The Study Of Bayesian Decision Model Based On Machine Learning For Human Action Recognition

Accurate identification of human movement changes is beneficial to health monitoring,clinical diagnosis and treatment of related diseases,rehabilitation evaluation and other applications,and its related study has catched much attention in the interdisciplinary fields of statistics,data mining and clinical medicine.How to effectively improve the accuracy of human action recognition in multi-category wearable Body Area Networks is a hot issue in recent years,and the difficulty lies in how to effectively explore the characteristics of the differential distribution of human action in multiple categories.So this paper takes the precise expression of multi-category human action difference separation feature space distribution as the entry point,and tries to integrate machine learning with Bayesian model,aiming to make full use of the excellent statistical learning characteristics of machine learning algorithms(such as Support Vector Machines,Random Forests,K-Nearest Neighbors,etc.)to accurately capture the multi-category human action difference probability distribution space and improve its related a priori accuracy,so as to improve the accuracy of Bayesian decision model accuracy.The international public wearable human action WARD database(13 categories in total)is selected to evaluate the presented post model.The related work is as follows:1.Propose a new method for extracting differential features of human action in multicategory body domain networks based on Random Forest algorithm,it use the excellent "voting mechanism" integrated learning characteristics of Random Forest algorithm to obtain differential separation features closely related to the inherent coordination,stability and other attributes of multi-category human action patterns,and effectively characterize the spatial distribution of the separation of multi-category human action patterns and facilitate the accuracy of Bayesian decision models.Results show that the post method can effectively characterize the spatial distribution of differential separation features of multicategory human action patterns,and the accuracy of the Bayesian decision model can reach90.64% in recognizing 13 categories of human actions,which give a supports for improving the accuracy of the multi-category human action Bayesian decision model.2.A new model for multi-category human action based on the fusion of Support Vector Machine and K-nearest neighbor algorithm(SVM-KNN)is proposed,which makes full use of the excellent statistical learning characteristics of SVM and KNN algorithm to obtain significant attribute features,effectively eliminates the misclassified samples mixed with the above Random Forest algorithm to characterize the separation feature space distribution of multi-category human action pattern discrepancy,and enhances its boundaries.The post model can improve the accuracy of the posterior distribution of the Bayesian decision model by weighting the boundary samples to optimize the posterior distribution.The experimental results show that the accuracy of the proposed model for recognizing 13 categories of human action patterns can reach 94.27%,compared with random forest,its accuracy increase 3.63%,effectively improving the accuracy of the Bayesian decision model and providing a new idea and method for accurately recognizing multi-category human actions.