Abstract: In order to promote the efficiency of automatic sorting in a tobacco logistics center, a rapid recognition algorithm based on vision technology was proposed to match the high-speed automatic cigarette carton sorting line. A new lighting means combining a dome light source with a coaxial light source was designed, a high-speed color camera was used to capture the image information of cigarette cartons, and a feature description method based on AGAST (Adaptive and Generic Accelerated Segment Test)corner domain was proposed. A database of cigarette carton images was established on the basis of the extracted features, and extreme learning machine (ELM) was adopted for training and cigarette carton recognition. Comparing with SIFT and SVM algorithms, the proposed algorithm boasted a recognition rate of 100% and shorter recognition time, it met the requirements of 10 frames/s required by the automatic sorting line and also provided a reference for effectively promoting the inspection accuracy in the case of abnormal cigarette cartons.