Graph-Structure-Based Multigranular Belief Fusion for Human Activity Recognition.

Area of Science:

• Artificial Intelligence
• Information Fusion
• Uncertainty Quantification

Background:

• Belief functions (BFs) are used for modeling uncertainty but face computational challenges with many focal elements.
• Existing methods to reduce complexity include simplifying basic belief assignments (BBAs) or using simpler combination rules, potentially losing specificity.

Purpose of the Study:

• To propose a novel and efficient multigranular belief fusion (MGBF) method to address the high computational complexity of BFs.
• To reduce the number of focal elements in BFs through a graph-based granulation technique.

Main Methods:

• A new BBA granulation method inspired by community clustering in graph networks is proposed.
• Focal elements are treated as graph nodes; distances identify community relationships.
• Nodes within decision-making communities are selected for efficient combination of multigranular evidence.

Main Results:

• The proposed graph-based MGBF method was applied to fuse outputs from CNN + Attention models in human activity recognition (HAR).
• Experimental results on real datasets demonstrate the effectiveness and feasibility of the MGBF strategy.
• The new approach shows potential compared to classical BF fusion methods.

Conclusions:

• The developed graph-based MGBF offers an efficient solution for complex information fusion tasks involving belief functions.
• This method effectively reduces computational complexity while maintaining the integrity of the fusion process.
• The approach is validated through successful application in HAR, highlighting its practical utility.