A self organizing map optimization based image recognition and processing model for bridge crack inspection
Introduction
In 2009, Typhoon Morakot wrought catastrophic damage on Taiwan, leaving 461 people dead and 192 others missing, with a cost of roughly 110 billion New Taiwan dollars (NTD), which is close to 3.3 billion United States dollars (USD) in damages. The extreme amount of rain triggered enormous mudslides and flooding throughout southern Taiwan. Typhoon Morakot not only tested how the Taiwan government could relieve the victims of a severe disaster, but also drew attention to the need to improve the safety of infrastructure to reduce the impact of disasters. There are over twenty thousand bridges located across Taiwan. As bridges have an important role in facilitating transportation, damage to bridges by disasters not only threatens the safety of users, but can also disrupt traffic flows and cause residents to be locked in place.
Cracking in concrete bridges is an inevitable problem resulting from natural processes and can invite spectacular failure of the entire bridge. Cracks not only provide access to harmful and corrosive chemicals inside concrete, but also allow water and deicing salts to penetrate through bridge decks, which can damage superstructures and bridge esthetics. Routine inspections are widely adopted and are carried out manually by certified bridge inspectors every two years, as stipulated in the National Bridge Inspection Standard by Federal Highway Administration (FHWA) in USA. Inspection results are mainly based on the inspectors' observations and visual assessment [1], [2]. However, such bridge detection methods have several limitations. The inspection process is laborious, time-consuming, and influenced by the subjective behavior of individual inspectors. The visual inspection only provides qualitative information on defects. Moreover, finding experienced bridge inspectors poses a challenge for the construction industry, which is now facing a pressing shortage of experienced and highly trained inspection personnel [2], [3].
In order to overcome these issues, considerable research has been conducted in an effort to develop automated bridge crack inspection tools to reduce the field work required for inspectors [4]. For example, Oh et al. suggest certain image processing algorithms for detecting and tracing cracks combined with the use of a robot mechanism [5]. Zhu et al. propose an automated bridge condition assessment system with a focus on detecting large-scale bridge concrete columns [6]. Yu et al. designed a robot which can detect fissures underneath a bridge. They provided a safe and effective machine vision technology to detect the bridge [7]. Bu et al. developed an automatic bridge inspection approach by employing Support Vector Machines to classify cracks based on wavelet-based image features. The researchers tested 50 different image samples, and both ‘complex’ and ‘normal’ images were considered. The resulting recognition accuracy rates of the crack ranged from 74% to 93.26%, varying according to the different image types, training set types, and the feature extraction methods used [8]. Li et al. also put forward a method consisting of crack extraction, an electronic distance measurement algorithm, and an image segmentation algorithm to detect cracks [3]. Further increasing the reliability and accuracy of the results remains an ongoing effort in this research area. The algorithm and the restricted conditions employed in this study contribute to this effort to improve crack detection. The objective of this study is to develop an automatic bridge crack detection method based on self-organizing map optimization through image processing technology. Several stages of image recognition for bridge cracks are utilized to process 216 images of randomly selected bridges located in northern Taiwan. The model presented here can improve the accuracy of bridge crack inspections and reduce the cost of labor.
Section snippets
Bridge crack inspection
Degradation often occurs during the final stage of reinforcement concrete structures' life cycle. The various degrees of maintenance and service conditions of structures in different natural environments derive from disparate degradation rates and consequences. In Taiwan, bridges very easily deteriorate due to high humidity, frequent earthquake loading, and overloading by heavy vehicles [9]. The acceleration of degradation in reinforced concrete structures can be attributed to natural factors
Data collection
This research employs a digital camera to capture concrete bridge cracks in order to develop an image recognition program and process image identification. Before shooting, the choice of the bridges included in the sample for the present study was based on the DERU visual inspection results. Ten concrete bridges in northern Taiwan are selected, and the crack images were randomly chosen from the artificial field shooting database. According to the manual of highway maintenance which was
Development of image recognition and processing model
The proposed image recognition and processing model was developed on the basis of the concepts of self-organizing feature map optimization (SOMO), fuzzy logic control, and hyper-rectangular composite neural networks (HRCNNs). The SOMO was developed by Su et al. in 2004 [13] and has been applied to several areas such as construction sequencing for building renovation and secant pile walls [14], [15]. The model development in this study starts with the creation of HRCNN integrated with fuzzy
Imaging processing
The image-processing steps performed in this study are as follows. First, apply image grayscaling to process the original color image. Second, use the high-pass filter to remove the low-frequency noise in the images to highlight the characteristics of a crack. Third, separate the subject and background through the binary process and eliminate unnecessary noise with labeling. Finally, employ MATLAB to develop a Local Directional Pattern (LDP) algorithm to capture crack features and mark the
Case implementation
In order to verify that the recognition program can be used for practical bridge inspection, the case study executed in the present research is the severely damaged Hsichou Bridge (shown in Fig. 4.) selected from the Taiwan Bridge Management System. Manual shooting was conducted in the field and then qualified images were selected to perform the recognition test. An example of an image analyzed for this case study is shown in Fig. 5.
50 pictures of the cracks on the Hsichou Bridge were taken for
Conclusions
Though visual inspection of bridges in Taiwan is relatively economical, this laborious and time-consuming method is easily influenced by the subjective behavior of individual inspectors. The automatic image recognition technique developed by this paper is aimed at effectively and efficiently identifying the cracks in concrete bridges using machine vision, rather than traditional human judgment, substituting subjective efforts with objective testing results.
This study selected appropriate
References (24)
- et al.
Long-distance precision inspection method for bridge cracks with image processing
Autom. Constr.
(2014) - et al.
Bridge inspection robot system with machine vision
Autom. Constr.
(2009) - et al.
Detection of large-scale concrete columns for automated bridge inspection
Autom. Constr.
(2010) - et al.
Image-based retrieval of concrete crack properties for bridge inspection
Autom. Constr.
(2014) - et al.
Comparison of SOM-based optimization and particle swarm optimization for minimizing the construction time of a secant pile wall
Autom. Constr.
(2009) A hybrid knowledge-sharing model for corporate foreign investment in China's construction market
Expert Syst. Appl.
(2012)Specification for the National Bridge Inventory Bridge Elements
(January 2014)Long-term Bridge Performance Primer, Rep. No. FHWA-HRT-13-051
(December 2013)- et al.
Vision-based automated crack detection for bridge inspection
Comput.-Aided Civ. Infrastruct. Eng.
(2015) - et al.
Study on a new bridge crack detection robot based on machine vision
Crack detection using a texture analysis-based technique for visual bridge inspection, special issue
Electron. J. Struct. Eng.
A planning multi-level bridge management system for Taiwan
Cited by (72)
MultiScaleCrackNet: A parallel multiscale deep CNN architecture for concrete crack classification
2024, Expert Systems with ApplicationsTrueDeep: A systematic approach of crack detection with less data[Formula presented]
2024, Expert Systems with ApplicationsMOD-YOLO: Rethinking the YOLO architecture at the level of feature information and applying it to crack detection
2024, Expert Systems with ApplicationsCrack recognition on concrete structures based on machine crafted and hand crafted features
2023, Expert Systems with ApplicationsSmart bridge maintenance using cluster merging algorithm based on self-organizing map optimization
2023, Automation in ConstructionPhysical interpretation of machine learning-based recognition of defects for the risk management of existing bridge heritage
2023, Engineering Failure Analysis