Research On Non-woven Fabric Automatic Detection Technology

In recent years,various highly contagious and highly pathogenic public health problems have occurred frequently,from the Ebola virus outbreak in Africa in 2014 to the new coronavirus spreading around the world at the end of 2019,all of which have caused troubles to human social development and physical health.In order to obtain adequate protection,the demand for various medical and sanitary products has increased rapidly.According to statistics,during the outbreak of new coronavirus,the daily output of domestic masks can reach 76 million.This is a very large number,but it still cannot fully meet domestic demand.It is not difficult to find that the production of masks and other personal hygiene protective products in our country occupies a large market,so academic research and project development around the production process of related products have extremely important significance and value.Based on this consideration,this article focuses on the detection of foreign matter defects in non-woven fabrics.Non-woven fabrics are raw materials for the production of disposable medical consumables such as masks and surgical caps.Due to the fact that the transportation process and storage environment cannot fully guarantee purity,there may be some foreign matter defects on the surface.This article refers specifically to two types of defects,hair and insects.Production enterprises need to screen out these foreign matter defects,otherwise it will greatly affect the quality of subsequent products and cause huge economic losses.In the past,the production of enterprises usually used traditional manual visual inspection methods to detect defects,but this method is inefficient,low precision,and will cost higher labor costs.It has been gradually eliminated by enterprises.Based on the above situation and the characteristics of non-woven fabric defects,this paper proposes a non-woven fabric defect detection algorithm based on traditional machinelearning and deep learning respectively,and designs an effective hardware and software platform to realize the non-woven fabric defect online automatic detection.After more than a year of actual operation and later modification and maintenance,the system has fully met the production requirements of the enterprise for its stability,accuracy and real-time performance.The main research contents of this article are as follows:Based on computer vision and machine learning knowledge,a non-woven defect detection machine learning algorithm based on the optimal Gabor filter is proposed.The direction and frequency of the optimal Gabor filter are selected using the maximum one-dimensional entropy and maximum variance for the insect and hair defect images Respectively,by calculating the length,width,defect duty cycle of the minimum circumscribed rectangle,image energy,and entropy of the image defect of the nonwoven fabric,construct the feature vector,construct the self-made nonwoven fabric defect data set,train the XGBoost model,on the test set An F1 score of 0.824 was obtained.Based on deep learning knowledge,a non-woven fabric defect detection algorithm based on the MobileNet-V2 network is proposed.A non-woven fabric defect data set is prepared for model training.For the problem of too little training data in the preparation process,Image processing method such as image rotation and translation and artificial manufacturing defect method are used for data enhancement.The prepared data set is used to fine-tuning the MobileNet V2 model using pre-training parameters.Experimental results prove that the defect detection method can effectively reduce the amount of calculation,and can ensure a high detection accuracy and good real-time performance.In the development of non-woven fabric defect detection system,developed and designed a complete set of non-woven fabric defectsInspection system,the whole system can realize automatic collection of non-woven fabric images,automatic detection of defects,and according to the defect detection result controls the execution mechanism to shut down to realize the screening of non-woven fabric defects.