This paper focuses on how to perspective the melting behavior of solid iron tailings in molten blast furnace slag and take a new non-contact visual analytical method to predict its melting law. The optimized convolution neural network (CNN) is used to track the moving target in charge coupled device (CCD) camera system efficiently and accurately, and the melting behavior of SiO 2 is described by coordinate translation transformation theory. Hierarchical agglomerative clustering (HAC) and delaunay triangulation were used to extract the characteristic parameters of the melting process of SiO 2. The prediction model of the melting rate of SiO 2 at high temperature was established by least square fitting (LSF) and dimensional analysis, and compared with the actual melting rate of SiO 2 obtained by experiments. The results show that the melting characteristics of SiO 2 at high temperature are in accord with certain function rule. The performance of optimized CNN in terms of processing time and accuracy is significantly improved, and the fusion rate prediction model of SiO 2 is verified by 100% accuracy. It provides theoretical support and model basis for the improvement of slag cotton preparation technology. INDEX TERMS Melting rate, target tracking, feature extraction, dimensional analysis, best match.