Knitted Glove Defect Detection System Based On Embedded Machine Vision

When knitting gloves on a glove machine,two or more yarns are usually combined to form a plied yarn for knitting.Currently,the yarn break detection device cannot detect the breakage of individual yarns in plied yarns.On the other hand,due to the high speed of the needles running in the needle bed,it is difficult to detect broken needle faults in a timely manner through direct detection methods.If these two most common but difficult to detect problems are not found in time,they will cause a large number of defective gloves and waste resources.To solve these two problems that are most likely to occur but cannot be directly detected,this paper conducts research on an indirect fault detection system based on glove product images,and also implements detection of three types of defects: glove holes,oil stains,and skipped needles.This creates conditions for automatic rejection of defective gloves in the subsequent packaging process.Firstly,this paper introduces the detection requirements of knitted gloves,analyzes the characteristics of glove defects,designs the overall system scheme,and parallel detection process for glove defects.According to the characteristics of embedded machine vision,the AM5708 chip is selected as the core of the system to implement the corresponding detection algorithm.An embedded operating system is configured,and an information exchange interface is designed for the transmission of various detection results,such as yarn breakage and needle breakage.Secondly,in response to the problem of needing to try and adjust camera parameters after changing gloves or environmental conditions,a knitted glove image quality evaluation index based on image gray histogram feature parameters and contour segmentation as the target is constructed.This index can quantitatively calculate various parameters of glove images,providing a basis and foundation for adjusting the shooting function of cameras for online detection to achieve the best shooting effect.The evaluation index is independent of factors such as the color of the gloves themselves,and has good generality,which can reduce the need for manual intervention.Then,the glove defect detection algorithm is studied.The image is further optimized through preprocessing such as image position correction and filtering.The defect detection algorithm uses color space conversion and edge detection to extract internal defects of gloves,and finally distinguishes different types of knitted glove defects based on features such as shape,size,and color.To address the problem that the commonly used Gaussian filter has poor generality,a specific convolution kernel is self-generated to achieve applicability to various gloves and reduce computational complexity.To improve the weak edge detection effect of the traditional Sobel operator,a custom convolution template is defined to improve the accuracy of the gradient algorithm.Finally,a detection experimental platform is built to test the performance and functionality of the defect detection system.The experimental results show that the proposed evaluation index has good consistency with subjective evaluation by human eyes,and the camera can automatically adjust parameters to capture the best image.The defect detection recognition accuracy is high,and the system can achieve an overall correct rate of 93% while keeping the detection time within 50 seconds,meeting the design requirements.When the system detects defects in knitted gloves,information transmission can be achieved through serial ports and other reserved ports,ensuring the practicality of the system..