ABSTRACT
This paper aims to establish a more defined diagnostic model based on image processing technique that can recognize more statuses rather than unsteady status. Definition and extraction of feature quantities are important preparatory work in this task. On the one hand, more well-defined quantities contribute to better recognition of combustion status; on the other hand, they increase the consumption of CPU. It is an advisable choice to find a balance between accuracy and consumption of resource, under precondition of meeting the production requirement. Base on the idea above, we use rough set as a feature selector to determine the optical feature combination. Support vector machine is chosen as classifier because of its performance of high accuracy and generalization capability in classification task.
