Research On Judgment Of Solid Wood Board Texture Based On Feature Fusion

With the development of the construction industry and the furniture industry, people put forward higher and higher requirements on the macro aesthetics and quality of the wood. Wood texture is an important factor to reflect the macro aesthetics and quality of the wood. In the timber industry, there are no national standards and industry guidelines which can describe the wood surface texture perfectly and the texture classification of wood plate has slower speed, low precision features. This study proposed a detection method, which was suitable to distinguish the texture of the wood. This method, fusing the six Tamura texture parameters from the perspective of visual psychology and the basic statistical characteristics, established parameters of texture characteristics of wood and laid a foundation for the automatic classification of wood surface texture.The idea of this study:first of all, Sample images of the oak wood were collected, split, and zoom. Secondly, wood surface texture could be divided according to the ruled, parabola and random pattern texture three textures. Thirdly, the parameters of the three textures were extracted and fused. Training samples were selected. At last, classifier was designed.Pretreatment:the size of Picture which was collected, divided, was512×512. To reduce the Interference of color, color texture images were transformed into grayscale texture images which have256levels. To increase the rate, the gray-scale image of wood was resized to a smaller one.Extracting parameters and parameters fusion:in this study. To overcome the shortcoming, which was that statistical methods were out of touch with visual mental model and could not use all information, and to increase the rate, the gray-scale image of wood texture was resized to smaller one, six Tamura texture parameters, from the perspective of visual psychology, extracted. This research extracted three basic statistics which could use the global information in the original gray-scale image quickly. Nine wood texture parameters of wood texture were fused into seven wood texture parameters by PCA which could eliminate the redundancy of data.Training samples selection:training samples of wood texture would affect the classification results of wood texture, bad samples should be excised. This study established a mapping, which used NLM based on Genetic Algorithms, from the nine parameters of wood texture to two parameters of wood texture. Good samples were selected by analyzing the two parameters in two-dimensional space.SVM was selected as the classifier. The seven parameters were input. The ruled, parabola and random pattern texture of wood surface textures were output. To get a better result of classification, penalty parameter of the SVM and the parameter of the RBF kernel function were optimized by PSO. Wood textures are classified by using the SVM with Parameters optimized and the SVM with Parameters which were not optimized. Comparing the results, the SVM with Parameters optimized had a better effect in the classification of wood textures.The simulation results showed that the method in this paper could reduce the time of exacting and fusing oak board texture features. The accuracy of classification of oak texture was91.43%by using the PSO-SVC. The accuracy of classification is88.1%by using the C-SVM. The accuracy is significantly improved. This method has the reference value for the development of wood processing automatic detection system.