Application Research Of Uncertain Information Processing In Image Target Recognition

Due to various subjective factors and objective factors,there is often some uncertainty in the information in the real world.As an uncertain information fusion theory,evidence theory can better represent,process and combine multi-source uncertain information.Applying evidence theory to image target recognition can better process some uncertain information in the image and achieve better results in target recognition and target tracking.This paper will study the application of generalized evidence theory in image target recognition.In the theoretical research of evidence theory,this paper proposes a generalized basic probability assignment(GBPA)generation method.The training sample is modeled by the Gaussian membership function,and then the single eigenvalue of the test sample is extended to the Gaussian membership function by the maximum error of the sensor.The GBPA is generated by calculating the intersection area between the test model and the proposition model.Finally,the GBPA under multiple attributes is fused by the weighted average method and the decision is made according to the fused GBPA.Experiments show the effectiveness of the method.In the open world,aiming at the problem of image unknown target recognition,this paper applies generalized evidence theory to multi-spectral image classification,which can simultaneously realize the recognition of known and unknown targets.Firstly,the spectral information is modeled and represented by triangular ambiguity number.For each pixel to be identified,the generalized basic probability assignment of the target category is generated in each band.Then,the GBPA generated by multiple bands is combined and made decision.The known and unknown targets in the image can be recognized more accurately,and the classification of the known and unknown targets in the image can be realized.Aiming at the problem of dynamic unknown target recognition and tracking in surface-space dynamic background,this paper realizes the recognition and tracking of known target and unknown target in surface-space dynamic background by combining target inter-frame displacement and block matching.The main contribution of this paper is to propose an open world GBPA generation method,and introduce it into image recognition of known and unknown targets,and establish a more complete implementation scheme and process.For video image sequences,this paper combines block matching and other methods to achieve current target tracking and unknown target recognition and tracking,and achieves good results.