Research On Bamboo Classification Method Based On Color And Texture Features

Bamboo texture smoothness and clear, is widely used for the manufacture of architectural and decorative products like bamboo tapestries, bamboo carpet, bamboo floor etc. The color shade of bamboo raw material is different after dealing with the same carbonization process, by place of origin, moisture content, density and other factors, but the color consistency is an important indicator to measure the quality of bamboo products. Therefore, bamboo raw materials must be graded by color, in order to ensure the quality of bamboo products. At present, it mainly depends on manual methods for color grading, which is labor intensive, inefficient and has greatly restricted the rapid development of bamboo products industry. This paper selects carbonized bamboo as the objects for study, researches the bamboo surface color, texture feature and classification method by using machine vision technology, digital image processing technique and pattern recognition theory, which will provides theoretical basis and technical base for the realization of bamboo automatic classification.The main content and conclusions of this paper were as follows:(1) The machine vision system applicable for bamboo classification was establishded. Images of eight levels bamboo samples were collected by using the machine vision system, and the standard sample library which contains 800 samples was establishded.(2) The bamboo original images were pretreated, such as image graying, median filtering, Otsu threshold, image Synthesis, by using digital image processing technique, and bamboo color images without background were obtained.(3) In the color model of HIS, bamboo surface color and texture features were described by using Color Histogram, Color Moment and Gray level Co-occurrence Matrix (GLCM) method respectively. Various building factors impact on GLCM and its parameters were research and analyzed by using the dissociable basis, and the way of building GLCM suitable to describe bamboo surface texture was established:the making step d equaled to 1, the gray-level of image g equaled to 16, the making directionθwas took as 0°,45°,90°,135°. The GLCM texture parameters value took the average of four directions to eliminate the impact of direction. Based on the foregoing research,9 color feature parameters of Color Moment and 6 texture feature parameters of GLCM were obtained, and a bamboo characteristic parameters system named feature parameters system I was founded. Another characteristic parameters system named feature parameters systemⅡwas founded after 15 characteristics of feature parameters systemⅠwere analyzed with principal component analysis of SPSS 16.0(4) In feature parameters systemⅠandⅡ, Bamboo samples were graded recognition by using BP neural network which is based on Levenberg-Marquardt (LM) algorithm. Feature parameters systemⅠwhich is more capable of identifying bamboo samples,was selected as optimal feature parameter system. Due to the shortage of BP neural network for the instability, easy to trap into local minimum point, GA-LMBP algorithm was proposed. In the proposed model, the genetic algorithm was firstly used to optimize the initial weights and thresholds of BP neural network, and then the LM algorithm was used in a small solution space located by the genetic algorithm to train the BP neural network. Test results showed that the genetic algorithm can not only overcome the shortage of BP neural network, but also can improve the training accuracy and generalization capacity of BP neural network, the average recognition accuracy of 8 levels bamboo samples reached 96.88%.(5) In feature parameters systemⅠ, Bamboo samples were graded recognition by using Least Squares Support Vector Machine (LS-SVM), and its recognition results were compared with the recognition results of BP neural network and GA-LABP neural network, the comparison results showed that LS-SVM was fastest and the most effective, the average recognition accuracy of 8 levels bamboo samples reached 97.5%.(6) A software system based on MATLAB7.0 for bamboo classification was developed. The functions of the software system included image pretreatment, feature extraction, bamboo classification and identification etc. Test results showed that the software system was operational simple and stable operation.