Hyperspectral Image Analysis for Mechanical and Chemical Properties of Concrete and Steel Surfaces (SN-5)

A typical human eye will respond to wavelengths from approximately 400 to 700 nm. A hyperspectral camera can extend the wavelength to as high as 2500 nm. This extension will allow engineers to find objects, identify materials, and detect processes on structural surface, which cannot be done with visual inspection. Hyperspectral images have been used in many fields (e.g. agriculture, geosciences, physics, surveillance, etc.). Their application in engineering structures is just beginning. A remote sensing method was applied to assess the degradation (carbonation, chlorides, and sulfates) of concrete and evaluate the strength of high performance concrete in situ. A support vector machine classifier was also explored for potential applications in paint condition (corrosion) assessment of steel structures. To date, all classifiers used to process large hyperspectral image data sets from satellite or remote sensing belong to semi-supervised learning. In bridge inspection, however, data are often collected from line elements such as girders and columns. As such, supervised learning may be practical, giving a more accurate assessment of condition. Hyperspectral imaging obtains an electromagnetic spectrum for each pixel in the image of a structural surface and processes information from across the spectrum. The spectral signatures, uniquely associated with certain objects, can be used to identify the materials that make up a scanned object. As an example, the characteristic wavelength at absorption peak can be used as a spectral feature for certain chemicals generated during corrosion process. Image analysis will be done in two phases: learning and classification. In the learning phase, training images are taken, labelled with a rating (the state of paint in steel members) or a value (degradation depth or chloride concentration), extracted for spectral signatures, and used to train a classifier (e.g. support vector machine) that relates the labelled parameter to the signatures. In the classification phase, hyperspectral images are taken, extracted for spectral features, and classified into a rating or a characteristic value using the classifier trained in the learning phase. This project aims to develop an open-source catalogue of concrete and steel surfaces and their spectral/spatial features (discoloration, characteristic wavelength, roughness, texture, shape, etc.), extract spatial/spectral features of hyperspectral images, and develop/train a multi-class classification or regression classifier through machine learnings (supervised and/or semi-supervised), and validate the classifier as a decision-making tool for the assessment of concrete crack and degradation processes, in-situ concrete properties, and corrosion process in steel bridges. Scope of Work in Year 1: (1) Prepare and test a set of RC specimens under freeze-thaw and corrosion conditions to characterize various aging and electrochemical effects on concrete surfaces, (2) Take and categorize hyperspectral images of the specimens corresponding to various concrete conditions (crack, degradation, discoloration, etc.), (3) Extract characteristic features from the images, and (4) Correlate the concrete conditions with spatial and spectral features of the images. Scope of Work in Year 2: (1) Understand the effect of pixel size on detection precision and accuracy associated with the focal distance of a co-aligned VNIR and SWIR camera as well as the effect of line light source (intensity and incident angle) on detection sensitivity using known objects of various size, (2) Conduct carbonation tests for various exposure periods (thus carbonation depths) and characterize hyperspectral image changes, and (3) Correlate the hyperspectral features with the carbonation depth of the tested specimens. Scope of Work in Year 3: (1) Prepare in pair the image features and structural conditions of various types of RC specimens for machine learning, (2) Develop and validate a supervised learning algorithm trained with the input-output data pairs for the prediction of steel bar corrosion, and (3) Investigate the dependence of the algorithm to the completeness of the training data set for steel bar corrosion in terms of causative relations between the images and their corresponding surface conditions.


  • English


  • Status: Completed
  • Funding: $108,821
  • Contract Numbers:


  • Sponsor Organizations:

    Office of the Assistant Secretary for Research and Technology

    University Transportation Centers Program
    Department of Transportation
    Washington, DC  United States  20590
  • Managing Organizations:

    Inspecting and Preserving Infrastructure through Robotic Exploration University Transportation Center

    Missouri University of Science and Technology
    Rolla, MO  United States  65409
  • Performing Organizations:

    Missouri University of Science & Technology, Rolla

    Department of Engineering
    202 University Center
    Rolla, MO    65409
  • Principal Investigators:

    Chen, Genda

  • Start Date: 20170301
  • Expected Completion Date: 20230630
  • Actual Completion Date: 0
  • USDOT Program: University Transportation Centers

Subject/Index Terms

Filing Info

  • Accession Number: 01646011
  • Record Type: Research project
  • Source Agency: Inspecting and Preserving Infrastructure through Robotic Exploration University Transportation Center
  • Contract Numbers: 69A3551747126
  • Files: UTC, RIP
  • Created Date: Sep 15 2017 8:38AM