Research And Design Of Online Identification System For Tobacco Stem

As one of the main materials of cigarette,tobacco stem can be processed into cut stem or stem grain with high filling value and great combustibility,which plays an important role in reducing harm and lowering tar,reducing formulation cost and improving physical quality of cigarette.The long-stemmed rates of tobacco stems,tobacco stems with leaves rates and thick tobacco stems rates and so on,which has obvious influence on the structure of tobacco stem,the rates of cut tobacco stem and the proprrtion of pure tobacco stem after cutting.thus affecting the physical quality of rolled and finished cigarettes.In order to accurately classify and identify tobacco stems in the tobacco stem production line,feedback the process control parameters of threshing and removing tobacco stems,combined with the popular deep learning technology,this paper carried out the research and design of tobacco stem online identification system,the following work has been done.(1)Firstly,an image collection platform for online recognition and classification of tobacco stem had been built.The platform was used to collect the original data of tobacco stem image for making image data set or providing image collection function for online identification and classification.(2)The traditional image algorithm was used to analyze the morphological characteristics of the tobacco stem,and the feature extractor and classifier were designed to classify different tobacco stem.To explore different methods of tobacco stem image identification and classification.The measurement method of tobacco stem parameters was designed by using image morphology method to obtain the long stem rate,thick stem rate and other quality indicators of the whole tobacco stem.(3)The image collected on the tobacco stem image acquisition platform was preprocessed to make the tobacco stem data set,and the data set was optimized by using data enhancement technology.(4)Different deep learning models were trained by tobacco stem data set,the trained deep learning model was used to identify and classify tobacco stem images,and the accuracy of tobacco stem is tested in online scene.The improved deep learning network model was carried out to improve the performance such as the accuracy and recall of the target tobacco stem identification and classification.The experimental results show that: the identification rate of the traditional image processing method for pure tobacco stem can reach 80.4%,but the identification rate for the tobacco stem with leaves is not well,only 62.1%.In this paper,we used several methods based on deep learning to identify three kinds of tobacco stem,the identification rate is higher than the traditional image processing method,and the accuracy more than 90%.Specially,the average identification rate of the improved yolov3 deep learning network model is more than 95%,which is nearly 6% higher than that of the unmodified model,which proves the feasibility of the deep learning method in the identification and classification of tobacco stems.This method can provide support for improving the efficiency and accuracy of tobacco stem classification.