Research On Jacquard Fabric Defect Detection Method Based On Visual Saliency And Convolutional Neural Network

The detection of fabric defects is essential for controlling the quality of textiles.The surface texture of jacquard fabrics is complex and the detection is difficult.The existed manual detection methods are inefficient and low in accuracy,which can no longer meet the current textile detection needs.At present,using machines to replace human eyes to detect fabric defects is the goal pursued by modern textile enterprises.Visual saliency means that humans can find regions of interest from a complex environment,highlight regions of interest and ignore regions of interest;convolutional neural network algorithms can simulate the mechanism of the human brain to analyze images to find internal laws and rules.Therefore,this paper carried out a research on the detection of defects in jacquard fabrics combining visual saliency and convolutional neural networks.The main research contents are as follows:(1)Jacquard fabrics have complex textures and diverse patterns.In order to eliminate background interference,this paper proposes a fabric image processing method based on visual saliency.This method inputs the collected fabric image into the visual saliency model,suppresses the background information of the image,and highlights the saliency of the defect area to obtain the saliency map of the image.Experiments were carried out on regular jacquard fabric images such as grids and stripes.This method can effectively highlight the saliency of defects in the image and suppress the saliency of the background.(2)The fabric image with defects can accurately display the size,position and shape of the defect after the visually significant image processing method,but for the defect-free image,after the visually significant image processing method,the background pattern is also highlighted Shown,resulting in a higher false detection rate.In order to solve this problem,this paper proposes a jacquard fabric defect detection method that combines visual saliency and convolutional neural network.First,use the visual saliency model to process the fabric image to obtain the saliency map of the fabric image;then,use transfer learning to classify the saliency map of the jacquard fabric image with the convolutional neural network model trained on the general data set.Through experiments on different fabric data,the results show that compared to using convolutional neural networks to directly classify jacquard fabric images,the accuracy of the method proposed in this paper is improved by 19%,and it is suitable for detecting defects in jacquard fabrics methods.