A Novel Motion State Recognition Method for Blast Furnace Burden Surface in Ironmaking Process.

Real-time and accurate motion state recognition of blast furnace (BF) burden surface (BS) is significant in the timely monitoring of abnormal furnace conditions and guiding top charging operations. However, unlike the general scene, achieving accurate and effective motion state recognition of the BS is currently unavailable and challenging due to the harsh environment inside the BF. To address this challenge, we propose a novel image-based BS motion state recognition method using the feature-point optical flow clustering in the saliency-driven target region. First, a high-quality BS image acquisition system is devised, including image acquisition using developed equipment and image enhancement using the illumination-guided camera response model (ICRM), and the various BS motion states are displayed through the enhanced images. Next, a target region detection model based on bidirectional prioritized random walks (BPRW) is constructed, and the feature-point optical flow in the target region is extracted. Finally, a maximum local density-based Gaussian mixture model (DGMM) of nonparametric estimation is constructed to recognize the motion state of the BS. Extensive experiments demonstrate that the proposed method can accurately and efficiently identify the different motion states of the BS, which provides furnace conditions data for guiding BF charging operation.