An On-Line Wood Surface Defect Detection And Classification System Based On Machine Vision

China is the major country of wood producing and processing,which is still laborintensive industry now.Besides of huge requirement of human labor,low accuracy and efficiency are also the problems,which hinder the increasing of output.With the development of computer vision and pattern recognition,it is the necessary access that applying them to wood producing and processing,which can further deepen the development of artificial intelligence.This thesis compares different color space model,analyzes the methods of different color feature extraction comparatively,presents the mask extraction method of the main colors of the space-based Lab,which accurately extracts color feature of the dominant color region;it describes the principle of GLCM and Gabor transform to achieve the two main texture features,and shows the effectiveness of Gabor filter template;it gives the algorithm testing process of different kinds of defects on wood surface,the multi-scale adaptive thresholding method is proposed in the scarring process of defect detection.To solve the problem that classification results are poor when the classifier is trained by global color feature with only one sample for each type,the thesis presents the mask extraction method of the main colors of the space-based Lab,which can help to effectively extract the color feature without parts of textures and defects.To solve the problem that GLCM is insensitive except the four directions of 0°?45°?90°?135°,Experiments show that Gabor filter templates of 5 scale 8 directions obtained by Gabor transform can accurately obtain the main direction of the wood grain.To meet the high efficiency of the production line,the paper proposes four fixed template instead of Gabor templates,the effect is significant.To solve the problem that traditional thresholding method does not take into account the detection of scarring defects with different shapes and different colors,the thesis presents a multi-scale adaptive thresholding method in defect detection of scarring,which can both detect the defect sizes by changing the window size dynamically.In practice,the classification results of different wood samples during different periods show that the classification error rate of color and texture is less than 0.01,and this system can accurately detect common pitfalls of timber.Thus,the timber line defect detection and classification system has a high value.The contributions of this thesis can be summarized as follows:? The thesis presents the mask extraction method of the main colors of the space-based Lab,which can get great classification results even if when the classifier is trained by global color feature with only one sample for each type.? The thesis optimizes the extraction method of main direction of the wood grain feature,which can meet the high efficiency production line and texture angle detection accuracy.? The thesis presents a multi-scale adaptive thresholding method in defect detection of scarring,which can both detect the defect sizes by changing the window size dynamically and Minimize undetected cases.