The Study On Wolfberry Image Classification And Its Application

Traditional wolfberry classification mainly adopts manual distinguishing mode depending on size,color and surface defects.In this mode,because of individual subjective error and the different level of fatigue,the classification process and standard can not be consistent and the process is more time-consuming,so it can not meet the needs of wolfberry classification.So currently implementing machine vision for wolfberry classification is not only improve classification efficiency but also realize nondestructive detection,it has became one of the focus research of wolfberry classification.This dissertation focus on method of wolfberry classification.Initially the wolfberry image that was preprocessed and denoised will be segmented,and then taking the segmented wolfberry particles' area and length-width ratio as the wolfberries' classification characteristic parameters,after the clustering analysis the wolfberries can be classified completely.Finally a software interface for wolfberry classification will be set up by using the presented method in this dissertation.The main job of this dissertation following several aspects:(1)Aiming at shadow noise produced in process of wolfberry image acquisition or some detail noise,inconsistent gray intensity and fuzzy boundaries in its gray image,extracting the red component to remove shadow noise,employing the algorithm of morphological multi-scale opening and closing reconstruction to smooth and maintain the information of contour edges of wolfberries.(2)Aiming at some accurate segmentation problems result from more particles' touching,a method based on concave points matching is proposed.Firstly extracting the contour edge of wolfberries by employing 8 neighborhood tracking algorithm,and then using a circular template to detect concave points in touching wolfberries' contour edge,taking the shortest Euclidean distance between concave points as matching condition to connect the concave points.Finally using the improved matching rules proposed in this dissertation to revise some mismatching concave points to complete touching wolfberry segmentation,and through contrast experiments the effectiveness of the proposed segmentation method was analyzed and verified.(3)A method for classifying wolfberry image has been proposed.Initially,depending on the R component distribution of the single wolfberry in the image,the mildew wolfberry particles will be removed,then using the area marking algorithm to scan and mark all segmented regions to obtain the areas of all target particles,and employing Hotelling algorithm to acquire particles' length-width ratio,taking the length-width ratio as the revised weighting of the area to do clustering-analysis by using K-means algorithm,based on the classification standard which obtained from wolfberry samples' training all the wolfberries can be classified,comparing to manual classification and analyzing quantitative results the accuracy and efficiency of the proposed method was verified.Finally the proposed approach is used to establish a software interface for the process of wolfberry classification by using Matlab Graphical User Interface,and this software is tested by using many of wolfberry images.The testing results show that the proposed approach has higher classification accuracy and efficiency for more wolfberry particles' touching conditions.