A method for detecting the largest cluster in EEG brain networks for emotion recognition

Abstract

The present invention belongs to the field of signal processing technology research, and provides an EEG brain network maximum cluster detection method for realizing emotion recognition, which solves the problem that the interaction between brain regions is unclear and the accuracy of emotion recognition is not high in existing brain network related research. First, by quantifying the phase synchronization degree between EEG signals, a functional brain network based on phase lock value is constructed; then, an improved Bron‑Kerbosch algorithm is used to detect the maximum cluster in the brain network. The method uses a rich‑club structure to screen effective network nodes. After all nodes are traversed, the maximum cluster with the largest number of nodes is obtained, that is, the maximum cluster of the brain network, which represents the most important intrinsic information processing module in the brain network; finally, the maximum cluster is thresholded, and topological attributes are extracted as features for emotion recognition. By using this method to construct an emotion recognition system, the features most relevant to emotional activities can be extracted, and the accuracy of emotion recognition can be improved.

Description

Technical Field

[0001] The present invention belongs to the research field of signal processing technology, and more specifically, relates to an EEG brain network maximum cluster detection method for realizing emotion recognition, which can be used to classify and identify the emotional state of an individual. Background Art

[0002] Nowadays, the field of affective computing has become a focus of the rapidly developing artificial intelligence research topic, and emotion recognition is the most valuable research point in affective computing. Emotion recognition is to identify the emotional state of a person by obtaining physiological or non-physiological signals, that is, to automatically classify the user's emotional state according to the input data, so as to make human-computer interaction more intelligent. Common emotion recognition methods can be divided into two categories: one is emotion recognition methods based on non-physiological signals such as voice, facial expressions and postures, and the other is emotion recognition based on physiological signals, mainly including peripheral nervous system and brain signals. Due to the differences between different cultures and countries, emotion recognition using physiological signal characteristics will produce more accurate results than non-physiological signals.

[0003] The human brain has the most advanced central nervous system, which plays a major role in the process of human emotion generation. In recent years, the development of non-invasive brain signal acquisition technology has promoted researchers' understanding of the relationship between brain mechanisms and emotions. Among various physiological signals, electroencephalogram (EEG) is a complex and unstable random signal generated in the cerebral cortex due to the electrophysiological activity of brain nerve cells. It is a comprehensive reflection of human brain neuronal activity, contains rich physiological and disease information, and can provide real brain mapping of spatiotemporal information about brain function. Non-invasive EEG has the advantages of convenient acquisition, high temporal resolution, real-time monitoring, and easy acquisition. It is captured directly from the cerebral cortex, which is conducive to reflecting the psychological state of people. Therefore, it has become one of the most widely used technologies for studying the brain mechanism of emotional arousal process.

[0004] However, the brain is a complex dynamic system. The dynamic changes in brain structure when different emotions are generated are elusive for researchers. The channel information collected by discrete electrodes in EEG equipment is not enough to reflect the coupling relationship between brain regions. In recent years, with the development of brain science, computational science, complex systems and other fields, EEG brain network research combined with complex network theory has gradually become a new research hotspot. Combining complex network analysis theory to construct EEG functional brain networks, study the brain function information processing mechanism of different human emotions, and the differences and characteristics of human brain functional networks under different emotional states will help people better understand the mechanisms of human brain information processing and perceptual feedback. Bressler et al. pointed out that cognition is generated by the dynamic interaction of distributed brain regions running in a large-scale network rather than the isolated operation of a single brain region. However, existing studies do not fully analyze the topological properties of brain networks, and the lack of information on the spatial communication between brain regions makes it difficult to extract effective key features in the process of emotion recognition, and the emotion recognition rate is not high. Summary of the invention

[0005] In view of the above situation, the present invention proposes an EEG brain network maximum cluster detection method for realizing emotion recognition, which can solve the problems of unclear interaction between brain regions and low emotion recognition accuracy in existing brain network related research.

[0006] The present invention provides an EEG brain network maximum cluster detection method for realizing emotion recognition, characterized in that it comprises the following steps:

[0007] Step 1: Construct an EEG-based functional brain network by estimating the phase synchronization relationship between EEG channel signals, define the scalp electrode positions as nodes, and the phase locking value (PLV) as the connection edge to measure the correlation between EEG channels. The weight represents the degree of phase synchronization between brain regions under emotional activity.

