Abstract: To detect tobacco insects (referring to Lasioderma serricorne in this study) in cigarette manufacturing factories accurately and efficiently, YOLOV3 model was adopted for real time recognition. To enhance the robustness of the model to recognize the position, cut tobacco and tobacco dust, sufficient data and ideal priori anchor frames provided by data augmentation techniques were adopted. In addition, the k-means++ clustering method was also used to train YOLOV3 model. The detection accuracy and anti-interference ability of the trained YOLOV3 model were analyzed, and the results showed that:1) After training, the YOLOV3 model could detect the tobacco insects in cigarette factories with above 95% detection accuracy for both validation set and test set. The detection speeds were 0.112 1 s and 0.129 2 s per frame respectively, which met the requirements for practical insect detection in a cigarette factory. 2) The trained YOLOV3 model had certain anti-interference ability to impurities, such as sticky position, cut tobacco and tobacco dust, suitable for installation and deployment in complex production environments. 3) The tobacco insect alarm system based on the YOLOV3 model could accurately detect the number of tobacco insects, the increment and position of tobacco insects at a given period of time. This system thus can be used to achieve satisfactory insect control in tobacco reclaimers and sliver separators.