A Study Of Affective States Recognition Based On Ant Colony Optimization From Physiological Signals

With the rapid development of computer technology, the wish for using computers for human services is stronger and stronger, emotion computing aims at enduing computer with the ability of recognition, comprehension, expression and adapting human emotions, and enabling it to perceive emotions, thus the computer can make a timely response and harmonious emotional communication with the users. Emotion recognition is an important part of emotion computing. Generally speaking, data acquisition and preprocessing is affected by objects and equipment, and the methods of feature extraction and classifier design are mature relatively, however, feature selection is worthy of considering. As an swarm intelligence-based global optimization algorithm, Ant Colony Optimization (ACO) is very suitable for solving combinatorial optimization problems. ACO has features such as simple code, fast calculating speed, good population diversity, easy to understand and implement. In order to obtain a higher emotion recognition ratio and an effective feature fusion, we adopt ACO with Fisher classifier and K-Nearest Neighbor (KNN) introducing increment K.The following works have been done:In order to extract features, we locate the position of QRS complex through the sum of amplitude absolute values of R-wave and S-wave, together with the duration between R-waves, the ratio reach 99.5%.Given the problem of combination optimization of features in the recognition of physiological signals, the thought of computational intelligence was introduced to the feature selection of emotional physiological signals, and the identification results of two groups of data set, using Ant System (AS) and Colony System (ACS) respectively, were compared. Local search strategy and variation were adopted to ACS for feature selection, and KNN was used to recognize happiness and sadness, in the hope of improve the recognition rate of emotions and the convergence rate of algorithm.The ACS-specific pseudo-random proportional rule, combined with local search and variation strategies, was introduced into MMAS (MAX-MIN Ant System) for feature selection, the improvement of convergence rate and the acquisition of high recognition rate and effective feature combination.Our lab collected EMG signals under 7 emotion states, including calmness happiness, surprise, disgust, grief, anger and fear. Through feature extraction, we use MMAS with Fisher classifier to select features, and do one-vs-rest classification simulations, thus obtained some useful feature combination.