[0008] Step 2: Calculate the node degree attributes of all nodes in the brain network;

[0009] Step 3: For each node in the brain network, calculate the weighted rich-club coefficient φ corresponding to its node degree k w (k), k represents the node degree of the node in the network, and w represents the weighted;

[0010] Step 4: Calculate the normalized rich-club coefficient φ norm w (k), norm stands for normalized;

[0011] Step 5: Filter the nodes participating in the maximum clique detection. If the normalized rich-club coefficient value is greater than 1, it indicates that a rich-club structure exists. Keep the nodes whose rich-club coefficients corresponding to the node degrees are greater than 1, and delete the rest of the nodes.

[0012] Step 6: Arrange the brain region nodes in order from small to large node degrees, and give priority to processing nodes with low node degrees during the algorithm execution, so that the calculation of the largest group in the PLV brain network can be completed with as few recursions as possible;

[0013] Step 7: Use the Bron-Kerbosch algorithm to detect the largest clusters in the brain network;

[0014] Step 8: Set the maximum cluster with the largest number of nodes as the final maximum cluster, which represents the key structure of the functional brain network under the task state;

[0015] Step 9: Threshold the largest cluster of the brain network according to the sparsity of 50% to obtain the maximum cluster structure after thresholding;

[0016] Step 10: Extract the topological features of the largest cluster structure of the brain network;

[0017] Step 11: Construct a complete emotion recognition system for classification and recognition.

[0018] Compared with the prior art, the EEG brain network maximum cluster detection method for realizing emotion recognition described in the technical solution has the following beneficial effects:

[0019] (1) Utilize the rich-club structure in the brain network to screen and construct effective nodes of the largest cluster, increase the accuracy of the results, and reduce the time cost;

[0020] (2) The Bron-Kerbosch maximum clique detection algorithm is optimized to minimize the number of algorithm recursions and reduce the time complexity of the algorithm, thereby improving efficiency;

[0021] (3) The emotion recognition system is constructed based on the largest cluster of the brain network. The topological features of the largest cluster that best represents the information interaction pattern when the brain processes emotional activities are extracted for emotion recognition, which can increase the accuracy of emotion recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 Flowchart of the implementation of a maximum cluster detection method for EEG brain networks for emotion recognition DETAILED DESCRIPTION

[0023] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the specific implementation modes of the present invention are further described in detail below in conjunction with the accompanying drawings.

[0024] EEG brain network maximum cluster detection method. Figure 1 As shown, the method comprises the following steps:

[0025] Constructing PLV brain network: constructing EEG-based functional brain network by estimating the phase synchronization relationship between EEG channel signals, defining scalp electrode positions as nodes, and phase lock values ​​as connecting edges to measure the correlation between EEG channels. The weights represent the degree of phase synchronization between brain regions under emotional activity.

[0026] Calculate the node degree attributes of all nodes in the brain network;

[0027] For each node in the brain network, calculate the weighted rich-club coefficient φ corresponding to its node degree k w (k);

[0028] Calculate the normalized rich-club coefficient φ norm w (k): For each network, construct 1000 random networks that have the same degree distribution and number of network connections as the PLV network, then calculate the rich-club coefficient of each random network under different given values ​​of k, and finally obtain the average value φ of the coefficients of these 1000 random networks random w The normalized rich-club coefficient is given by φ w (k) and φ random w It is expressed as the ratio of

[0029] Filter the nodes participating in the maximum clique detection: If the normalized rich-club coefficient value is greater than 1, it indicates that a rich-club structure exists. Keep the nodes whose rich-club coefficients corresponding to the node degrees are greater than 1, and delete the rest of the nodes.

[0030] Node sorting: Arrange the brain region nodes in order from small to large node degrees, and give priority to processing nodes with low node degrees during the algorithm execution, so that the calculation of the largest group in the PLV brain network can be completed with as few recursions as possible;

[0031] Use the Bron-Kerbosch algorithm to detect brain network maximal clusters;

[0032] The maximum cluster with the largest number of nodes is set as the final maximum cluster, which represents the key structure of the functional brain network under the task state;

[0033] The maximum cluster of the brain network is thresholded at 50% sparsity to obtain the maximum cluster structure after thresholding.

