Abstract:
Physiological signals play an important role in the detection of individual pressure, fatigue detection and other studies. Physiological signals contain abundant physiological or psychological state information of human body. Compared with other methods to detect the state of human body, physiological signals have higher objectivity and robustness and can be measured continuously. However, multi-source physiological signal detection also brings a lot of difficulties, such as a variety of signal acquisition, high feature dimension, and low computational efficiency. Aiming to the above problems, this paper presents a hybrid feature selection algorithm by combining ReliefF algorithm, the maximum relevance minimum redundancy (MRMR) algorithm, and the multi-population genetic algorithm (MPGA). Firstly, ReliefF algorithm is used to calculate the correlation weight value between each feature and label and select the feature subset with significant effect on the classification result. And then, the MRMR algorithm is employed to remove the redundancy in the feature subset so as to further streamline the feature subset. Besides, the coevolution algorithm based on MPGA is adopted to select the most effective feature subset. Finally, the proposed algorithm is verified and compared with other feature selection algorithms via the driving stress detection. The results show that the proposed algorithm can effectively eliminate redundant information embedded in high dimensional features, improve the computing efficiency of feature selection process, and achieve better recognition accuracy.