Spalling detection via deep transfer network

A spall is defined as flakes materials that are broken off of a large solid body. Spalling is a significant surface hazard that can undermine the durability of concrete structure and thus may affect structural safety of the building and sometimes cause serious construction problems with injuries and...

Full description

Saved in:
Bibliographic Details
Main Author: Wu, Ruida
Other Authors: Xie Lihua
Format: Final Year Project
Language:English
Published: 2019
Subjects:
Online Access:http://hdl.handle.net/10356/78138
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Nanyang Technological University
Language: English
Description
Summary:A spall is defined as flakes materials that are broken off of a large solid body. Spalling is a significant surface hazard that can undermine the durability of concrete structure and thus may affect structural safety of the building and sometimes cause serious construction problems with injuries and loss of properties. A number of image processing techniques (IPTs)[1] have been implemented for detecting civil infrastructure defects. These IPTs are able to extract flaw parts such as cracks from images, but face challenges while analyzing images under various environmental factors (e.g. lightening, shadow). In order to get a better performance in detecting civil infrastructure problem, a method based on deep learning for spalling detection could be developed. Due to the fast development of artificial intelligence and machine learning technologies, especially in applications of deep learning in computer vision, structural damage recognition could rely on such knowledge because such problem is indeed a repetitive work which requires learning knowledge from past experience. In this project, we will apply technologies related to knowledge transfer to develop a model that does spall recognition on images taken by Unmanned Aerial Vehicle (UAV) with high accuracy. In order to improve the model, we will do transfer learning, data assemble and knowledge distillation to this model so that it is portable as well as robust enough.