Tobacco Feeding Uniformity Testing System Based On Digital Image Processing

Detection methods, detection devices and the method of calculating for indicators of tobacco casing uniformity is one of the seven key contents of "tunnel-type multi-nozzle casing machine", which was proposed by State Tobacco Monopoly Bureau as development project in tobacco industry in 2007. By studying a batch of tobacco leaves before and after casing process from Nanjing cigarette factory, a casing uniformity detection system in primary processing line is built in this paper.The system is composed of three parts: hardware, software and detection algorithm. The hardware part: The computer gets the images of tobacco leaves in the black box through a digital camera which is connected to the computer by USB; The software of the system is programmed mainly by VC++6.0, including the image read, pre-processing, image enhancement, feature extraction and so on. The application of wavelet filter and neural network is programmed by Matlab7.0. By using Matlab7.0 toolbox functions, we can reduce the work intensity of programming, streamline the source code and improve the efficiency of the programming. And Access databases are added to the system to facilitate the management of the collected data; the detection algorithm part: We use the changes of the histogram peaks of the tobacco images from different channels before and after casing process as image features. By using image features of the tobacco leaves, the neural network can judge the comparability of the casing uniformity between the test leaves and the training leaves. It can be concluded that the whole casing uniformity is good if the casing uniformity of test leaves and training leaves is similar.Neural network method has special application for the complex, ambiguous and crossed information. And digital image processing technology is already widely used in industry, health care, aerospace, military and other fields, playing an increasingly important role in the national economy. In this paper, after theoretical analysis, algorithm research and a large number of experiments, it is concluded that the combination of digital image processing and neural network pattern recognition to detect the casing uniformity of tobacco leaves is feasible.