Surface Defect Detection Of Breathing Mask Based On Deep Learning

In recent years,with the rapid development of deep learning theory in the field of computer vision,deep learning-based technology is widely used in image recognition and detection problems,such as the defects detection of rail surface,green coffee surface,glass surface and so on.With the improvement of China's economy and the increasing use of cars in cities,the haze becomes more and more serious.More citizens need to wear breathing mask to go outside.For doctors and nurses in the hospital,breathing mask is also a necessity.At present,the majority of breathing masks on the market are produced by non-woven fabrics.During the production of non-woven breathing masks,there are some defects,such as knotting in the ear band,the ear band falling off,hair and stains on the breathing mask body,etc.These defects will affect the mask user's wearing and health.At present,traditional image processing methods are widely used in factory production line.These methods require manual features setting,which lacks universality and expansibility.In order to solve these disadvantages in traditional methods of image processing,this paper uses deep learning algorithms to detect and recognize surface defects of breathing mask.Unlike the traditional methods of image processing,convolutional neural network(CNN)in deep learning algorithm doesn't need manual features setting.It can automatically learn and extract useful features information of image.This paper mainly studies on below contents.1)A deep CNN recognition algorithm of breathing mask surface defects based on transfer learning is proposed.This algorithm makes a small adjustment to the vgg-16 model adding BN normalization and GELU activation function.This algorithm is used to identify and detect surface defect image of breathing mask using data enhancement,image preprocessing and transfer learning.The experiment result shows that this method has higher recognition rate than the traditional image processing algorithms.2)This paper proposes a deep CNN recognition algorithm of breathing mask surface defects based on multi-feature fusion.This algorithm,which is based on last algorithm,uses the way of features fusion to make up for the traditional CNN deficiency of losing some part of feature information in the process of feature extraction.Compare with the previousalgorithm,it shows that this algorithm has a high recognition rate for the surface defect images of breathing mask and other industrial parts,and its generalization is better.3)This paper presents a fast and accurate CNN algorithm for surface defect recognition of breathing masks.This algorithm uses dilated convolution to extract and keep more image feature information.With the use of BN and IN normalization and PReLU activation function in this method,it not only has a high recognition rate for the defect image of breathing mask,but also has a faster recognition speed for each defect image of breathing mask than the traditional image processing algorithm.