A Research Of Affection Recognition Into RSP Signals Based On ACO Algorithm

Affective Computing means that computers are endowed with human emotion, as well as the ability to understand the state of human emotion and the ability to recognize it. Affection Recognition is an important branch of Affective Computing, which is one basis to establish harmonious human-machine environment. The study of Affection Recognition covers speech sounds, facial expressions, postures, text, physical signals etc, among which physical signals is the most difficult one. In the process of studying physical signals, Respiration Signal (RSP) is one of the main research subjects. The change of RSP signals, to a certain extent, can reflect the most important and real part of human emotion, therefore by looking into RSP signals, we can recognize human inner emotion and its changes, which lays a solid foundation for the establishment of Affection Recognition System.RSP signals are rich in affective state information and can obviously reflect human changes of affective state, so this paper, by extracting features of collected RSP signals and adopting intelligent optimization algorithm to select affective features, has effectively solved many limitations of traditional methods in dealing with feature selection in a certain degree. On this basis, this paper conducted a further research to the feature combination of particular mood in RSP signals.What have done in this paper can be classified into following parts:1. Collecting RSP signals. By collecting respiration signals data of subjects in six affective states (i.e. happiness, anger, surprise, disgust, grief and fear) on collecting device MP150, we established a basic emotion database of respiration signals. Materials are movie clips rich in various kinds of emotion. Subjects are freshmen in Southwest University.2. Pretreating RSP signals and extracting corresponding statistical nature and wavelet feature. Extracting effective RSP signal affective feature is extremely important to the subsequent recognition of affective.state. Before extracting features, the experiment pretreated the RSP signals, such as denoise, etc. Pretreatment is a method that uses Butterworth low pass filter to filter interfering signals and normalizes numerical value according to baseline value. As an effective time-frequency analysis method, Wavelet Transform has good frequency characteristic. This paper adopted Wavelet Transform method, extracted 84 wavelet features from respiration signals and combined the features of respiration signal itself to extract 87 statistical natures, and finally a original collection of 171 features was formed.3. Selecting feature by using Ant Colony Algorithm. Feature selection is actually similar to combinational optimization, so we can solve the problem of Feature selection by solving Combinational Optimization problem. Ant Colony Algorithm is a algorithm that bases on ant colony intelligence global optimal algorithm; has characteristics of fast-computing, easy-coding, good-ant diversity, easy-understand and realize, etc; and it is suitable to solve Combinational Optimization problem. This paper took Respiration Signal as the research subject, led local search and mutation strategy to Ant Colony Algorithm, combined Fisher classifier with Emotion Recognition, and it not only showed a good result, but provided an effective feature subset for recognizing affective state.