[0034] Extracting topological features of the largest cluster structure of brain networks.

[0035] Figure 1 The two steps indicated by number 1 (calculating node degree and sorting nodes) are optimization operations for Bron-Kerbosch. Since the node degree represents the number of edges associated with this node, the nodes are sorted based on the node degree so that nodes with small node degrees enter the iteration process first. This optimization idea effectively avoids too many redundant steps during the execution of the algorithm.

[0036] Figure 1 The three steps indicated by number 2 (calculating the rich-club coefficient, calculating the normalized rich-club coefficient, and screening nodes) are intended to ensure that the largest cluster detected in the end comes from the presence of a rich-club structure in the network. This is because the emergence of a rich-club structure represents that the brain regions in the brain network are interconnected and work together to process specific tasks. At this time, the largest cluster detected is the key architecture that best represents the brain's processing of information interaction.

[0037] Figure 1 The step indicated by number 3 (maximum cluster thresholding) uses thresholding to remove weak connections, sets a sparsity of 50% to filter the connected edges in the maximum cluster, and leaves only the top 50% of the connected edges with larger weights. This processing method ensures that the maximum cluster structure no longer has the fully connected characteristic while retaining effective strong connections, laying the foundation for subsequent topological feature extraction.

[0038] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A method for detecting the maximum cluster of EEG brain networks for emotion recognition, characterized in that: The following steps are involved: Step 1: By estimating the phase synchronization relationship between EEG channel signals, a functional brain network based on EEG is constructed. The scalp electrode positions are defined as nodes, and the phase locking value (PLV) is used as the connection edge to measure the correlation between EEG channels. The weight represents the degree of phase synchronization between brain regions under emotional activity. Step 2: Calculate the node degree attributes of all nodes in the brain network; Step 3: For each node in the brain network, calculate the weighted rich-club coefficient φ corresponding to its node degree k w (k), k represents the node degree of the node in the network, and w represents the weighted; Step 4: Calculate the normalized rich-club coefficient φ norm w (k), norm stands for normalized; Step 5: Filter the nodes participating in the maximum clique detection. If the normalized rich-club coefficient value is greater than 1, it indicates that a rich-club structure exists. Keep the nodes whose rich-club coefficients corresponding to the node degrees are greater than 1, and delete the rest of the nodes. Step 6: Arrange the brain region nodes in order from small to large node degrees, and give priority to processing nodes with low node degrees during the algorithm execution, so that the calculation of the largest group in the PLV brain network can be completed with as few recursions as possible; Step 7: Use the Bron-Kerbosch algorithm to detect the largest clusters in the brain network; Step 8: Set the maximum cluster with the largest number of nodes as the final maximum cluster, which represents the key structure of the functional brain network under the task state; Step 9: threshold the largest cluster of the brain network according to the sparsity of 50% to obtain the largest cluster structure after thresholding; Step 10: Extract the topological features of the largest cluster structure of the brain network; Step 11: Construct a complete emotion recognition system for classification and recognition.

2. The EEG brain network maximum cluster detection method for realizing emotion recognition according to claim 1, characterized in that: In step 4, the normalized rich-club coefficient φ is calculated norm w The method of (k) is as follows: for each network, 1000 random networks are constructed, which have the same degree distribution and number of network connections as the PLV network. Then the rich-club coefficient of each random network is calculated under different given values ​​of k. Finally, the average value of the rich-club coefficient of these 1000 random networks is obtained. random w , random represents random, and the normalized rich-club coefficient is φ w (k) and φ random w The ratio is expressed as: Among them, φ norm w (k) represents the normalized rich-club coefficient, φ w (k) represents the weighted rich-club coefficient, φ random w (k) represents the average of the rich-club coefficients of 1000 random networks.

3. The EEG brain network maximum cluster detection method for realizing emotion recognition according to claim 1, characterized in that: The topological features of the largest cluster of the brain network extracted in step 10 specifically include functional integration measures calculated by features such as characteristic path length and global efficiency, and functional separation measures calculated by features such as clustering coefficient, local efficiency, transitivity, and modularity.

Images