Considering the present situation of the basic oxygen furnace (BOF) steelmaking endpoint control, a neural network model was established to judge the steelmaking endpoint. Based on the furnace mouth radiation information acquisition platform, the spectrum and image characteristics were analyzed using the fiber spectrum division multiplexing technology and the color space conversion method. The results indicate that they are similar at early-middle stage but inverse at the steelmaking late stage. Some appropriate variables were selected from the law curve as the neural network model parameters and the model was trained and forecasted on the basis of an improved back propagation (BP) neural network correction coefficient algorithm. The experimental results show the response time is less than 2 s which meets the requirements of online endpoint judgment, and the prediction accuracy of the proposed algorithm is better than that of the conventional algorithm. The system works stably and the anticipated effect is achieved.