|
The production process of modern tobacco machinery is highly precise and complex, and when unpredictable failures occur in the production line, they can lead to economic losses for tobacco companies. Artificial Intelligence online monitoring technology can provide real-time fault alarms for tobacco production lines in production, and at the same time, without affecting the overall operation of the production line, it can improve the overall operational efficiency of the tobacco enterprises, which in turn can generate economic benefits. The fundamental process of visual inspection of cigarette package appearance quality is to classify the inspection image to distinguish qualified and unqualified packages, how to accurately judge the “qualified” and “unqualified” packages is the key to improve the accuracy and reduce the false reject rate. How to make accurate judgment on “qualified” and “unqualified” cigarette packs is the key to improve accuracy and reduce false reject rate. By analyzing the collected images of strip cigarettes, there are certain feature differences between qualified and unqualified images, in order to effectively determine the abnormal strip cigarettes, machine learning-based detection methods can be used to analyze the image feature information, extract the feature parameters, establish the cigarette packaging classification model, judge the defective strip cigarette products and reject them. The main content of the research in this chapter is to extract the image data features by combining wavelet transform and grayscale covariance matrix, and then use support vector machine and BP neural network algorithms respectively to sample and learn the training sample set of cigarette, and then classify and predict the test sample set to verify the classification performance and compare the classification effect of the two classifiers.
|
