Knowledge representation and reasoning based on probabilistic graphical models embedded in an agent for object detection

摘要

A new method is proposed for uncertain and dynamic multisource knowledge representation and reasoning based on probabilistic graphical models embedded in an agent cycle for object detection problems. Firstly, factor graphs, a kind of probabilistic graphical model, is extended from the basic uncertainty measure function and nesting definition. Then, it is embedded in an agent cycle with the description of problems and intentions. At last, problem driving knowledge selection methods are proposed. And scalable multilevel precision knowledge reasoning is easy to be realized based on Monte Carlo simulation. Results from object detection in remote sensing images, a knowledge processing problem with massive data, validate our methods.