Target identification by adaptive combination of conflicting evidence

The observed information in an air target identification system is always uncertain and conflicting. Dempster's rule can achieve reasonable and specific combination results, but it involves counter-intuitive behaviors when the information high conflicts. Dubois & Prade (DP) rule and belief functions discounted method can palliate this drawback, but their computation is a bit complex and performance of convergence is not good enough. In order to take advantage of the various rules while avoiding their respective drawbacks, an adaptive combination approach is proposed. The most adapted rule is automatically selected among these rules according to the amount of conflict. Both conflicting beliefs and evidence distance are used to measure the conflict from different aspects. If the conflict is low, Dempster's rule can be used. Otherwise, DP rule or belief functions discounted method will be selected depending on the details of conflict. An air target identification experiment shows that the adaptive combination approach can get reasonable combination results with good performance of convergence in case of high conflict, which is beneficial to improving the speed and rate of target identification.