Key Technology Based On Machine Vision Tobacco Impurity

The production of modern cigarettes roughly goes through seven major processes, including tobacco curing, roasting, fermenting and cutting, and after that the cigarette products are put into the market and finally provided with consumers. Currently, the impurity control in the procedures of tobacco curing and roasting is mainly finished manually, which would cost a lot of time and finally affect the efficiency of tobacco cutting. So it’s of great industrial significance to automate the removal of tobacco impurities.In this paper, according to the features of near-infrared and visible images of tobacco leaves, and combined with theoretical knowledge of digital image processing and pattern recognition, automatic detection of tobacco stem and tobacco leaf classification together with other algorithms are implemented. And an impurity removal system of tobacco leaves based on machine vision is built. The major work of this paper includes:Firstly, we verified the feasibility of detecting tobacco stem by near-infrared images. Based on fuzzy enhancement and transition region (threshold selection), we gave a method for tobacco infarction stem segmentation. After segmentation, we obtained near-infrared tobacco infarction stem which was analyzed through chain code tracking and skeleton thinning respectively. And both methods reached ideal results.Secondly, we verified the validity of Gist feature in tobacco leaf classification. Then combined with the features of normal tobacco and moldy tobacco images, we improved the extraction method of Gist feature and added the rotation invariance in the direction graph of Gist feature, and got color feature map by linear functions. Finally, we used support vector machine (SVM) to implement tobacco leaf classification and the recognition rate met industrial requirements.The last but not the least, we designed a tobacco impurity removal system based on machine vision. Combined with the requirements of cigarette industrial production line, we gave the complete system structure and function modules. Furthermore, we introduced the designs of the critical hardware and software of the system in depth, and proved the effectiveness of the proposed algorithms in industrial production.