Research On Wood Defect Recognition Method Based On Convolutional Neural Network

Recognition and identification of wood defects is an important part of Wood Physics and wood environmental science.It has important scientific research significance and practical value.How to identify wood defects efficiently and accurately has become a difficult problem in wood science.Traditional visual inspection and image processing methods are not competent,so it is of great significance and application value to find a technology that can replace traditional methods for wood defect recognition and classification.In order to deeply study the image processing and pattern recognition of wood defects,aiming at the characteristics and image extraction of wood defects,the convolution neural network algorithm is used to recognize and analyze the image of wood defects,and the convolution neural network is introduced to solve this problem,which has both theoretical and practical feasibility.In this paper,a convolutional neural network algorithm is used to identify wood defects.The work is as follows:First,there are many kinds of trees selected and their defects are different,which makes the number of each sample smaller.It is difficult to identify each sample by using traditional convolution neural network method.To solve this problem,a cross-layer convolution neural network model is proposed.The difference between this model and traditional convolution neural network is that the model can add low-level features of the network to the classifier,so that the recognition efficiency of convolution neural network is higher.Second,recognition of wood defects is slow because of the complexity of the image and the excessive parameters involved.To solve this problem,an improved stochastic gradient descent method is used to optimize the convolutional neural network,so that the learning rate of the network changes adaptively with the change of the network.The optimized network can identify the defects in the image more quickly and improve the recognition efficiency.Finally,the improved stochastic gradient descent algorithm is applied to the cross-layer convolution neural network model,which can further optimize the network.The cross-layer idea is introduced to improve the accuracy of the improved convolution neural network.The optimization algorithm can improve the convergence speed of the network.Combining the two methods can identify wood defects more effectively.The experimental results show that according to the image characteristics of wood surface defects,the improved convolution neural network is used to solve the problem of wood surface defects recognition,so that the recognition accuracy of the network reaches a higher level.