Research On The Method Of Extracting Sawtimber Surface Defect Feature Based On LVQ Neural Network

Wood defect is not only an important index which assesses timber grading, but also the main factor of affecting the wood quality and reducing the saw timber commodity value. To realize the rapidly non-destructively and on-line way for wood surface defects testing is an effective approach to increase the lumber commodity value and speed up timber processing automation.Based on the analysis of analyzes the status quo of non-destructive testing techniques home and abroad, this paper trying to combine digital image processing technology with LVQ neural network pattern recognition theory, and applying them to the wood surface defect detection process.(1) This paper focuses on the wood surface defects identify recognition algorithm of the control recognition system. By analyzing the impact of the wood quality from the different types of wood surface defects, two common defects, that are round knot and huge hole, were selected to be identified.(2) The original wood defect images were processed by the following steps: gray level transformation, median filter and threshold of segmentation, which relying on the image processing technology. Seeking the best method of wood defect digital image pretreatment through theoretical analysis and experimental contrast.(3) This paper described the principle of wood defect feature extraction and analyzed the characteristic quantity which is common used to be extracted and the factors of wood defects formation, thus grayscale mean, grayscale variance and defect outline character points were chosen as wood defect pattern characteristics to be the multidimensional input of the LVQ neural network.(4) Based on the algorithms of wood defects feature extraction, the software of wood defects feature extraction system was developed. For ensuring the expansibility of the source code, each algorithm was designed by modular programming.(5) According to numerous specific wood defect images which were collected and analyzed a wood defect LVQ neural network pattern recognition system was established by carrying out Matlab engine calling the functions provided by Matlab neural network toolbox in VC++6.0 development environment. The feasibility of the system could be proved from three areas which included the recognition accuracy and minimum resolution.