Intelligent Detection Of Welding Defects Based On X-ray Image

Welding is the main way to combine steel structure.X-ray flaw detection,as an intuitive and reliable detection method,can almost comprehensively inspect these areas.It is also the mainstream detection method in the engineering field today.If it can be combined with popular intelligent detection technologies such as image detection,it will bring great benefits to the society.The method flow of intelligent detection is studied by means of image processing and pattern recognition for radiographic images with welding defects.The image preprocessing mainly includes denoising and contrast enhancement of welding ray image.The wavelet filter is used for denoising and contrast enhancement,and compared with the classical denoising algorithms,it is found that the wavelet filter has certain advantages in the field of ray image denoising.In order to improve the operation speed and recognition accuracy of image processing,the region of interest of defects is firstly extracted.During the extraction,it was found that the common edge detection algorithms were not satisfactory,and the region of interest was extracted by combining threshold segmentation and edge detection method,and the ideal weld edge was obtained by using Hough transform.The morphological watershed algorithm was used to segment the defects,and artificial line fusion based on endpoint judgment was designed to solve the discontinuous defect after the linear defect segmentation,and the continuous long linear defect segmentation results were obtained.Through the random operation of the samples,the sample set is expanded to get 2193 samples and the sample is normalized,which basically meets the requirements of image classification training.In terms of traditional machine learning ray image defect classification,texture features such as local binary mode and gradient direction histogram are extracted,and then the classical binary tree support vector machine is selected for image multi-classification.After classification experiment,it is found that the recognition accuracy is below 80%.Then,a multi-feature fusion method is designed for feature extraction.After classification experiment,the recognition accuracy rate is improved to 86.1%,which can basically meet the requirements of the occasion is not high,but for the occasion of high accuracy needs to be higher recognition accuracy.In the aspect of deep learning image recognition,the VGGNet convolutional neural network with 13 convolutional layers and 3 fully connected layers was used to classify the defects,and the classification accuracy reached 92.8%.In order to reduce the time and cost of classification for practical application,a two-stage classification model was proposed.The accuracy of classification was increased by 1% after the determination of defects and the subdivision of defects.Two-stage classification avoids the time consuming of subdivision on normal image and the interference of normal image noise,and also improves the accuracy of classification.After several comparative experiments,this paper selects the required intelligent welding defect detection method based on X-ray image,and obtains the accuracy rate of welding defect image classification of more than 90% through the experiment,further narrowing the error rate and shortening the gap of its practical application in production.