[2I13-17] New Methods and Technologies

[2I14] Image Processing for Grout Loss Detection in Bridge Inspections

*Hengxi Chen¹, Shuichi Hasegawa^1,2, Akio Sugita², Masahiro Abe² (1. The University of Tokyo, 2. Nuclear Professional School, The University of Tokyo)

Keywords:Grout Loss Detection, Machine Learning, Light Intensity Correction, Geant4 Simulation, Noise Reduction

This research presents an innovative method for detecting grout loss in radiographic images of bridges, combining machine learning with sophisticated image processing techniques. Our approach involves noise reduction via transfer learning from a pre-trained Deep learning model, addressing the challenge of limited training data. Image enhancement is achieved through grayscale to RGB conversion and light intensity correction, enhancing perceptual detail and countering point-source X-ray uneven illumination. Using the Geant4 toolkit, we simulate grout loss, training a machine learning model on this data, supplemented with scarce real-world examples. This model can then recognize grout loss instances within images. Testing the model with real-world data and our experiment in Tokai validates its effectiveness, indicating the potential in improving bridge maintenance and safety procedures.