion signal data of the reference object; The using the emotion recognition model to process the heartbeat signal data of the reference object to obtain the emotion recognition result of the target object includes: The converted signal data of the reference object is processed using the emotion recognition model to obtain the emotion recognition result of the target object.

6. The method according to claim 5, characterized in that The performing feature alignment processing on the ballistocardiographic signal data of the target object and the reference object based on the balanced distribution adaptive optimization function to obtain the conversion signal data of the reference object comprises: Determining a target edge probability distribution and a target conditional probability distribution according to the ballistocardiographic signal data of the target object; Determining a reference edge probability distribution and a reference conditional probability distribution according to the ballistocardiographic signal data of the reference object; Processing the target edge probability distribution, the target conditional probability distribution, the reference edge probability distribution and the reference conditional probability distribution using a balanced distribution adaptive optimization function to obtain a conversion matrix; The conversion signal data of the reference object is obtained according to the conversion matrix and the ballistocardiographic signal data of the reference object.

7. The method according to claim 6, characterized in that The use of the balanced distribution adaptive optimization function to process the target edge probability distribution, the target conditional probability distribution, the reference edge probability distribution and the reference conditional probability distribution to obtain a conversion matrix includes: Calculating the difference between the target edge probability distribution and the reference edge probability distribution using a mean difference function to obtain an edge difference value; Calculating the difference between the target conditional probability distribution and the reference conditional probability distribution using the mean difference function to obtain a conditional difference value; The edge difference value and the conditional difference value are processed by using a balanced distribution adaptive optimization function to obtain a conversion matrix, wherein the balanced distribution adaptive optimization function is used to optimize the distribution difference between the heart ballistic signal data of the target object and the reference object respectively.

8. The method according to claim 1, characterized in that The emotion recognition model includes N binary classifiers, where N is a positive integer. The number of the binary classifiers is determined based on the number of sample emotion categories. The emotion recognition model is trained based on the following operations: Acquire a training sample, wherein the training sample includes conversion signal data of a sample reference object and a sample emotion category label of a sample target object, wherein the sample reference object is an object associated with the sample target object; Extracting features from the converted signal data of the sample reference object to obtain signal features; Using N binary classifiers to process the signal features respectively, and obtaining a sample emotion category prediction value corresponding to each of the binary classifiers; Based on a statistical function, statistics are performed on the N sample prediction values ​​to obtain a prediction score corresponding to each sample emotion category; Determining the emotion category corresponding to the maximum prediction score among the plurality of prediction scores as a sample emotion recognition result, wherein the sample emotion recognition result represents the emotion category prediction result of the sample target object; An initial emotion recognition model is trained according to the sample emotion recognition results and the sample emotion category labels to obtain the emotion recognition model.

9. The method according to claim 1, characterized in that: The heartbeat signal data includes at least one of the following: Heart rate variability data, respiratory change data, beat-to-beat heart rate statistics data, and beat-to-beat heart rate nonlinearity data.

10. An emotion recognition device based on a cardiac impulse signal, the device comprising: An acquisition module, used to acquire the ballistocardiographic signal data of the target object and the ballistocardiographic signal data of each of the multiple candidate objects; A screening module, configured to perform similarity processing on the cardiac perturbation signal data of the target object and the cardiac perturbation signal data of each of the plurality of candidate objects, determine a candidate object associated with the target object from the plurality of candidate objects, and obtain at least one reference object; The recognition module is used to process the heart ballistic signal data of the reference object using the emotion recognition model to obtain the emotion recognition result of the target object.

Images