Evidence Fusion Method Based On Correlation Measurement Coefficient And Its Application

Dempster-Shafer(DS)evidence theory can be used to express uncertain information.Without considering prior probability,uncertain information can still be modeled to help people realize multi-source information fusion and decision-making.However,since the combination rules of evidence theory do not consider the conflict of compatible focal elements among evidences,fusing evidence of conflicting information will appear counterintuitive results,resulting in reduced decision accuracy.In addition,how to construct the basic probability assignment function is also a difficulty in evidence theory.Therefore,the correlation measurement coefficient is proposed to measure the degree of similarity between evidence and the degree of correlation between monad sets and multiple subsets,and a new evidence fusion method is proposed based on it.In addition,this method is also applied to the identification of impulse noise and significance detection,so as to improve the accuracy of detection and recognition.The main work is as follows:(1)A new Evidence fusion method based on Evidence Correlation Measure(ECM method)is proposed.Firstly,the definition of correlation measure coefficient is given,which is used to measure the similarity degree between different evidence,and express the difference between monad set and multi-subset evidence.At the same time,it also has the mathematical properties of non-negative,symmetrical and non-degenerate.Secondly,according to the correlation measurement coefficient,the credibility weight of evidence is constructed,and the information volume of evidence is combined with the weight correction of evidence.Finally,using Dempster’s combination rule to combine new evidence,effectively reducing the probability of counterintuitive evidence conflicts and improving the accuracy of evidence fusion.(2)Aiming at the problem of uncertainty mutation caused by impulse noise impact,a multi-feature criterion fusion pulsed noise target recognition method is proposed.By analyzing the difference between the pulse noise and the surrounding signal points,this paper concludes that the pulse noise has extreme characteristics,dissimilar characteristics and discontinuity characteristics.Based on these three characteristics,the mass function is established to quantify the possibility of the original pixel changing into the pulse pixel.At the same time,the ECM method proposed in this paper is used to recognize pulse noise.The experimental results show that this method can still maintain the high accuracy of pulse noise recognition and provide a better basis for image restoration when the image is impacted by high-density noise.(3)In saliency detection,this paper presents a significance detection method based on superpixel multi-feature fusion.Firstly,the convex hull is constructed by using corner detection to form the rough foreground or background.Secondly,extracting the texture features,color features and position features of the superpixel,and the mass functions is constructed.Finally,the multi-feature fusion is realized by combining the ECM method,and the superpixel under test is accurately classified as the foreground(background),so that the significant area can be displayed more clearly from the image.The simulation results show that the proposed method can preserve more precise target boundary,and its stability is better than other comparison algorithms.