The work aims to propose a new adaptive character detection system to solve the problems of high error rate and poor adaptive ability of existing character detection system. In view of the influence of illumination change and variety of product specifications on the quality of image acquisition in the process of product development and production lot number detection system of a large instant noodle manufacturer in China, the corresponding adaptive solutions were proposed respectively. By improving the local threshold algorithm, the robustness of the algorithm was enhanced, and the adaptive acquisition and correction algorithm based on real-time image acquisition was further proposed. Through the experimental test and field verification, the detection system had strong adaptability and robustness, the detection speed was 10 pcs/s, the error rejection rate was less than 0.005%, and the missing rejection rate was 0. The system is more stable and reliable than the traditional character detection system, and has a good application prospect.