Research On Wood Classification Method Based On Convolution Neural Network And Visual Word Bag Model

Wood is not only an important natural renewable polymer compound,but also an essential living resource.It plays an important role in the national economy.Due to the complexity of wood composition and the fact that most of the work of wood species classification is completed manually,there are common problems of low classification efficiency and low classification accuracy.Therefore,it is very important to propose an effective wood species identification method.Wood cross-section microscopic images contain a large number of micro cell structure information.Based on this,automatic species recognition and classification has shown a certain potential in recent years.Word bag model is a technology applied to text classification.In this paper,word bag model is extended from the field of natural language processing and analysis to the field of image processing and analysis.For any image,the bag of visual word(Bo VW)model extracts the basic elements in the image,and counts the frequency of these basic elements in the image,which is expressed in the form of histogram.Because the visual word bag model can work effectively only when the provided feature extractor is well matched,this paper uses convolutional neural network(CNN)to optimize the proposed feature extractor,so as to learn more appropriate visual words from wood crosssection microscopic images,and applies piecewise linear activation function in the feature extractor based on convolutional neural network,Verify its performance on wood cross-section micro image data set.The full text mainly focuses on the following contents:(1)Aiming at the problem of species identification of wood microstructure by using adaptive feature extractor in visual word bag model.In this paper,convolutional neural network is used to optimize the proposed feature extractor in order to learn more appropriate visual words from wood cross-section microscopic images.The feature extractor based on convolutional neural network and image interpretation based on visual word bag model are combined into wood species classification.Two wood microscopic image data sets are used to test the effectiveness and feasibility of the improved CNN-Bo VW algorithm.(2)Aiming at the problem that the general activation function only extracts significant features and cannot accurately model and classify subtle visual differences,this paper uses the piecewise linear activation function to improve the selection ability of the existence degree and deletion degree of features in the improved CNN-Bo VW algorithm,and searches the optimal slope in the optimized feature selection model,which is tested by different data sets,The superiority of CNN-Bo VW algorithm using piecewise linear activation function in image recognition is proved.(3)The visual word bag model based on improved convolution neural network is applied to the cross-sectional microscopic image species classification of more than 70 kinds of commercial wood in Central Africa.The experimental results show that the experimental accuracy of this model is 89.41%,which is better than that of the original visual word bag model,which is 78.70%,and has a more accurate classification effect than CNN-Bo VW algorithm using feature extractors with different activation functions,It is proved that the algorithm in this paper is effective.