Assessing sensor reliability for multisensor data fusion within the transferable belief model

Area of Science:

• Artificial Intelligence
• Machine Learning
• Data Science

Background:

• Sensor reliability is crucial for accurate classification tasks.
• The transferable belief model (TBM) offers a framework for reasoning under uncertainty.
• Existing methods may not fully capture sensor reliability in joint operational contexts.

Purpose of the Study:

• To develop a novel method for assessing individual sensor reliability.
• To extend this method for evaluating the reliability of multiple, jointly operating sensors.
• To enhance classification accuracy through improved sensor reliability assessment.

Main Methods:

• Developed a method to find a discounting factor for individual sensor reliability based on minimizing the distance between pignistic probabilities and actual data.
• Extended the method to assess joint sensor reliability by computing discounting factors that minimize distance for aggregated belief functions.
• Utilized the transferable belief model for belief function manipulation and pignistic probability computation.

Main Results:

• Successfully quantified individual sensor reliability by identifying optimal discounting factors.
• Demonstrated a method to assess the aggregated reliability of multiple sensors working in concert.
• The proposed methods provide a quantitative measure of sensor performance in classification.

Conclusions:

• The developed methods effectively assess sensor reliability within the transferable belief model framework.
• The approach is applicable to both single and multiple sensor systems in classification.
• This work contributes to more robust and reliable machine learning systems through better sensor evaluation.