Automated Steel Bridge Coating Rust Defect Recognition Method Based on U-Net Fully Convolutional Networks

Abstract

Nowadays, bridges are significant infrastructure in most countries, and it is crucial to come up with an effective corrosion detection method for steel bridge inspection. A crucial issue on rust recognition is to distinguish real rust corrosion spots and areas. A fully convolutional neural network, namely U-Net, is explored to develop an image semantic segmentation model, which provides a wide range of rust image recognition.