Research On Wood Surface Defect Detection Based On Fractal Theory

Disfigurement detection of the wood surface is a multi-disciplinary and intersectional technology,which has upper applied values In the manufacture field and many spheres of farther processing. The paper researched deeply disfigurement detection of the wood surface around the fractal theory. And the paper researched texture segmentation, feature extraction and Pattern recognition of wood surface disfigurement with combining fractal parameter, analysis of the wavelet multi- distinguishability with artificial neural network.Texture segmentation is the most important issue for image processing. The paper put forward segmentation method of wood surface disfigurement based on fractal parameter and the adjunction of spring and mass of matrix directing at disadvantages of segmentation method of traditional surface disfigurement and calculation methods of fractal parameter.Extraction of characteristic quantity will affect straightway the distinguishability ratio of Wood disfigurement detection system.The paper carried on two-dimension wavelet Twice Decomposition to recognise Wood disfigurement image, Calculated autocorrelation fractal dimension for eight widths images of a single-branch reconstruction,and constituted characteristic quantity of pattern recognition with autocorrelation fractal dimension of undecomposed disfigurement image corporately. Results showed that fractal dimensions of different disfigurement categories have better distributivity aiming at the mean values and mean square values of nine characteristic quantities,and it was feasible to use feature parameter to describe wood surface disfigurement.The character values of the wood disfigurement styles distinguishability based on connecting the wavelet multi- distinguishability with autocorrelation fractal dimention, uses four kinds disfigurements of knag , bark pocket , wormhole and molder as output of distinguishability typtles,uses L-M Calculation methods to train BP neural network. After having trained the network, we analyzed recognition ability to get linear Regression curved line, the related coefficients of target export exceeded 0.9.When we adopted 200 samples to test, the distinguishability ratio reached 95.6 percents. Results showed that it can improve recognition precision of Wood surface disfigurement detection to use fractal theory to carry on disfigurement segmentation and characteristic extraction.