Deep Architecture Based Spalling Severity Detection System Using Encoder-Decoder Networks

Abstract

Proper maintenance of concrete structures is a significant issue to avoid any hazardous situation in civil infrastructure. Spalling is a significant surface concrete distress in bridges and buildings. Correctly detecting the severity level of spalling can make it happen to detect and maintain the harmful spalling promptly to avoid any accidents [10]. While previous works have been on surface defects, like cracks and spallings, few have addressed spalling severity detection. In this paper, we have proposed a deep learning-based approach to detect the exact location of spalling according to severity level by using pixel-by-pixel classification. Our network labels each pixel as no-spalling, small, or large spalling. To get the optimal proposed deep architecture, we tested several encoder-decoder networks to compare and analyze the performance of the detection processes.