Detection and Classification of Norway Spruce Compression Wood in Reflected Light by Means of Hyperspectral Image Analysis

Abstract

A methodology has been developed based on reflected light to detect compression wood in stem cross sections of Norway spruce (Picea abies [L.] Karst.). In addition to quantify the spatial distribution of compression wood, the chronological pattern of its formation is recorded by cross linking the pixel classification to the tree ring sequence. An imaging spectrometer is used to record the spectral characteristics in the visible light and near infrared of the cross-sectional surface. Cross-sectional areas are classified by hyperspectral image analysis into severe compression wood, moderate compression wood, normal wood, and background/cracks. The classification is performed by the Spectral Angle Mapper algorithm, which compares the standardized spectrum of each pixel with reference spectra stored in a spectral library. The reference spectra are obtained from selected training areas of the different compression wood severity classes identified by cell characteristics under a light microscope. The tree ring boundaries are located in a grey scale image which shows the spatial information at wavelength 435 nm and the annual radial increment is measured. The classification accuracy is tested by a confusion matrix and cross-analysed with High-Frequency Densitometry.