Wood Surface Defect Recognition Based On Wavelet Transform And LBP

Nowadays,with the rapid development of wood processing and production,the intensive production of the processing industry has led to a substantial increase in the volume of timber production.In this process,the requirements of the quality of the wood surface are becoming more and more strict,and the surface quality of wood has a direct impact on the economic benefits and sensory value of wood products,so it is difficult to complete the task with the method of traditional artificial observation and detection.In this paper,it is very meaningful to explore and study the identification and classification of wood surface defects based on this starting point.This paper takes dry-knot,encased-knot,crack of wood as the main research object,and deeply study the image processing and pattern recognition of wood surface defects detection.what's more,the feature extraction algorithm based on wavelet transform and LBP is proposed,which is applied to the multi-scale analysis of wood surface defect recognition.Image segmentation is the primary defects of image recognition,this paper introduces and discusses several kinds of threshold segmentation technology,selecting the optimal threshold segmentation iterative algorithm used in wood defect image segmentation.At the same time,several kinds of edge detection algorithms are compared,and the edge operator which has a good effect is selected.This paper focuses on the feature extraction of image of wood defects,in order to extract local texture feature of defect,this paper proposes the extraction method of surface defects of wood texture feature based on wavelet transform and LBP algorithm,and a multi-scale two-stage wood defect classification method is proposed on this basis.(1)A wavelet decomposition of the wood defect image is carried out,and a low frequency sub image is obtained,and the 3 high frequency sub images are obtained.(2)Extracting the shape feature of the defect area on the low frequency sub image,and the uniform LBP texture feature is extracted on each high frequency sub image.(3)Classifying knots and crack based on shape features of defects,and further distinguish dry-knot and encased-knot by the LBP multidimensional eigenvector.In this paper,BP neural network and support vector machine(SVM)are used to identify the two kinds of classifiers.The experimental results show that the rate of the two kinds of classifiers for identification of crack and knot reached more than 90%.However,because the sample characteristics of high dimension in the identification of dry-knot and encased-knot,support vector machine has more advantages in the process of recognition with shorter time and more accurate recognition accuracy.The experimental results show that the proposed algorithm based on wavelet transform and LBP can effectively solve the problem of defect identification.