The Study Of Wood Internal Defects Detection Based On Deeping Learning Method

The traditional wood defect nondestructive testing technology has a series of problems,such as low recognition accuracy,high cost and too complicated operation,and the traditional testing methods can not accurately show the specific location and size of wood internal defects.In this paper,a non-destructive identification and classification method of wood internal defects based on eigen matrix is proposed,in order to improve the efficiency and accuracy of wood internal defects detection and classification,and to study the correlation between the types of wood internal defects and the accuracy of model recognition.The main contents of this paper are as follows:Build the data set of wood internal characteristic matrix.In this paper,an Eigen Matrix Net image data set of wood internal eigen matrix is constructed by combining the stress wave wood nondestructive testing instrument and the feature matrix image simulation program.It contains12000 samples of three different defect categories(cavity,crack,decay)and no defect healthy trees.It is used for training,verification and testing of deep learning model.The data set adopts the methods of flipping and clipping to enhance the image data.After the data is enhanced,a large number of new images are generated.The detection model of internal defects of wood is established to identify the internal defects of wood.Based on Eigen Matrix Net image data set of wood internal feature matrix,fast r-cnn is used to extract the regional visual features,obtain the wood defect features,and train the feature model to complete the detection task.This paper analyzes the reasons why the recognition accuracy of different defect categories is different in the recognition task,and obtains the correlation between the defect categories and the classification accuracy.At present,the recognition ability of various detection methods in this field is obtained by multi-party comparison.By comparing the performance of different depth learning classification methods in data sets and the recognition ability of current mainstream wood nondestructive testing technology and traditional wood nondestructive testing technology,we can judge the advantages and disadvantages of this method and other detection methods.To realize a mobile application of nondestructive testing of wood internal defects,and to explore the development of relevant information.The experimental results show that the classification and detection model of wood internal defects generated by this method has the accuracy of 99.8% for judging whether there are defectsin wood,74.3% for detecting the location of wood internal defects,and less than 1% for center deviation.The results show that the model generated by this method is effective in identifying and classifying wood internal defects,and can reflect the correlation between the types of some wood internal defects and the accuracy of model identification.