Automated recognition of surface defects using digital color image processing
Introduction
Deteriorating infrastructure systems, insufficient budget amounts, and increasing public demand for safer transportation systems have motivated federal/state agencies to find more objective and computerized infrastructure inspection methods [1]. Conventional inspection methods consist of visual inspections by inspectors. While a careful visual inspection can provide valuable information, it has inherent limitations in identifying the structural integrity and assessing the level of damage of a facility. Inspectors cannot access every part and space of infrastructure. Therefore, many problems can be neglected until they become serious and require costly repair.
One of the computerized technologies for advanced infrastructure inspection methods is the application of digital image processing. Digital image processing methods have been developed for steel bridge coating inspections for the past few years. The rust percentages on steel bridge coating surfaces can be reliably computed through the use of digital image processing methods. However, most researches solely focused on the calculation of the degree of rust defects on the steel surfaces in percentage. Therefore, an automated processor that can recognize the existence of bridge painting rust defects needs to be developed. Once this processor is created, it can be combined with the processor for calculating the degree of rust defects. In other words, the integrated system for coating surface condition assessment can be developed with two image processors equipped: the pre-processor and the main processor. The pre-processor is required to check whether a given digital image contains defective parts. If the image is judged as defective, the image is sent to the next stage, the main processor, to make an accurate defect assessment.
In the civil engineering domain, not much research was performed for defect recognition using digital images. Many examples can be found in pavement areas [2], [3], [4]. Those previous research studies focused on the recognition of pavement cracks and the classification of crack types applying the projection method. This method projects the intensity values of a grayscale image horizontally and vertically. Fig. 1 shows an example of the projection method using a 5 by 5 pixel digital image. The numbers in each pixel indicate the light intensity values where 0 means white and 255 means black. Two pixels with darker areas are defective areas, showing high intensity values. The horizontal projection map can be generated by averaging the sum of each row pixel values. Likewise, the vertical projection map can be obtained by averaging the sum of each column values. The existence of defective areas can be identified by comparing pixel values in the horizontal and vertical projection maps.
This method seems to be effective for the defects whose types have a long linear shape such as pavement cracks. However, bridge painting rust defects are often characterized as small scattered spots. In addition, coating surfaces experience non-uniform illuminations and noises arising from foreign materials. Such factors can cause non-defective images to be classified into defective images from the use of the projection method. Therefore, this article proposes a novel approach to recognize the existence of bridge coating rust defects by processing digital color images for statistical data acquisition and performing a multivariate statistical analysis.
Section snippets
Research methodology
The methodology for the development of a rust defect recognition method, called the RUDR method hereinafter, can be classified into four stages: data preparation, data analysis, statistical modeling, and testing and validation. Fig. 2 shows the whole research steps for the image processor that are sequentially linked together. The detailed description of each stage is given as follows.
In the data preparation stage, steel bridge coating images have to be taken first. For this task, total 20
Image conversion
For the exploration of RUDR method, bridge painting images were prepared and divided into two groups: non-defective and defective. Each group has 30 digital images for data acquisition. The images are expressed as the most fundamental color space, the RGB color space. The color space, expressed as cube, has three primary colors: red, green, and blue as shown in Fig. 4 [6]. Each primary color axis has 256 (28) levels of color shade, which means a total of 224 colors can be generated from the
Data analysis
The collected data are analyzed to choose discriminating variables that are significant for the separation of two or more groups. For the purpose of selecting efficient variables, two kinds of analyses are performed: Wilks' lambda analysis and data range analysis.
Multivariate discriminant functions
Multivariate discriminant functions are a very useful method to separate two or more classes or populations and assign a new observation to one of two or more classes. To model successful discriminant functions, appropriate variables have to be selected on the basis of a number of measurements. The examples applying the discriminant functions can be found in many areas [5]. They include: purchasers of new products and those to purchase slowly, successful or unsuccessful college students,
Model testing and validation
In this stage, the developed RUDR method is validated by using new painting images to test the model efficiency. For the validation process, a validation set is created. The set comprises of 20 coating images: 10 as a non-defective group and 10 as a defective group. The conditions of validation images were confirmed by 8 bridge field inspectors in Indiana out of total 14 bridge field inspectors. For the model validation, each validation image was processed to obtain the values of three major
Application of rust defect recognition method
This section describes the implementation of the RUDR method in practice to classify given digital images into a correct class. The following stepwise application plan contains three steps, from image acquisition to decision-making on bridge coating conditions.
Summary and conclusions
This paper presented a novel approach to recognize the existence of bridge coating rust defects by utilizing the digital color space for statistical data acquisition and performing a multivariate statistical analysis.
The rust defect recognition method was realized by following four stages: data acquisition, data analysis, statistical modeling, testing and validation. In the data acquisition stage, bridge painting digital images were prepared to generate two types of data sets: defective and
References (9)
- et al.
Performance comparison of bridge coating defect recognition methods
Corrosion
(2005) - et al.
ARIA: an operating system of pavement distress diagnosis by image processing
- et al.
Novel approach to pavement cracking detection based on fuzzy set theory
Journal of Computing in Civil Engineering
(1999) - et al.
Position-invariant neural network for digital pavement crack analysis
Computer-Aided Civil and Infrastructure Engineering
(2004)
Cited by (103)
Application of artificial neural networks throughout the entire life cycle of coatings: A comprehensive review
2024, Progress in Organic CoatingsPhysical interpretation of machine learning-based recognition of defects for the risk management of existing bridge heritage
2023, Engineering Failure AnalysisA Review on Recent Advances in Vision-based Defect Recognition towards Industrial Intelligence
2022, Journal of Manufacturing SystemsSingle-shot multibox detector- and building information modeling-based quality inspection model for construction projects
2021, Journal of Building EngineeringAutomated bughole detection and quality performance assessment of concrete using image processing and deep convolutional neural networks
2021, Construction and Building MaterialsScene understanding in construction and buildings using image processing methods: A comprehensive review and a case study
2021, Journal of Building EngineeringCitation Excerpt :The following sections of this review paper analyze the use of color, geometry and feature diagnostics for global reasoning and inferencing. Image processing methods have been developed to capture certain clues in the scene and to elicit useful information from crude imagery data [30,31]. Classic applications of such methods (without machine learning) have been discussed in this paper.