Design And Implementation Of Non-woven Fabric Defect Detection Algorithm Based On Deep Learning

As an important production industry in China,the textile industry has been an important force in China's economic development.As one of the textiles,the demand for non-woven fabrics is increasing day by day.Non-woven fabric in the production process will inevitably produce different kinds of defects due to extrusion,friction,oil contamination and mistakes in the bleaching process,so defect detection is very important for the evaluation of the quality of non-woven fabric.With the development of computer vision technology and artificial intelligence technology,non-woven fabric defect detection based on computer vision has achieved very excellent results.But the current defect detection algorithm still has the problems of low detection efficiency and poor adaptability.Therefore,the deep learning algorithm is introduced into the application of non-woven fabric defect detection,and in-depth research is carried out on the non-woven fabric defect detection algorithm based on deep learning.First of all,considering the low configuration of IPC in the factory production site,the detection algorithm should not be too complicated.Therefore,two detection schemes based on adaptive binarization and convolutional neural network are proposed under the premise of ensuring detection speed and accuracy.Adaptive binarization is used to locate the collected image and cut the defect area.In the first scheme,the obtained defect image is sent to the packet convolutional Alex Net network for recognition.The advantage of this algorithm is that the detection speed is fast.In the other scheme,the wavelet transform is performed on the defect image,and the weighted fusion is used to the transformed component,then the simplified residual network is used to recognize the fusion result.The advantage of this algorithm is that the recognition accuracy is higher.Both of the two algorithms can be used for detection with simple equipment and meet the requirements of industrial detection speed and accuracy.Secondly,in view of the slow detection speed of YOLOv3 algorithm and low detection accuracy of small targets,a non-woven fabric defect detection algorithm based on improved YOLOv3 is proposed.The lightweight convolutional neural network Ghost Net is used as the feature extraction network to accelerate the computing speed of the network.At the same time,the spatial attention module is added to the feature extraction network,so that the network can learn the significant regions better.Furthermore the position regression loss function is replaced with CIOU,and the feature layer is adjusted according to the actual area of the defect.Compared with the original YOLOv3 algorithm,the detection speed and accuracy of this algorithm have been greatly improved,with m AP value reaching 0.971,F1 value reaching 0.958,and detection speed reaching 0.053 seconds per frame.Finally,the deep learning based non-woven defect detection software system is built.The software can realize the analysis of online non-woven images,and realize the precise positioning and identification of defects.The software system has a simple interface,good interactivity,simple operation process,and a report form will be automatically generated after the inspection is completed.The result of joint debugging proves that the software system can run stably and has good adaptability.