Enhanced Weld Defect Detection Through Advanced Image Preprocessing Techniques

Weld defect detection is vital for ensuring structural integrity, but challenges like noise and low contrast in weld images often hinder accuracy. The traditional manual inspection methods usually have difficulty in maintaining consistency and accuracy. Deep learning has recently been the subject of...

Full description

Saved in:
Bibliographic Details
Published in2025 International Conference on Electronics, Computing, Communication and Control Technology (ICECCC) pp. 1 - 6
Main Authors Gaidhane, Vaishali, Kolhe, Sujata
Format Conference Proceeding
LanguageEnglish
Published IEEE 01.05.2025
Subjects
Accuracy
CLAHE
deep learning
defect
Defect detection
Image edge detection
Image preprocessing
Image quality
Inspection
MSE
NLM
Noise
Pipelines
Radiography
SSIM
weld
Welding
Online AccessGet full text
DOI10.1109/ICECCC65144.2025.11063979

Cover

More Information
Summary:Weld defect detection is vital for ensuring structural integrity, but challenges like noise and low contrast in weld images often hinder accuracy. The traditional manual inspection methods usually have difficulty in maintaining consistency and accuracy. Deep learning has recently been the subject of ongoing efforts to improve fault detection at various corporate processing sites. During radiographic non-destructive inspection, pre-processing should be given higher priority in order to automatically detect welding defects using deep learning. This study presents an advanced image preprocessing framework for better detection of defects on weld images to build models for machine learning. For the preprocessing of weld images, we employed a combination of advanced methods, including Contrast Limited Adaptive Histogram Equalization (CLAHE) for contrast enhancement, Non-Local Means Filtering for noise reduction, Canny Edge Detection for edge preservation, Adaptive Thresholding for precise segmentation, and Morphological Operations to refine image structures. These methods collectively improve image quality and enhance defect visibility. We further show that these methods indeed improve image quality and defect visibility by comparing low MSEs, high SSIM scores. This pipeline for pre-processing seems really promising in the case of automating defect recognition and to increase efficiency of inspection processes.
DOI:10.1109/ICECCC65144.2025.11063979