Industrial image classification using a randomized neural-net ensemble and feedback mechanism

Accurate burning state recognition plays an important role in rotary kiln sintering process control. In order to avoid significant discrepancy between human feedback cognitive mechanisms and traditional open-loop recognition methods, a novel intelligent cognitive model based on a variable granularity simulated feedback mechanism is explored in this paper, where evaluation indexes of cognitive results are established to freely regulate cognitive granularity, and the variable granularity simulated feedback mechanism is constructed to update cognitive features and cognitive rules with different granularities. The proposed cognitive model is applied to improve burning state recognition accuracy. With the initial granularity, a burning state recognition decision information system is developed using extracted flame image features. Random vector functional-linker (RVFL) network ensembles are employed to build the initial burning state recognition rules. By using cognitive errors and granularity transformation rules, a heuristic feedback mechanism is proposed to update the decision information system and recognition rules. The experimental results show that our method is effective and outperforms other open-loop recognition techniques.