Research On Defect Detection Algorithm Of Uneven Illumination Fabric Image Based On Deep Learning

Fabric defect detection is the key of textile quality control.The light source changes in the process of visual inspection will cause uneven illumination in the fabric image.However,traditional fabric defect detection methods do not work well in complex lighting environments.With the rapid development of deep learning in recent years,it has a wide range of applications in the field of image enhancement and target detection.The deep learning technology is used to study the defect detection algorithm of fabric images under complex illumination.In order to get close to the fabric image in the real environment,the research objects in this paper include two fabric types and three common defect types.The image is collected by self-built image acquisition system,and then the sample number is expanded by data enhancement method after fusion with the same type of open source fabric data set TILDA.Finally,fabric data sets containing complex lighting effects are constructed to provide sufficient samples for subsequent deep learning training.In this paper,deep learning is applied to fabric image preprocessing.The convolutional layer and transpose convolutional layer are used to extract features and reconstruct images respectively,and then the appropriate loss function is selected to form a new convolutional neural network framework which can be used to remove uneven illumination in fabric images.Compared with the traditional image enhancement method,the experimental results show that the image processed by the proposed method is clear and of higher quality.According to the characteristics of fabric database,an improved YOLOV4 algorithm is used to detect defects in fabric images.In order to solve the contradiction between the feature map and the receptive field in the case of high-resolution input of the fabric image,the FPN part of the YOLOv4 framework is replaced with AC-FPN;The Focal Loss function is introduced into the YOLOv4 loss function,which can solve the imbalance between the target and the background.By comparing the improved YOLOV4 algorithm with the original YOLOV4 algorithm,the results show that the MAP value increases by 5%when IOU is 0.5.A detection method based on deep learning is proposed to detect defects in fabric images under complex illumination.This method can be used for grey cloth and printed cloth with high accuracy and robustness.And it is suitable for fabric image defect detection under complex illumination.