Matematiske realfag og teknologi
IMTs forskere vant pris for beste poster
Ingunn Burud
Med posteren 'Classification of Frost Damage on Brick Walls using Multivariate Image Analysis' gikk Knut Kvaal, Andreas Flø, Ole Mathis Kruse og Cecilia Futsæther til topps på 12th Scandinavian Symposium on Chemometrics.
Gratulerer så mye til Knut, Andreas, Ole Mathis og Cecilia !
Her kan du se posteren
Her kan du lese om konferansen.
Knut Kvaal mottar prisen for beste poster.
Foto: Per Waaben Hansen
Abstract
Frost damage often occurs on bricks of buildings in northern climates due to repeated freeze-thawing cycles. The frost-damaged bricks often display cracks, fractures and chipping at various size scales. Frost damage can affect the structural integrity as well as diminish the aesthetics of the wall. RGB images of brick samples of an apartment building in Oslo, Norway were captured using a high resolution digital SLR camera. Each brick was extracted separately from the wall giving an image stack of individual bricks. AMT (Angle Measure Technique) and related techniques were used as pre-processing algorithms to transform the images from the 2-D domain into the unfolded 1-D domain, in principal, by calculating angles and distances for a set of given points of the image. The resulting feature vectors (AMT) were used for further analysis. Likewise, a GLCM (Gray Level Co-occurrence Matrix) was calculated for each image in the stack from which Haralick’s texture features for each image were computed. Principal component analysis (PCA) was used to search for similarities and differences between frost damaged and intact bricks based on either the AMT or the GLCM texture features. PLS-DA was used to classify bricks into two groups: (1) damaged or (2) intact bricks. PCA showed that damaged and intact bricks separated into two clusters which could easily be identified in the score plots. The PLS-DA models could classify a brick as damaged or intact to a precision of 90%. Thus, the results indicate that it should be possible to develop automated techniques for identifying intact and damaged bricks in a brick wall.
Oppdatert: 16.06.11
Utskriftsvennlig versjon
Del med en venn: