Abstract: In order to increase productivity, reduce subjective grading error and improve quality, it proposed a method of using machine vision for automatic ratings of feather color. First, the eigenvalues of images were extracted and analyzed, based on the YU‘V’ color space. Then, an optimized binary tree of multiclass classifiers was built, based on SVM, combined with its characteristic value. It was used to classify the feather color. Finally, the classification error was analyzed. The experimental results show that the designed classifier has a high correct rate, and that it can efficiently determine the feather color level, thus satisfying the production requirement for automatic grading of feather.

Key words: machine vision; feather; color space; SVM; binary tree；multiclass classifiers