Researchers from University of Damascus developed a Convolutional Neural Network (CNN) architecture that offers robust results for both subject-dependent and subject-independent human emotion recognition
Emotion recognition is a vital part of Active and Assisted Living (AAL), Driver Assistance Systems (DAS), and other automated systems. Several studies have focused on improving the performance of emotion recognition approaches. Now, a team of researchers from University of Damascus, Research Center Borstel—Leibniz Lung Center, and Dresden University of Technology developed a robust emotion recognition system. The team used a grid search technique, which is a model hyperparameter optimization technique to fine-tune the parameters of the proposed Convolutional Neural Network (CNN) architecture, to train the CNN model.
To test the proposed robust emotion recognition system, the team used public benchmark datasets: MAHNOB and DEAP. Distributed Evolutionary Algorithms in Python (DEAP) is a multimodal dataset used to analyze human emotional states. MAHNOB is an audiovisual laughter database that contains 22 subjects who were recorded while watching stimulus material, using two microphones, a video camera and a thermal camera. The team collected data from 30 young healthy adults (17 female and 13 males). The new CNN architecture has three convolutional layers, three subsampling layers in between, and an output layer. The proposed CNN architecture was compared with current models such as Support Vector Machine, K-Nearest Neighbor, Naive Bayes, and Random Forest.
The team used only electrodermal activity (EDA) signals to develop a subject-independent human emotion recognition with a promising recognition rate. The team observed an increase in the f-measure (a measure of a test’s accuracy) for subject-independent classification for MAHNOB and DEAP. The team also reported an emotion recognition analysis using only the EDA signal for subject-dependent with an accuracy of 56.5% for the arousal dimension and 50.5% for the valence dimension based on four songs stimuli. In further research, the team plans to focus on human emotion recognition using EDA with respect to different lab–settings and integration of Stacked Sparse Auto Encoders – (artificial neural network used to learn efficient data codings in an unsupervised manner) with CNN. The research was published in the journal MDPI Sensors on April 7, 2019.