Research On Multi-sensor Information Fusion Target Recognition Algorithms

In recent years, the performance of the sensor is improved a lot with the development of information science and sensor technology. A large number of different kinds of multi-sensor systems are widely used in the complex background, which makes the research of multi-sensor information fusion attract broad attention by more and more people. Multi-sensor information fusion is also called information fusion, and is to produce meaningful information by dealing with data from multiple sensors in different levels and grades. Multi-sensor information fusion has a great application prospect not only in the military field but also in the civil field. In the military field, the purpose of the information fusion is positioning, characterization and identification of the observation, including multi-sensor detection, multi-sensor integrated tracking and state estimation, multi-sensor target recognition fusion, situation description, threat estimation, sensor and database management, etc. Multi-sensor target recognition includes fusing the data in the detection and investigation information to get the target recognition results and credibility, which will be studied. Evidence theory is a kind of uncertain inference methods, and has a wide range of applications in detection and diagnosis and artificial intelligence, etc. Especially in the multi-sensor information fusion field, evidence theory has become a basic and important fusion algorithm. Target recognition is studied by using the theory.The study includes:(1) The paper studies the multi-sensor information fusion theory and introduces common basic target recognition algorithms.(2) D-S evidence theory is helpful to deal with uncertain information, but the classical D-S algorithm also has one ticket veto and evidence conflict problems, which makes the recognition accuracy lower. Therefore this paper proposes two improved algorithm of evidence theory to deal with the conflict. They are combination method of conflict evidences based on each evidence value and D-S evidence theory based on the improved evidence conflict coefficient. And the paper shows simulation and analysis about the two improved algorithms.(3)The basic probability assignment of the improved target recognition algorithm is still difficult to obtain. For this, an improved adaptive genetic algorithm is used to optimize BP neural network to obtain the basic probability assignment. And the target recognition algorithm based on sensor confidence is put forward to improve the target recognition algorithm further. The simulation shows that the two methods are both effective.