Design And Implementation Of Detection System For Surface Of Cups Ofinstant Noodles

The competition of instant noodles market is becoming increasingly fierce; the manufacturers are paying more attention to the packing quality of the cups of instant noodles and therefore the packaging quality of surface detection is becoming more and more important. At present the traditional inspection methods are manual inspection and there are many disadvantages such as low-speed, low-precision and high labor costs. Study shows that the detection system for surface of cups of instant noodles based on the machine vision inspection technology with high-speed, high-precision, and strong anti-interference ability is the preferred solution. In this thesis, the application of machine vision is provided to detect packaging quality of surface about the cups of instant noodles. The main work and contributions were as follows.(1) On the analysis of packaging of surface about the cups of instant noodles, the hardware architecture and software architecture of the quality inspection system were designed from the perspective of machine vision.(2) The segmentation and preprocessing of images. This is the foundation of the whole software system. In this thesis several classical algorithms were discussed and the algorithm which combining the Otsu and the entropy principle was adapted at last. Having run some tests, it could be found that this algorithm could get better results.(3) The search of component-labeling algorithm and extraction of features. This is the key part of the whole software system. In this thesis two algorithms including the pixel-based component-labeling algorithm and the run-based component-labeling algorithm were tested, what is more, the algorithm based on the pixel was improved in this thesis. After the search of component-labeling, some features such as area、numbers、perimeter and so on can be worked out and the most helpful features were selected as the input of the classifier at last.(4) The design of classifier. This is the kernel part in the whole software. In this thesis two different classifiers were discussed such as C4.5 decision tree and the Bayes classifier. The calculation was optimized based on the L’Hospital’s rule in C4.5 decision tree. According to the test, it could be found that the decision tree could achieve a high rate of classification under the condition that the numbers of categories and features are all known.According to the results of the above, the algorithm which was suitable for packaging quality of surface detection was proposed. Finally, the system had been tested and the results indicated that the performance could meet the requirements of production including detection precision、stability and real-time. At present the whole system had already consigned a usage and it will bring huge economic benefits to the enterprises in the long run.