EEE emotion recognition: the brain function connection network is combined with local activation information.

In this paper, the author Zhang Weiwei's school Northeast University graduate research direction emotion recognition introduction the change of EEG power spectrum is usually activated by the change of motivation emotional state.in recent years, it has been found that the difference of EEG spectrum caused by different emotional states is obvious in the alpha band of EEG, and can be captured in the anterior part of the brain.in addition, emotional response is also related to the spectral changes between different brain regions. Therefore, the changes of EEG spectrum in different brain regions can predict different emotional states of subjects.as mentioned above, the activation mode can capture the power difference between different brain regions when they feel different emotions, and the connection mode between different brain regions can show the information interaction process of the brain during emotional processing.at present, there are few researches on emotion recognition using the pattern fusion of the two compensation information.in this paper, the information network transmission patterns of different brain regions and the differences of brain activation are integrated to improve the ability of emotion recognition.methods: an emotion related brain network with phase locking value was constructed, and the compensation activation information and connection information were combined for emotion recognition by multi feature fusion method.paper method} Figure 1. The activation mode of emotion recognition process mainly reflects the energy difference of subjects in different emotional states.the author mainly studies the performance of power spectral density (PSD), differential entropy (DE), dasm, RASM, ASM and dcau under different emotions, as shown in Figure 2.among them, we can see that there is a significant difference between beta and gamma bands.} Fig. 2. The connection mode of EEG activation modes under different emotions is not a simple linear relationship between different brain regions. In this paper, the author proposes a phase locking (PLV) method to capture nonlinear phase synchronization and statistically measure the phase synchronization of two brain regions.the definition of PLV is as follows: in order to construct brain function network map, 32 electrodes are used as network nodes, and four kinds of EEG based network patterns (ENP) are extracted by using the knowledge of graph theory.the clustering index quantifies the ratio of the number of connections existing between the nearest neighbors of a node to the maximum number of possible connections. The clustering index of node I is defined in: SP length represents the minimum number of edges that must be traversed from one node to another.where g-efficient measures the extent to which information is propagated throughout the network and is the reciprocal of the harmonic mean of the minimum path length between any two nodes.l-efficient measures the regional specialization ability of the network by observing the good connection of its sub networks.} Fig. 3. The pattern fusion activation pattern of brain function network under different emotions captures the local activities that respond to emotion, while the connection pattern mines the interaction between related brain regions.the difficulty of fusing the two patterns is that the redundant information in the two feature types may affect the recognition effect unexpectedly. in this work, F-score and classifier dependency structure are combined to realize feature selection, as shown in Fig. 1, which not only ensures the resolution of the selected feature subset, but also saves the calculation. the F-score of the i-th feature is defined as: the larger the f (I), the stronger the discrimination ability of the corresponding feature. for the descending order of all F-score values, this paper applies the support vector machine (SVM) and graph regularized extreme value learning machine (gelm) model, and uses 10 times cross validation scheme to select the generalized features based on the training set. the steps of feature selection are as follows: 1. Calculate the F-score values of all features in the training set, sort them in descending order, define them as F, and initialize a feature subset fsel to be empty. defines the best classification result as MAXR and sets it to 0. initializes the optimal feature index to null. 2. Select the feature vector with the highest f score from F and add the selected vector to fsel. 3. 4. Repeat steps 2 to 4 until the traversal is completed. } Fig. 4. The feature selection process maximizes the above loss function of the recognizer, and the feature extraction process will produce domain invariant features, thus reducing the domain differences in emotion recognition. results in order to evaluate the performance of the proposed method, experiments were conducted on three benchmark emotion databases based on multimodality, namely mahnob-hci, deap and seed. in essence, the performance improvement of multimodal information method is due to the use of compensation information obtained from activation mode and connection mode. the experimental results are shown in Fig. 5. Compared with the single-mode method, the multimodal method can improve the recognition performance. in addition to improving the performance of the fusion strategy, we also need to pay attention to two other aspects that affect emotion recognition. One is that different accuracy is observed on different data sets, and seed data set performs best. another point is that due to different design purposes, different feature selection methods may lead to differences in the best feature subset. Experimental results conclusion in this work, the author proposes to combine the topological structure of brain network with the activation pattern of power spectrum based on EEG data for emotion recognition. this method not only captures the local emotional response activities, but also explores the interaction between the relevant brain regions. click the following title to view more previous contents: spatiotemporal hierarchical neural network EEG emotion recognition model emotionmeter: more accurate recognition of human emotions ICLR 2020: from de-noising from encoder to generative model variational inference How can we make more high-quality content reach the reader group in a shorter path and shorten the cost of finding high-quality content? The answer is: people you don't know. there are always people you don't know who know what you want to know. paperweekly may serve as a bridge to promote scholars and academic inspiration from different backgrounds and directions to collide with each other, and burst out more possibilities. paperweekly encourages university laboratories or individuals to share all kinds of high-quality content on our platform, which can be the interpretation of the latest papers, learning experience or technical dry goods. we have only one purpose, to make knowledge flow. 📝 Contribution criteria: · the manuscript is a personal original work, and the author's personal information (name + school / work unit + education / position + research direction) should be indicated in the contribution. If the article is not the first one, please remind and attach all the published links when submitting the article. · paperweekly defaults that every article is the first one, and the "original" logo will be added 📬 Submission email: · submission email: hr@paperweekly.site Please send the pictures of all articles separately in the attachment. Please leave the instant contact information (wechat or mobile phone) so that we can communicate with the author when editing and publishing 🔍 Now, we can also be found in Zhihu. Go to the homepage of Zhihu, search for "paperweekly" and click "follow" to subscribe to our column. About paperweekly, paperweekly is an academic platform for recommending, interpreting, discussing and reporting the achievements of advanced papers on artificial intelligence. if you study or engage in AI, welcome to click on the "official account" in the background of public numbers, and the little assistant will bring you into the communication group of PaperWeekly. ▽ click | read the original | download the paper