Abstract: The existing tobacco sorting system failed to reject unqualified stripes,stems,and foreign matters in a single system,therefore an on-line fine sorting system was designed and developed,which was composed of 4 subsystems responsible for rejecting agglomerated strips,stem,greenish or mildewed strips and foreign matter respectively.The agglomerated strips subsystem adopted a rejecting device of star-shaped rollers,and the other three adopted machine vision technologies separately incorporating different imaging means,such as low energy X-ray,visible light,etc.,to detect undesired matters by intelligent image algorithm with machine learning function and reject those matters by high pressure air jets.The results of experiments showed that:1) The rejecting rate of the system for agglomerated strips was over 95% at different tobacco throughput;2) Taking processing capacity and sorting efficiency into account,the preferable processing capacity was 1000 kg/hour,at which its rejecting rates for stem,greenish or mildewed strips and foreign matters were 91%,75% and 92%,respectively,which were higher than the levels of hand picking.The new system well satisfied the requirements of normal production,it improved the efficiency of strips sorting,and reduced labor cost.