Research On Recognizing Tea-leaves And Impurities Based On Image Processing And Pattern Classification

As the technology of sorting tea-leaves becomes more and more automated, it has a very brightfeature to apply the computer vision into sorting tea-leaves. It can deeply increase the automationdegree and reduce the manual working and cost. In the processing of picking tea-leaves there will besome stones, sticks, weeds and other impurities infiltrating into the tea-leaves. This paper bases onimage processing and pattern classification to recognize the impurities in the image. As a result, it cansuccessfully recognize the impurities. The impurities in the image include stick type and ball type.Different methods will be adopted to identify the different type impurities.In order to identify the impurities, process the image to highlight the characters needed andsuppress the interference information. The operation includes: graying image, smoothing image andsplitting image to separate the objects and background. Moreover, remove the shadows existing in theimage, which will influence the identification work in the later period. At last, detect the edge ofobjects in the image to get a binary image. Because of the stick type impurities in the image have linecharacter which doesn’t exist in the tea-leaves, so if the line in the impurities can be detected then theimpurities can be identified. Hough transform which has good robustness can detect lines in images. Itcan avoid the influence of voice as well. So this paper applies the Hough transform to detect lines andoptimizes the algorithm as well. As a result, the algorithm is speeded up and the stick type impuritiescan be detected very well. On the other hand, the ball type impurities in the image have differentshape with the tea-leaves. Therefore, it can be identified by a classifier. This paper introduces severalclassifiers’ design principles. And the more mature Support Vector Machine (SVM) algorithm isapplied to design a classifier. The method is labeling the connected component in the binary imagewhich has separated the objects and background at first. Then calculate the feature data of everycomponent which means every object. In the end select the optimal feature subset to design theclassifier which is on the basis of the SVM.On the above research foundation, the identification system is designed, which can identify theabove-mentioned two type impurities. And then20images which obtain tea-leaves and two typeimpurities are applied to test the reliability of designed system. And the experiment data indicates thatthe accuracy is more and90%, which demonstrate the system can complete the identification targetvery well.