Study On Color Classification Of Bamboo Pieces By Computer Vision

In the bamboo products such as bed-mat, bamboo pieces is a basic component. After bamboo is cut into pieces, color classification is a necessary process which ensures the color consistency of the final product. Traditionally color classification is assessed through visual inspection by human inspectors. This will lead to non-consistency in color classification and increase labor cost. In order to recognize and classify bamboo pieces automatically, the paper studies the color feature of the bamboo pieces, based on which the color classification method is discussed theoretically and verified through experiments.The main research work and results:(1) A computer vision system was established to detect the bamboo pieces surface feature and classify the color of the bamboo pieces, on which the sample images were acquired.(2) Algorithms of image processing are studied. Comparing the background segmentation results of sample images in R, G, B channels by Otsu and iteration Method, Otsu method to segment R channel image is selected to segment the background. After boundary tracing and region marking, every single bamboo pieces are extracted. Then Tilt correction based on radon transformation and the segmentation of the upper surface of bamboo piece region are carried out before the effective region for color classification is extracted.(3) After analyzing the color of bamboo pieces, the color feature parameters are calculated. Based on color histogram in HIS color space, the frequencies of different Hue ranges the pixels belong to are counted firstly. And the the means and standard deviation of the pixels in the effective region for color classification are calculated for each component of the HIS space separately. Based on Blemish segmentation, the feature of blemish class was calculated.By feature selection in color classification experiment, eight feature are selected as the color features for classification, which are the means and standard deviation of the pixels in the effective region for color classification for each component of the HIS space,μH,μS,μI,σH,σS,σI, and two feature Based on Blemish segmentation, Q1 and Q2.(4) BP Neural Network was selected to classify the bamboo pieces on the basis of their color. Given the shortcoming of traditional BP network, the Levenberg-Marquardt optimization method are chosen as network training Algorithm after comparing with other method in the experiment. When the neural number of the hidden layer is eight in the experiment, the convergence rate and the classification correctness of the BP network reaches the highest point at the same time. The average classification correctness for different classed is 94.5% and the average time for one bamboo pieces classification is 0.1914S.The research provides essential theory foundation and technology support for classify bamboo pieces based on their color automatically. The methods discussed in the paper are significant and have great practical value for strengthening bamboo product's competition in the international markets and improving the development of export trade.