Research On Long Bamboo Batten Surface Defect Detection And Color Classification Based On Machine Vision

With the development of bamboo industry in China,bamboo products are more and more popular for people,but the problems of backward bamboo products processing technology are existed,low production efficiency,unified judgment standard in our country has seriously hindered the bamboo industry market.Therefore,based on the national spark plan project,long bamboo batten surface defect detection and color classification systems and methods are studied,in order to make it possible to approach the long bamboo surface defect detection and color classification.This approach can be of great significance in improving the bamboo earnings and production efficiency,the main content of the study is as follows:(1)Design a long bamboo surface image acquisition scheme based on the characteristics of the shape and length,long bamboo surface image acquisition device is designed.By selecting the appropriate acquisitive equipment,set long bamboo surface defect detection of machine vision system up,collect bamboo library including common bamboo defects and three kinds of color types.Image acquisition of dynamic movement for long bamboo surface is developed.(2)Analysis and design of the types of long bamboo batten surface defect.design various common bamboo defects(including bug,crack,greener,whiteness)detection method and process.Long bamboo surface defective types are analyzed and divided.In addition,bug,crack,greener and whiteness were detected directly.First of all,the preprocessed algorithms of denoising,grayscale,threshold,tilt correction for the original image were designed,then develop the bamboo defects detection algorithms,such as wormhole detection based on morphology and difference,crack detection based on adaptive Canny of double threshold,greener defects detection based on vertical projection combined with edge detection on the big/small bamboo and whiteness based on region growing algorithm.(3)A method was put forward to detect the whiteness based on adaptive region growing block and distinguish white defects and knots by jump characteristics.In this way,rapid detection of bamboo white defects can be achieved,avoiding the influence on knot area between bamboo white defect detection and improving the efficiency.(4)Combined with gray scale,texture,color characteristics on the long bamboo batten surface,the detection methods between the sides of bamboo was studied.In order to reduce theinfluence of bamboo surface for defect detection,a three layer(3 inputs and 2outputs)identification model of network structure has been set up.By extracting 16 surface features from bamboo to input to the network for training,green or yellow surface of bamboo was judged according to the output.Finally it can make detection efficient and the detect detection rate can reach above 99%.(5)According to the reasons of the long bamboo categorization,bamboo color classification method was studied.In view of the uneven distribution of long bamboo color area,a feature extraction method based on the main color was proposed in this paper.In addition,a combined classifier was designed to classify different colors between bamboo battens based on the KNN algorithm combined with machine learning algorithm.The designed method is stable and effective,reducing the influence of uneven distribution of bamboo surface color,improving the anti-interference ability and detection rate.The classified rate can reach 91.58%.