Abstract
Nowadays, biological recognition technologies attract more attention in electrocardiograph (ECG) signals, which vary among different people and are difficult to counterfeit. However, the robustness of recognition cannot be well sustained in the case of diversified application scenarios and huge human crowds. In order to tackle this problem, this paper puts forward a fiducial and non-fiducial mixed feature extraction method, which can effectively complete the multidimensional feature modeling of ECG signal. In addition, this paper proposes a linear discriminant analysis (LDA) based on multiple features (LOMF) algorithm based on ECG mixed feature to solve time-overhead problem of big data training. LOMF includes ECG signal preprocessing, sub-block division, and block training. By combining the MapReduce distributed computing framework and the secondary retrieval method based on the multidimensional feature space, LOMF is parallelized to improve recognition rate and computing efficiency at the same time. The experiment results show that, in the diversified scenarios, utilizing ECG mixed feature can return a higher recognition rate than the traditional ECG 1-D feature. Moreover, compared with the traditional LDA and support vector machine algorithms, the precision of LOMF increases by 7%-8%, which depends on the most competitive advantage of using LOMF. LOMF fits MapReduce parallel framework well so it is more effective than traditional algorithms, especially on diversified application scenarios (such as Internet) where the amount of data grows rapidly.