棉 花 学 报     Cotton Science     2007,19(2):119-123

 

Color Features Selecting and Grades Clustering for Preharvest Cotton Images with Dark Background
WANG Ling1,JI Chang-ying1*,CHEN Bing-lin2
(1.College of Engineering,Nanjing Agricultural UniversityNanjing 210031,China;2. Key Laboratory of Crop RegulationMinistry of AgriculureNanjing Agricultural UniversityNanjing 210095,China)

Abstract:In order to assess the grades of preharvest cottons objectively,boll size and fiber color features,including yellow region,yellow degree,white degree and brightness contrast of cottons with/without bracteoles in six typical color spaces,were investigated based on Chinese government grading standards and machine vision technologies. The results show that better discriminations of feature parameters can be obtained in RGB,NTSC,Hunter and HSI color spaces,white degree is unvalued according to the correlation analysis between boll size and fiber color features. K-means clustering were performed based on valid image features to grade cotton samples into seven categories in each color. The results indicate that clustering results are independent of color spaces,bracteoles color contributes to their grades obviously and HSI color space maybe the best color spaces for grade clustering because of higher and more uniform correlations between grades and features as well as shorter runtime can be obtained in this space. Machine vision can be used to improve sorting accuracy of preharvest cottons since its more exact discriminability comparing with that of human eyes.
Key Words:image features;color space;cotton grade;clustering;validity  [Full Text,2092KB]