Research On Fire Detection Based On Multi-sensors Data Fusion Technology

Most traditional detection methods carry on the judgment and the recognition through gathering sole fire characteristic parameter information.As disturbed by many factors the rate of false alarm has been relatively high. The method that, in view of non-structural characteristics, based on intelligent information processing has the abilities of self-learning and self-adaptive and now has become the study of fire detection technology.On the background of the project of Chongqing forest health monitoring system and combining with the Jinyun mountain ecosystem,we proposed two kinds of two-stage fire detection systems based on data fusion technology.They are the two-stage fusion system based on BP neural network and evidence theory and the two-stage fusion system based on evidence theory.The two fusion systems are proposed based on two different fusion ideas:The first one is proposed based on the consideration of making use of the neural network to solve the nonlinear structural problem;and the other is proposed based on the thought of dimensionality reduction and statistics.Before building the two-stage fusion system,we have intensive study the basic theory of neural network and evidence theory. Why we choose to use BP neural network to solve nonlinear structural problems is that BP neural network is the network that is the most mature, owning high training accuracy and satisfactory generalization results and the most widely applied.In the practicle of our application,in order to overcome the existent defects of the traditional BP learning algorithm,such as slow convergence speed, easily falling into local minimum etc,we choose a improved learning algorithm beased on L-M algorithm to achieve the purpose of training neural network after studing a variety of improved learning algorithms. And the simulation results show that the convergence speed of improved learning algorithm has been greatly improved. All along,the research on DS evidence focuses on the following three aspects:①the construction of basic probability assignment function;②fusion rules of high conflict evidence;③the problem of combinatorial explosion in evidence synthesis.In this paper,we focused on the first two points. First we proposed a basic probability assignment function based on an experience database;Aiming at the issues of high-conflict evidence integration,we summarized and analyzed the typical domestic and foreign literature, considering the implement and complexity of algorithm and select an effective improvement of the fusion rule.And these improved methods proved to be effective and practical after applying in the two-stage fusion system.