During the tobacco threshing and redrying processes, separation quality of tobacco leaves and stems has a significant effect on the purity of output tobacco lamina. For the traditional separation method by air classifier, separation efficiency is limited by the finite difference in suspension velocities of stem tobacco leaves. Developing an accurate detection method for stems is of key importance for improvement of tobacco lamina quality by upgrading separation efficiency from stems and leaves. In the present work, an on-line detection and separation system of stem from tobacco lamina was developed using low-energy X-ray transmission imaging and digital imaging analysis technology. Optimal testing conditions were determined as an X-ray radiation intensity of 70KeV, a detection belt speed of 1.6 m/s, as well as the gray segmentation threshold of 40,000 used in the image recognition algorithm. The free stems and stems with leaves were respectively used as target samples to test the validation of the detection method. Test results showed that both the recognition rates for free stems and stems with leaves were higher than 98%. When stems with leaves were mixed with pure leaves and then detected, the recognition rate was still as high as 94.5%, and the separation rate of stems was 92%. The experiments indicated the feasibility of the on-line detection method of separating stems from tobacco lamina based low-energy X-ray transmission imaging, which has potential applications for the accurate separation of stems and leaves in the tobacco threshing and redrying processes.