Research And Application Of Multi-sensor Data Fusion Algorithm

With science and technology advancing, various sensor technologies emerge endlessly. For the practical needs of the growing, sensor systems get to complex. Traditional systems are simple, usually obtaining information from single sensor. In spite of multi-sensor it reflects the target information from multiple sides individually. Many modern systems are complex and deal with the information from a plurality of signal sources. So it needs use multi-sensor observation information in combination by the optimization criteria to make up one-sidedness of single sensor. And, multi-senor’s complementary and denoising can improve the accuracy, fault tolerance and comprehensiveness. Thus, multi-sensor data fusion technology is necessary.Multi-sensor data fusion is a multi-source data processing technology, and a multi-level, multi-faceted process. This article first reviews the basic theory of data fusion and introduces the classification and model. Then, fusion algorithm of theoretical study and experimental demonstration are been done from two aspects, which are data level fusion and decision level fusion. On data-level fusion, the article studies the adaptive weighted fusion algorithm principle, noise and accuracy problems, and carries out the experiment. On decision level fusion, the article studies the basic principles of BP neural network, network design analysis as well as the problems existed, and carries on the glass quality prediction experiment. Meanwhile, the article also studies the basic principles, combination rules and problems of D-S evidence theory, which is done experiment on the glass quality prediction.In the data-level fusion research, the article proposes stepped dynamic weighted fusion algorithm according to the feature of fusion accuracy related to fusion and sensor numbers. Through experiment contrast it has higher accuracy. In the decision level fusion research, it uses BP neural network to solve the credibility allocation problem and D-S evidence theory to make decisions and carries on the quality prediction experiment. The result shows it can improve the ability of predicting the glass quality. Finally, the improved algorithms are applied to the glass production process visualization system, and conduct a preliminary analysis and design.