Abstract:

Machine-harvested cotton impurities are multiply. It is important to detect the type and content of impurities for adjustment of the processing technique of cotton. An impurity detection method based on region color segmentation was presented. During image segmentation stage color gradient image was obtained based on filtered image firstly. Marking image was achieved by H-minima transform, and initial segmentation image was acquired based on modified gradient image by watershed algorithm. Then region merging was conducted for initial segmentation image. Region adjacency, region color feature and region area were considered for region merging. Region color features such as saturation, intensity, region color vector module and color similarity were used. Repeated merging was adopted, and information of color feature was updated in different merging. Finally various features including color, texture and shape were extracted through support vector machines algorithm for impurity recognition. Experimental result show that a successful recognition rate of 94% for natural impurities is achieved.

Key words: machine-harvested cotton, image segmentation, impurity recognition, marker-controlled watershed, rigion merging, color feature