基于孪生网络模型的织物疵点检测方法OA

To solve the issue of lower efficiency in fabric defect detection in industrial settings,a fabric defect detection method based on a siamese network model was proposed.By analyzing the similarity detection mechanism of siamese networks,a similarity detection model based on siamese networks was constructed,which included three modules:feature extraction module,similarity measurement module and defect classification module.Firstly,Inception structure was introduced to construct feature extraction module,sample features were extracted at multiple scales.Secondly,in similarity measurement module,the abstract features of the sample was translated into difference calculations,the sensitivity of the siamese network to the difference was used to realize the rapid identification of fabric images,to complete the first stage of judging whether the fabric had defects.Finally,in the defects classification module,defects images were input into YOLOv4 network to achieve accurate defects classification,to complete the second stage of classification task.Experiments on public datasets showed that the proposed method can effectively detect fabric defects,mAP value was reached 87.54％and detection speed was reached 54.8 frames per second,the better balance between detection accuracy and speed could be reached,the practical industrial production detection requirements of textile enterprises could be met,which could also provide a novel solution for fabric defect detection in the textile industry.