Learning fuzzy modelling through genetic algorithm for objectrecognition

摘要

The paper proposes a genetic-algorithm-based learning strategythat models membership functions of the fuzzy attributes of surfaces ina model based machine vision system. The objective function aims atenhancing recognition performance in terms of maximizing the degree ofdiscrimination among classes. As a result, the accuracy of recognizingknown instances of objects and generalisation capability by recognizingunknown instances of known objects are greatly improved. The performanceenhancement of a model based object recognition system consisting of aset of synthetic range images is established by incorporating a dynamicoff-line learning mechanism using a genetic algorithm in the feedbackpath of the system