| Summary: | 碩士 === 國立金門大學 === 土木與工程管理學系碩士班 === 100 === Most important civil infrastructures are made of concrete, so accurate information by
routine inspection is necessary for structure maintenance. Currently most infrastructure
inspections are implemented by inspectors. However, sometimes manual inspection would be
inefficient and unsafe while the inspections of skyscraper or substructure of bridge are
executed. In last decade, image-based techniques were applied to crack detection and
measurement for concrete structures, such as principal component analysis, co-occurrence
matrix, wavelet analysis, and statistical texture. However, the present detection accuracy
between 68.7 and 76.5% is unsatisfied for practice applications. This research proposes a
morphology-based image processing technique to attempt to automatically detect cracks in
concrete structures. The morphology-based image processing technique consisting of
weighted median filter, image opening operation, and image segmentation was used to
transform the grey images of concrete structure into the binary images. To segment the image
regions of complete crack from a noisy environment, two critical morphological features
including area and eccentricity were measured for each segmented image region. Then, a
sensitivity analysis based on the measured morphological features was applied to determine
the appropriate criteria for crack detection. In this thesis, 100 images of the concrete roads and
50 ones of the concrete buildings were acquired to be the training and testing samples,
respectively. The experimental result indicates that the optimal training and testing accuracies
are 90% and 76%, respectively.
|
|---|
