In the process of preparing fused magnesia in fused magnesium furnace, different working conditions such as smelting condition, feeding condition and semi-fused condition alternately occur. Among them, semi-fused condition is the most difficult and critical to distinguish. At present, the identification of semi-fused conditions mainly depends on manual experience. The accuracy of this method depends on the experience level and physiological state of workers, in addition, the labor intensity of the workers is high and it is easy to miss detection and misdetect. Therefore, based on the dynamic characteristics of the furnace flame image under different working conditions, this paper proposes a working condition recognition technology based on the dynamic characteristics of the B-spline dynamic network. Firstly the linear dynamic system model of the furnace flame is established to describe the dynamic characteristics of the system. Then, the kernel function based on subspace principal angles is designed to measure the similarity of the flame dynamic models. The comparison experiment shows that the design of the working condition recognition technology based on the dynamic characteristics of the B-Spline dynamic network has the better classification accuracy and higher efficiency.