Deep Learning-Based Method for Coal Ash Fusibility Determination

In order to solve the problem of limited feature expression ability of existing coal ash fusibility determination methods leading to low determination accuracy, this paper proposes a coal ash fusibility determination method based on deep learning. To meet the data requirements for model training, we use frame segmentation technology to generate temporal image sequences from coal ash fusibility videos. We then extract images of the ash cone at various temperature characteristics and label them with corresponding temperature tags, thereby constructing a comprehensive coal ash fusion image database. We subsequently propose a multi-module collaborative deep learning model comprising two key components: 1) The residual-attention module mitigates the gradient vanishing problem using residual connections while also leveraging channel attention mechanisms to dynamically adjust feature weights; 2) The feature pyramid module effectively amalgamates high-level semantic information with low-level detailed features via multi-level feature fusion. Experimental results reveal that our proposed method outperforms existing models in coal ash fusibility determination, offering intelligent and precise coal ash fusibility measurement.