With the development of the steel industry, there are higher requirements for casting technology, and the addition of melting agents and crystallizers at suitable key nodes helps to improve the purity and quality of steel. Therefore, the study of mold melting and crystallization is a crucial part of the entire casting process, and the casting temperature regulation that relies on the naked eye and experience to identify key nodes in the past can no longer be satisfied. In this paper, an image information extraction technique is proposed to quantify the casting process and study the influence of temperature change on melting rate and crystallization rate based on time series model. Based on the image extraction technology, three eigenvalues of image change are extracted: mean gray value, differential statistical mean of image grayscale, and statistical entropy value of gray difference, and divide the whole casting process into two parts: melting and crystallization, and determine the start and end of melting and the beginning and end of crystallization through the change of three eigenvalues. After research, the change of the values of the three indicators is counted, and it is found that the steel completely enters the liquid state at 1 49 seconds, enters cooling in 142 seconds, and is completely crystallized in 671 seconds, and by studying the linear relationship between the above indicators and melting and crystallization, The relationship between temperature and melting rate was determined, i.e. the melting rate peaked at 980 °C, and the melting rate reached 90% as the temperature rose continuously, and the flux was added at this time; With the decrease of temperature, the crystallization rate gradually increases, reaching a threshold around 800°C, and the crystallization rate rises to 83 %, at which time the crystallizing agent is added. Adding fluxes and crystallizers at the right time allows the purity and quality of the mold to be at its best.