Research on cascade detection technology of tobacco impurities images based on computer vision and machine learning.

To improve the accuracy of tobacco impurity detection and removal, a cascade detection method for tobacco impurity images based on computer vision and machine learning was designed. This method used color features and gradient energy computer vision methods to locate tobacco impurities. Combining HOG and LBP features with a cascade Adaboost classifier, a multi-window detection algorithm was designed to detect tobacco impurities in real time. The experimental results showed that the accuracy of the static impurity detection method based on color features was higher than that of the gradient energy method. After combining HOG features and multi-level cascade Adaboost classifier algorithm, the detection results were very stable, and the accuracy of tobacco impurity detection reached 97. 33%. In the actual operation process, there was no need to manually adjust the algorithm parameters, ensuring the accuracy and effectiveness of the algorithm while reducing the time cost. [ABSTRACT FROM AUTHOR]