Abstract:

Objective Jacquard knitting fabric with its diversified pattern design can meet consumers’ personalized fashion and pursuit. However, defects in the production process would seriously damage the integrity of the pattern and reduce the aesthetics of products. Due to the pattern color and edge texture, it is difficult to distinguish the normal fabric texture from the minor defects by the traditional detection method, which increases the detection difficulty and complexity. Therefore, strengthening the quality monitoring of jacquard knitting products and identifying the defects on the fabric quickly and accurately become the key link to ensure product quality.

Method Based on the obvious difference between the defect area and texture area of jacquard knitted fabric in the gray value and its distribution, the information entropy and statistics were introduced into the two-dimension feature extraction technology. The weighting fusion mechanism was designed to integrate the multi-source features, so that the optimal reconstruction of features and the geometric outline and statistical feature framework for capturing defects were achieved. These features were used as input data for the support vector machine(SVM) classifier. To optimize the performance of SVM, a new algorithm integrating arithmetic optimization and Levy flight strategy was introduced.

Results The information entropy feature and feature quantity feature were used to represent the defect image on the gray scale. The location of defects was determined by analyzing the difference between texture and defect in spatial distribution. The detection process was mainly composed of three modules. The image pre-processing module was responsible for the preliminary processing of the input jacquard fabric image, including noise removal, contrast enhancement and other steps, so as to improve the image quality and lay a foundation for the subsequent feature extraction. The feature extraction module was the algorithm using information entropy feature and statistical feature to divide the texture area and defect area by weighting fusion dual feature area, and extract information that can represent defect characteristics. The classification module was based on the feature data provided by the feature extraction module, using the improved SVM classification algorithm to classify the defects and mark the position of the defects. High detection rate was achieved for broken yarns and holes, but low detection rate was obtained for low strength defects such as knot and yarn jumping. The combined algorithm significantly improved the detection rate after image pre-processing. Single-feature algorithm was efficient for specific defects, but the algorithm in this paper maintained a high detection rate for all kinds of defects, and verified the effectiveness and robustness of its pretreatment and detection algorithm for complex texture and defects of jacquard knitted fabric. Under the input condition of multiple characteristic parameters combination, the Levy optimization algorithm-support vector machine(LAOA-SVM) classification algorithm and its comparison algorithm designed in this paper showed the highest classification accuracy. The LAOA-SVM classifier performed well on the independent test set, with an accuracy rate of 99.4%, and the time of 0.93 s was less than that of other methods, indicating that the algorithm was efficient and accurate in the fabric defect classification task. The LAOA-SVM classifier showed good classification performance after training, with only a little misjudgment between yarn jumping and knot. Up to 100% accuracy was reached for other fabric defect classifications. Especially for the regional defects with obvious similarity in characteristics, such as yarn breaking and hole breaking, the advantage of LAOA-SVM was highlighted. Other classification algorithms demonstrated a high misclassification rate when dealing with different types of defects. This result strongly demonstrated the high efficiency of LAOA-SVM classifier in fabric defect recognition.

Conclusion The image pre-processing stage effectively enhances the contrast between the defect and the texture through grayscale and filter denoising. Then, the information entropy and statistical features are used to extract the defect information, and the feature optimization is achieved by double-feature weighting fusion. The designed LAOA-SVM classification algorithm and its comparison algorithm show the highest classification accuracy. The LAOA-SVM classifier performs well on the independent test set, with an accuracy rate of 99.4%, and the time of 0.93 s is less than that of other methods, which indicates that the algorithm is efficient and accurate in the fabric defect classification task. The experimental results show that, compared with the prior art, the method performs excellently in the detection of defects such as holes, flying sparks, yarn jumping, greasy dirt and yarn breaking, with a detection rate of up to 99%, and a classification accuracy of 99.4%, especially when dealing with defects with similar characteristics.

Key words: jacquard knitted fabric defect, grayscale statistics, information entropy feature, statistical characteristic, weighted fusion, detection and classification