Mobile hyperspectral imaging for structural damage detection
Metadata[+] Show full item record
Numerous optical-imaging and machine-vision based inspection methods are found that aim to replace visual and human-based inspection with an automated or a highly efficient procedure. However, these machine-vision systems have not been entirely endorsed by civil engineers towards deploying these techniques in practice, partially due to their poor performance in object detection when structural cracks coexist with other complex scenes. A mobile hyperspectral imaging system is developed in this work, which captures hundreds of spectral reflectance values at a pixel in the visible and near-infrared (VNIR) portion of the electromagnetic spectrum bands. To prove its potential in discriminating complex objects, a machine learning methodology is developed with classification models that are characterized by four different feature extraction processes. Experimental validation with quantitative measures proves that hyperspectral pixels, when used conjunctly with dimensionality reduction, possess outstanding potential in recognizing eight different structural surface objects including cracks for concrete and asphalt surfaces, and outperform the gray-values that characterize the texture/shape of the objects. The authors envision the advent of computational hyperspectral imaging for automating structural damage inspection, especially when dealing with complex structural scenes in practice.
Table of Contents
Introduction -- Hyperspectral Image -- Preprocessing -- Methodology -- Machine Learning Approach -- Discussion -- Appendix 1-2
M.S. (Master of Science)