Computer vision-based crack-like object detection has many useful applications, such as inspecting/monitoring pavement surface, underground pipeline, bridge cracks, railway tracks etc. However, in most contexts, cracks appear as thin, irregular long-narrow objects, and often are buried in complex, textured background with high diversity which make the crack detection very challenging. During the past a few years, deep learning technique has achieved great success and has been utilized for solving a variety of object detection problems.
This book discusses crack-like object detection problem comprehensively. It starts by discussing traditional image processing approaches for solving this problem, and then introduces deep learning-based methods. It provides a detailed review of object detection problems and focuses on the most challenging problem, crack-like object detection, to dig deep into the Deep Learning method. It includes examples of real-world problems, which are easy to understand and could be a good tutorial for introducing Computer Vision and Machine Learning.
Many books discuss Computer Vision and Machine Learning from the aspect of theory, algorithm and its applications with some simple examples that are far from practical engineering. The author worked on Computer Vision, pattern recognition, and image processing research over 10 years and the group leader,
who is also the second author of the book, focused on making an industrial pavement surface inspection product with computer vision technology during the past 30 years. However, until 2015, there were still some tough problems in pavement crack detection that were not well solved using traditional image processing approach due to the complexity and diversity of different pavement surface conditions. Since 2015, the group focused on addressing the issue with deep learning, and soon in 2018, great progress was made and an efficient method was invented and tested on large-scale data. It is applicable to industry.
This book discusses Deep Learning and its applications based on a practical engineering problem: crack-like object detection. The advantage is that we conducted many tests and trials in practice and obtained many valuable engineering experiences, which cannot be found in a regular text-book. We selected five classic problems in crack detection that cover the knowledge necessary for a beginner to quickly become familiar with Deep Learning and how it is used in Computer Vision. The main research topics include image classification, transfer learning, weakly supervised learning, generative adversarial networks, fully convolutional network, domain adaptation, deep edge computing, etc.