Real-time Detection Algorithm Of Wood Board Based On Machine Vision

Wood classification is an important part of wooden furniture production.The brightness of the wood surface is different,the texture structure is fine and natural and irregular,so the classification task based on the brightness and texture characteristics of the wood surface has always been a frontier topic in the field of machine vision and image processing.The surface defects of wood have a great influence on the quality of wooden furniture.There are many types of wood surface defects and complex shapes.The detection method of manual inspection cannot be achieved by conventional methods.How to improve the classification efficiency and sorting quality of wood is the production process of wooden furniture.An important part of it.In recent years,visual inspection technology has developed rapidly,making it possible to complete wood sorting through machine vision technology.A plywood imaging experimental platform was designed.The hardware of the experimental platform consists of a camera,a light source,and a board simulation production line.Introduce the types of cameras and light sources.Select the camera,lens and light source hardware through imaging requirements,and discuss and design the lighting scheme under different surfaces.The software part of the experimental platform is image acquisition software.This thesis introduces the function of the software and collects the wood sample library.The thesis completed three aspects of work,1,wood imaging acquisition system design;2,wood classification method research and design;3,wood board defect recognition algorithm design.In the aspect of imaging system design,a cemented wood board imaging experimental platform was designed.The experimental platform consists of a camera,lens,light source and board simulation production line.The thesis introduces the important parameters of camera,lens and light source.The imaging platform with the best sampling effect is designed by experiment and theoretical calculation.The human-computer interaction interface of real-time classification system of wood board is designed according to the actual needs of the project.In the design of wood classification algorithm,this thesis proposes a classification method based on multi-feature fusion from the perspective of human eye bionics.The wood is classified according to the comprehensive brightness and texture characteristics of the surface.The board area in the image is identified by dynamic threshold segmentation,feature screening,morphological processing,and the like.A variety of eigenvalues based on gray level co-occurrence matrix and human eye visual classification criteria were extracted,and multi-layer neural networks were constructed for classification and recognition.In the design of the board defect recognition algorithm,a blue-defect defect recognition algorithm based on the color difference method is designed for the problem of blue-defect defects on the board surface.Aiming at the defects of dead knots and cracks on the surface of wooden boards,a thesis based on convolutional neural network is proposed.In order to verify the stability and effectiveness of the real-time detection system of the board,the thesis designed a variety of targeted comparison tests to verify.In order to solve the problem of multiple differences in board image acquisition,the thesis compares different light source lighting schemes.In order to verify the efficiency of the board texture extraction algorithm,a comparative experiment based on different segmentation algorithms is designed.Based on the effectiveness of the classification algorithm,a comparative experiment based on different feature extraction methods was designed.To verify the stability of the wood classification algorithm in the thesis,a robust contrast experiment was designed when the illumination intensity was insufficient.The experiment proves that the algorithm can complete the real-time classification and defect recognition task of the board,and has a comprehensive classification accuracy of more than 94.17%.