An improved semisupervised learning for top gas flow state classification to reduce emission and improve production efficiency in ironmaking blast furnaces

Developing a deep learning classifier to identify gas flow states in blast furnaces using top-camera images typically requires a substantial number of labeled images. However, manual labeling is both time-consuming and costly, making it impractical for large datasets. To address this challenge, this study proposes an improved semisupervised learning model based on the Mean Teacher framework. The model integrates a novel feature loss module to achieve high classification accuracy with minimal labeled data. In the datasets with 20%, 30%, and 40% labeled ratios, using a single training iteration with labeled data, the model achieves the accuracies of 78.61%, 82.21%, and 85.2%, respectively, while the accuracies are 82.09%, 81.97%, and 81.59%, respectively, for the multiple cycle training iterations. The results show that the proposed method can accurately identify the gas flow state at the top of a blast furnace, outperforming both the original Mean Teacher model and a fully supervised approach. Additionally, tailored training schemes are provided for various application scenarios. The findings show that the proposed technique can significantly reduce the reliance on labeled images, promoting the application of intelligent models for blast furnace operation.