Abstract:

Objective Conventional manual testing relies on the subjective experience and rating standards of inspectors to complete the appearance quality testing and evaluation of fabrics, which has problems such as backward productivity, poor detection accuracy, low efficiency and easy fatigue. Fabric defects automatic detection technology is one of the key links for textile enterprises to develop into intelligent manufacturing. Thus, this paper intends to develop a real-time detection system in order to achieve the automatic detection of fabric defects so as to overcome the disadvantages seen in manual detection.

Method The system adopts the motor drive to realize the fabric winding and the automatic transfer of the roll. Unwinding and transmission can be stably, with high automation and accuracy. In order to meet the different lighting requirements, three rows of LED lights are installed, and they have more lighting modes than other systems. Eight industrial cameras are arranged side by side to realize the image acquisition of the fabric. The acquired images were rapidly detected by the image defects based on an improved Itti salient model fault detection algorithm. The model has shorter detection time for fabric image and has higher accuracy, which can meet the real-time detection requirements of fabric defects.

Results The schematic diagram of the fabric image acquisition system is established (Fig. 1). The fabric is rewound by the motor and can be stably transmitted to the image acquisition area in a specific route. In the image acquisition area, fabric images of different thickness fabrics with different light sources are obtained (Fig. 3). It can be seen that the sharpness of the images taken by different light sources is different, which meets the detection requirements of different thickness fabrics. It also indicates that the installed camera has a high shooting definition. The images were detected based on the improved Itti salient model. Different directions can effectively extract the features of the fabric image and detect the edge information inside the image. The fabric fault significance graph is obtained by manipulating the normalized brightness and orientation feature, and the significant graph is divided by the custom threshold to effectively detect the defect information(Fig. 8, Fig. 9). It can effectively detect fabric defects in industrial grey fabric and denim, such as oil and holes. The defect detection rate is 93%. Compared with other fabric defect detection algorithms, the detection accuracy is higher. At the same time, it can be seen that the detection time of this method is short (Tab. 3), and the detection speed is 48 m/min. The real-time detection speed is further improved.

Conclusion In order to improve the disadvantages of the convenitional artificial fabric fault detection, a fabric image acquisition platform and a real-time fabric fault detection system based on the significance detection algorithm are proposed. The fabric platform can be driven by a motor, which is more stable than the previous roll transmission system and takes clearer photos. The improved significance detection algorithm detects the images and achieves good detection results. By comparison, the method has high detection accuracy and real-time performance, and the detection time of the algorithm meets the requirements of dynamic detection. The designed fabric real-time detection platform can run effectively and stably, and have higher real-time detection performance.

Key words: defect detection, real-time detection, Gabor filtering, Gaussian pyramid, Itti saliency