The Application Of Data Fusion Technique Based On BP Neural Network To Distributed Intelligent Pension System

With an ageing society gradually approaches, the pension issue of elderly has become increasingly important. Families don’t know whether their older people are well taken care of because of the nursing due to poor facilities in the nursing homes, lack of standardization and combination of only few information management systems. The distributed intelligent pensions system is adopted to solve the problem meeting real time care need which is unattainable using scattered and isolated information management systems in present nursing homes and old-age service facilities. With this distributed system officers can not only monitor physical conditions in real-time, but can also provide information to families about the health of the elderly through the combined use of RFID technology and video Coordinated Supervisory Control monitoring.People in the nursing houses wear RFID wristbands to gather signs information in real time. There provides huge data to the system due to RFID wristbands being reading continuously. Simultaneously the video surveillance system collect the images of the elderly in real-time. In order to know whether the elderly are well taken of, the system needs to deal with these data. Data fusion is a powerful tool to solve the problem of gets the optimal decision from multiple attribute data. There are different kinds of algorithm of data fusion home and abroad. The neural networks have advantage at fault tolerance and resilience, which means have lesser requirements of the system’s prior probability to deal with incomplete or inaccurate information. Combined with multi-source and mass characteristics of data in the distributed system and data processing real-time accuracy characteristics, also according to basic use which has characteristics of high recognition accuracy to some extent, BP neural network algorithm is chosen in this article.However, taking large amounts of data directly as input to the neural network before data fusion makes the neural network’s input dimension become too high and structurally complicated requiring a longer training time. Redundant properties and noises have bad influence on training results.Rough Set has a strong advantage at deal with incomplete, uncertain and redundant data which means not only can it effectively get the best feature with less operation and high precision but also remove the inaccuracy made by personal subjectivity. Integrating advantages of both by combing rough set with neural networks. First, using rough set for data reduction and let the reduction of data as input to the neural network. The whole process reduces network complexity, thereby reducing training time network while making the whole system a certain degree of fault tolerance and anti-jamming capabilities. The attitude of the human body features reduced to k represents after deal with images information based on foreground image information processing. First the data needs to be processed before data processing.The traditional BP algorithm based on the gradient of the steepest descent method which makes the network easily trapped into local minima and slowly convergence of learning process. In this paper select additional momentum algorithm is chosen to improve the traditional BP algorithm. The advanced algorithm increases the amount of amendments to the weights and thresholds in the same gradient direction to ensure algorithm went in convergence direction.