The Research On Defect Detection And Classification Algorithm Of Masks Based On Deep Learning

With the rapid development of urban development in China,environmental pollution has intensified,resulting in an increase in the number of cases of respiratory diseases and increased risk of bacterial infections.Masks have evolved from medical,industrial,and specialty applications to mass consumption.The demand for supplies and masks gradually increase.In the production process of masks,various kinds of defects such as loss of ear cords,surface damage,etc.will occur,and the current mask standards require the appearance of the masks to be intact.Therefore,the detection of surface defects on masks has certain practical significance.In recent years,deep learning has achieved great success on many issues of computer vision such as image recognition,target positioning,face recognition,and the like.It is worth noting that the current major progress and application of the industrial sector is largely dependent on high-cost supervised deep learning.However,in many practical scenarios,there are problems that the cost of data acquisition is too high or even difficult to obtain.Therefore,this article will explore how to use deep learning technology to detect and classify defects in the case of less data.Firstly,for the mask defect images collected by black-and-white camera,the traditional gray threshold segmentation algorithm is used to detect the defects,and the gray cooccurrence matrix(GLCM)features and LBP features of the images are further extracted.Then the BP neural network classifier is used to classify the defect types..The defect detection algorithm uses a variety of image enhancement techniques to preprocess images,thereby reducing the influence of noise and background information on the detection of mask defects.Finally,the mask is divided into sub-thresholds.The classification algorithm calculates the GLCM in the four directions of the image,and then extracts the four statistics of the GLCM as the features of the image.Finally,a three-layer BP neural network is constructed to classify the defects.Secondly,aiming at the difficulty of extracting image features manually by the traditional BP network and the limited sample size,combined with the convolutional selfencoder(CAE),a stack denoising autoencoder algorithm(FSDAE)based on the Fisher criterion is proposed.The algorithm first intercepts several small images from the original image,and uses sparse autoencoder(SAE)training to get the sparse features of the small image.Second,using this feature to initialize the CAE network parameters and extract the low-dimensional features of the original image;This feature data is sent to the FSDAE network for defect detection and classification.The masks and two kinds of fabrics were tested.The experimental results show that the proposed algorithm can effectively improve the detection rate of masks and has certain universality.Another problem is that it is difficult to detect color mask defect images.In this paper,by analyzing the advantages and disadvantages of the current mainstream convolutional network model and combining migration learning theory,this paper proposes a classification detection algorithm based on deep convolutional network.And using the separation of thoughts to improve the recognition rate of the masks,this idea was then extended to fabric testing,and good detection results were obtained.Finally,by analyzing the differences between different models and network models in different hardware,it provides a certain guiding significance for industrial detection of masks.Finally,a complete inspection system was designed based on the overall algorithm,and the currently completed research work was reviewed and the future development trend of the project was summarized.