Automatic weld defect detection of pipes with YOLO v۱۱ and preprocessing with Gaussian filter and background removal Method

Welding of oil and gas pipeline is of great importance and contributes significantly to the safety of natural gas transportation. Currently, radiographic examination with film and detectors is used for the quality inspection of welds. However, inspection methods for automatic detection with high accuracy for different sizes and types of weld defects face difficulties. To address this challenge, this research presents an intelligent method for defect detection based on a deep learning approach. First, a contrast enhancement method is applied to a set of radiographic images processed based on the Gaussian filter algorithm and background removal method. Then, a YOLO v۱۱ network architecture is developed from these enhanced radiographs to learn the feature distribution of several common defects. This model uses multi-scale feature fusion methods to effectively identify different sizes and types of defects. The results obtained from the network trained on a set of radiograph images show that using YOLO v۱۱ with processed images based on the Gaussian filter algorithm and background removal method compared to using YOLO v۱۱ on unprocessed radiographic images achieves a ۲۰% improvement in automatic weld defect detection and can effectively improve inspection efficiency and promote the development of automatic X-ray detection.

کلیدواژه ها:

Automatic detection ، Industrial Radiography ، weld defects ، pipes ، YOLO v۱۱ ، Gaussian filter ، background removal Method