Research On Target Recognition Technology Based On Multi Sensor Feature Information Fusion

In the modern battlefield,since the target recognition technology of a single sensor has been far from satisfying the current limitations of the needs in all areas.While the feature based on multi-sensor multi-target recognition technology utilization characteristics of each sensor which can be accomplished more multi-task,Recognition based on feature fusion technology can be used in many uncertain environment and complex background,characterized by a plurality of sensors for effective information integration,not only can improve recognition rate,but also can reduce the limited data bandwidth.Characterized by a plurality of sensors for effective information integration not only can improve the recognition rate,and can reduce the data bandwidth limited.At present,research on feature-based multi-sensor information fusion target recognition technology is relatively small,it is very urgent to wait for its depth and effective theoretical and technological research.To this end,we went to explore and study a few content from the following:1)Introduces several feature extraction algorithms,data obtained by the external sensors,use the feature extraction algorithm to extract data effectively,which mainly for hu moments,affine invariant moments,wavelet moment and Zernike moment four main feature extraction algorithm for a detailed analysis and simulation.2)The main study of the covariance matrix algorithm and based on the covariance matrix of its characteristics to studied the covariance matrix of image feature fusion method.To carry out the integration of the characteristic information of the image area by constructing efficient and reliable covariance matrix,so that the feature information fusion has a better rotation invariance,scale invariance,robustness,class separability and other characteristics.3)Mainly analysis and research the BP neural network of its slow convergence and easy to fall into local minimum problem,First,we proposeed an BP neural made improvements through the use of additional momentum to adjust its weights and add learning rate adaptive learning algorithm selection,Mainly BP neural network analysis and research,slow and easy to fall into local minima problems for its convergence speed.First,BP neural made improvements through the use of additional momentum to adjust its weights and add learning rate adaptive learning algorithm selection,and then through theoretical analysis proposed and studied using particle swarm algorithm to optimize the parameters of the improved BP neural network to further improve the learning speed of neural networks,and finally using the above feature information fusion for target recognition.The results showed that: compared to conventional neural network,particle swarm optimization to improve BP neural network to some extent,the formula can significantly improve object recognition rate,while the practical application of this algorithm to the identification system,the algorithm proved feasible and effectiveness.