A Novel Evidence Combination Method Based on Improved Pignistic Probability.

Area of Science:

• Artificial Intelligence
• Information Fusion
• Pattern Recognition

Background:

• Evidence theory is crucial for uncertain information fusion.
• Conflicting evidence fusion in single target recognition is a significant challenge.
• Existing methods struggle with computational complexity and information loss.

Purpose of the Study:

• To propose a novel evidence combination method for single target recognition.
• To address the challenge of fusing conflicting evidence effectively.
• To improve accuracy and reduce computational burden in evidence fusion.

Main Methods:

• Developed an improved pignistic probability function for evidence redistribution.
• Utilized Manhattan distance and evidence angle for certainty and mutual support extraction.
• Employed entropy for uncertainty calculation and weighted averaging for evidence correction.
• Applied Dempster's combination rule for final evidence fusion.

Main Results:

• The proposed method demonstrated improved convergence compared to existing techniques.
• Achieved an average accuracy improvement of 0.51% and 2.43% over benchmark methods.
• Effectively handled highly conflicting evidence in single- and multi-subset propositions.

Conclusions:

• The novel evidence combination method offers a robust solution for conflicting evidence fusion.
• The approach enhances accuracy and efficiency in single target recognition.
• This work contributes to advancing the field of information fusion under uncertainty.