Research On Wood Defect Detection System Based On Image Processing

China is a large timber importing country,and the demand for wood has increased year by year.However,the comprehensive utilization rate of timber in China is low,and the supply and demand of timber is tight.Different products have different requirements on wood texture and hardness,so the detection of wood defects is essential in the production process.The existing wood defect detection technology is mostly based on manual detection,and the false detection rate is high and the efficiency is low.To solve this problem,this paper proposes a research on wood defect detection system based on image processing,and has completed the following main research work:Firstly,the three kinds of threshold segmentation methods,including maximum inter-class difference method,fuzzy entropy method and cross entropy method,and K-Means,Mean Shift and K-Medoids clustering methods are used to segment the wood defect image.The results show that cross-entropy method has more advantages than other methods.At the same time,the open-close operation of binary image is introduced,and the optimal result is obtained by using the open operation.In order to improve the segmentation effect of one-dimensional cross-entropy thresholding algorithm when there is little difference between defect and wood background color texture,such as leaf knot.At the same time,in order to improve the speed of operation,the Satin Blue Gardener Bird Optimization(SBO)algorithm is improved,and a two-dimensional cross-entropy threshold segmentation algorithm based on chaotic SBO is proposed.The experimental results show that the algorithm has better segmentation ability for wood defects like leaf knot which texture is similar to background color texture,and has more efficient operation speed than the traditional two-dimensional cross-entropy segmentation algorithm.Secondly,based on the texture characteristics of wood defects,feature extraction of wood defect images is performed using Nonsubsampled Contourlet Transform(NSCT).The mean,variance,energy and skewness of the decomposed eight-directional subgraph were calculated to obtain the NSCT characteristic quantities of wood defects.At the same time,the application of local binary mode(LBP),Tamura and gray level co-occurrence matrix(GLCM)on the feature extraction of wood defect images is studied.The recognition effects of each feature extraction method in BP neural network and support vector(SVM)are compared.The results show that NSCT contains better recognition ability in single feature quantity.Subsequently,the multi-features were tested in SVM,and it was found that the combination of Tamura,GLCM and NSCT showed the best performance,which could reach 95.9% accuracy.Finally,the hardware and software system of the wood defect detection system is designed,including the complete operation process,hardware structure,algorithm migration and PC software system design of the system.The improved two-dimensional cross entropy segmentation algorithm is written into the embedded control system,and the system is tested.The results illustrate that the whole process from image capture to obtain sawing calibration information only needs 0.984 s,which shows real-time detection capability.