Research On Information Extraction Algorithm Of Weld Image

In the era of industry 4.0,the intelligent manufacturing is regarded as the core,the industrial intelligent robot industry market presents an explosive growth momentum,in which machine vision plays an important role as the "golden eye" of industrial robot.The characteristics of machine vision system are to improve the flexibility and automation of production,and reach high precision,high efficiency,and they have played an important role in many occasions.Because of these characteristics,the manual detection often exist false detection,missing detection and other situations,machine vision began to be applied in weld defect detection,and it gets good results.In weld defect detection,the segmentation and extraction of weld defect image is very important,which is related to the accuracy of detection.In this paper,the X-ray weld image is taken as the primary research object,and the traditional image processing and deep learning methods are used to segment and extract the weld defects.In the traditional image processing,according to the steps of graying,image filtering,image enhancement and image segmentation,the segmentation of weld defects is completed.Aiming at the possible noise of weld image,the adaptive median filter is used to eliminate the noise of the image;the contrast of weld image is relatively low.In this paper,the image enhancement method based on incomplete Beta function and sparrow search algorithm is adopted to increase the gray difference between foreground and background,and it improve the image quality effectively.In image segmentation,the edge detection,threshold segmentation and region based segmentation are compared.The traditional image processing is based on the gray level of the image to complete the task of image segmentation.The gray level difference between the foreground and background of the weld defect image is small,so this method can not achieve the ideal segmentation effect.In this paper,several deep learning models are introduced.Deeplabv3+ deep semantic segmentation model combines the advantages of coding and decoding structure and empty space pyramid pooling,and it has played very well in the task of semantic segmentation of natural scenes.The defects in the weld image are relatively small,so the Deeplabv3+ semantic segmentation model is improved,and the multi-scale feature extraction is carried out with smaller parameters.After training,the accuracy of the model can reach 90%.Using the improved Deeplabv3+ model to segment the weld defect image,the result is closer to the ideal manual segmentation weld defect than the traditional image processing.