A Collision and Reaction Model of Feature Fusion: Mechanism and Realization
Weimin Peng, Huifang Deng · IEEE Intelligent Systems · 2015
Duplicate data must be properly handled in the process of intelligent search, surveillance, and tracking. To better detect and fuse this data, the authors take a new perspective on feature fusion based on the "collision and reaction"' mechanism of data objects. This mechanism can greatly facilitate differentiating duplicate and nonduplicate data in the fusion (reaction) probability. With this perspective, the fusion probability of a feature sample in a dataset is the cumulative effect of the microscopic collision (interaction) results between this sample and other samples. The operations on duplicate detection and feature fusion depend on the fusion probability and are finally realized when the collision process is finished. This new quantum-inspired feature fusion model gives interesting experimental results and could have great implications for some intelligent systems.