1. Nghiên cứu và phân loại tín hiệu điện não (EEG)
- An survey on Ensemble Method to classify Electroencephalogram
This research focuses on the feasibility of synthetic algorithms, including Boosted Trees, Bagged Trees, Subspace KNN, Subspace Discriminant, RUSBoosted Trees for identifying brain wave signal patterns. With two datasets used, it is the one that measures the four types of human emotions (valence, arousal, dominance, like). The receiver consists of 11 states consisting of groups of mechanical, normal, and thinking signals. The research focuses on researching the above algorithms, using the wavelet transform to find out the signal's characteristics, then classifying, comparing the results, improving, and reaching a conclusion.
- Application of LSTM algorithm in classification electroencephalogram
This research aims to recognize emotion from different participants using the electroencephalogram by using Long short term memory networks. For the processing signal step, with the DEAP dataset as the data input, we use the Differential Entropy to calculate the signal's complexness. Then all these features are divided into a small frame with equal lengths. Finally, we combine all these frames into a variety of videos of 32 participants. Our result focus on the accuracy and time consuming for the training of the different architecture of the network, then we aim to find the suitable variable for the long short term memory network to classify the emotion of human.
- Classification of Electroencephalography using Neural Network Algorithms
Electroencephalography is one of the most promising methods in the field of brain-computer interfaces due to its full time-domain resolution and the availability of advanced and portable sensor technology. This research attempts to use the electroencephalography signal based on Physiological Signals extracted from the Database for Emotion Analysis to classify the emotion of the subjects by classifier neural network algorithms. In order to maintain spatial information among channels, we used a three-dimensional (3D) model built from the EEG signal segment to display their features at different frequency bands. In addition, two types of neural network including 3D convolution neural network and hybrid deep learning one (hybrid network) were applied to train and test its ability of emotion states classification. As a result, the 3D hybrid network gave the most efficient classification with an accuracy of around 80%, which was better than other algorithms e.g. Support Vector Machine, Navie Bayes, Convolution Neural Network. Furthermore, our results also showed that the accuracies were achieved differently at various frequency bands, in which beta band gave the highest accuracy. Combining signals of different frequencies helped to improve the classification efficiency
2. Nghiên cứu và phân loại tín hiệu điện cơ (EMG)
- A Study of Finger Movement Classification Based On 2-sEMG Channels
This research focuses on the feasibility of synthetic algorithms, including Boosted Trees, Bagged Trees, Subspace KNN, Subspace Discriminant, RUSBoosted Trees for identifying brain wave signal patterns. With two datasets used, it is the one that measures the four types of human emotions (valence, arousal, dominance, like). The receiver consists of 11 states consisting of groups of mechanical, normal, and thinking signals. The research focuses on researching the above algorithms, using the wavelet transform to find out the signal's characteristics, then classifying, comparing the results, improving, and reaching a conclusion.
