| In recent years,air pollution is serious,and the production of non-woven masks is increasing year by year.Non-woven masks have higher requirements on hygiene and quality,but there are many defects in the production process of non-woven masks,such as white silk,insects,stains,hair,holes,nose strip defects,ear rope defects,etc.The traditional manual visual inspection method is difficult to meet the requirements of high precision,high speed and good stability in industrial production.Although the research on woven fabric defect detection based on image processing and deep learning has made good progress,it can not meet the needs of real-time detection of non-woven masks in terms of accuracy,speed and stability,and there is no mature and reliable automatic online inspection system to monitor and detect all aspects of non-woven mask production.In view of the above difficulties,this paper designs a complete automatic on-line detection system for non-woven masks based on the principle of machine vision and deep learning,and realizes the automatic online detection of non-woven masks.Through half a year of online test operation and improvement,it can fully meet the needs of automatic online detection of non-woven masks in terms of accuracy,speed and stability.The system has great application value and can be used as a model for automatic online inspection systems for other industrial products.The main research contents of this paper are as follows:1)Research on automatic on-line detection technology of non-woven fabrics.The texture of the non-woven fabric itself is disorderly and uneven,and the defects are too small and the shape,size and position of the defects are greatly different.This paper combines the generative adversarial network and the convolutional neural network to detect defects.First,the image is randomly cropped to obtain more defect samples.Second,the image is augmented by rotating,translating,changing contrast and brightness,adding noise,etc.,and then generating a defect pair by using an anti-generation network pair to augment the data set.Finally,convolutional neural networks are used to classify defects.Through the automatic online test for half a year,the average detection rate for defects of non-woven fabrics reached 98.10%,and the false alarm rate was below 0.01%.2)Research on automatic on-line detection technology of non-woven masks.There are many types of non-woven masks,many of which are artificially defined,which leads to a vague definition of defects.Due to the characteristics of flexible materials of non-woven masks,it is difficult to accurately extract the characteristics of non-woven mask defects.In view of the above situation,this paper proposes a detection method for specific defects of non-woven masks.Firstly,this paper proposes the positioning and extraction of non-woven masks based on Hough transform.Secondly,this paper proposes to use the sub-region OTSU for region segmentation,and then use the Canny operator for edge detection to extract the ear loop.Thirdly,this paper uses the prior knowledge of gray value statistics and the position of the non-woven mask to detect the ear band defects,nose strip defects and other defects of the non-woven mask.Through the automatic online test for half a year,the average detection rate of the system for non-woven mask defects reached 99.2%,and false positive rate was below 0.2%.3)Design automatic inspection system for non-woven masks.First,the core hardware devices such as cameras,lenses,light sources,and computers are selected according to actual needs.Secondly,a lighting scheme suitable for industrial site and non-woven mask detection is proposed.Third,through some auxiliary hardware such as relays,valves,lights,buzzers,etc.,to achieve control of the production process.Fourth,design the software module of the automatic online detection system for non-woven masks.The hardware and software of the detection system have been running stably for half a year,and the indicators in all aspects are good. |
PDF Full Text Request