In order to improve the degree of automation in the detection process of yarn cone, this paper proposed an on-line detection system for detecting yarn cone defects a on the top surface and sides. The system is composed of two industrial cameras,an LED strip light, a photoelectric sensor and a computer. Firstly, the top images under the overexposure mode and normal mode, the side images under the overexposure mode were time-sharing collected by the camera and light combinations. Secondly, for the top overexposure image, the center of the yarn cone was located in by applying adaptive segmentation method. Thirdly, for the top image of normal mode, the transformed image was carried out after polar coordinate transformation, and the curly core yarn defect, the multi-source yarn defect and the net yarn defect were respectively detected based on analyzing the projective features of vertical edge distribution, analyzing texture and intensity distribution consistency, and local texture direction histogram. Finally, for the side overexposed image, the boundary position was quickly located by projection method, and then the multi-layer defect was determined by analyzing fitting degree of the contours. The experiment results show that the proposed system can identify some kinds of yarn cone defects, including muilti-layer, net yarn, curly core yarn and muilti-source yarn, with high detecting accuracy.
Template