Intelligent Evaluation System For Similarity Of Industrial Ray Digital Negatives

Because of the complex environment and the long distance of the welding joints,it is easy to have some welding joints not detected or repeated detection,which leads to the welding quality can not be evaluated,and there is a huge potential safety hazard.At present,traditional manual evaluation is used in repetitive testing of radiographic film in industry,which has many problems,such as high error rate,serious waste of human and material resources,non-standard file processing and storage format,which is not good for industrial production and application.This paper studies the DICONDE standard of digital image processing and communication,and designs a set of intelligent evaluation system of similarity of industrial radiographic digital film,which solves the problem of repeatability detection of radiographic film by means of repetition detection of effective information of negative image and contrast of similarity of negative image.The main works are as follows:1)The DICONDE file standard of digital image processing and communication is studied in view of the non-standard file processing and transmission format in the field of nondestructive testing.The DICONDE file format and TIFF file format are analyzed and studied.On the basis of the research,the DICONDE file construction,display and analysis are realized.The conversion between TIFF file format and DICONDE file format is completed.2)Aiming at the problem of repetitive detection of valid information in negative images,the image features are first preprocessed,including HE equalization,binarization,etc.;then character recognition is performed,which mainly includes the construction of character data sets,and the establishment,training and recognition of paddleocr network;finally,the repetition detection of the valid information identification result and the stored negative number in the database is completed.3)Aiming at the problem of similarity comparison of negative images,a method for extracting region of interest based on Gaussian difference was designed,the weld area was extracted and processed in blocks,a twin neural network model based on each block was constructed.The block-weighted image similarity comprehensive evaluation method,and the visualization function based on the Qt framework is used to complete the repetitive detection of radiographs.Through the system test,the intelligent evaluation system of industrial ray digital film similarity designed in this paper has perfect function and runs stably.The accuracy of image similarity comparison is 85%,and the character recognition average accuracy is 90%.The transformed DICONDE file meets the national standard.All the indicators meet the requirements of the cooperative enterprises.