Research On Defect Detection Technology In Wood Processing Automation

The traditional wood defect detection is mainly done by manual work, which consumes a lot of manpower and material resources and the efficiency is very low. Automatic identification of wood defects by machine learning methods overcomes the influence of fatigue and subjective factors of artificial recognition. This method realizes the accurate location of wood surface defects and improves the utilization ratio of the plate. However, the core identification technology of machine learning method needs further study.This thesis studies the method of wood image feature extraction, and then performs cluster analysis on the extracted data in aim to detect the type and location of wood surface defects.The main research contents of this thesis are as follows:Firstly, three kinds of wood image feature extraction methods are proposed. 1. Two kinds of color feature extraction methods, including the histogram and color moment, realize the dimension reduction of the data. Then this thesis uses clustering algorithm to identify the defect types of wood and locate the defects. 2. Two texture feature extraction methods, based on GLCM and Gabor wavelet transform are proposed. Then this thesis uses clustering algorithm to recognize wood surface defects. 3. After the Gabor wavelet transform, a new method of wood image feature extraction based on Gabor+ PCA is proposed. The experimental part is based on K-means, SOM and AP algorithm to cluster the data set extracted from this method. The result shows that the texture feature extraction method based on Gabor +PCA has better recognition accuracy.Secondly, a wood defect recognition method based on affinity propagation(AP) clustering algorithm is proposed. This advanced algorithm improves the clustering efficiency and accuracy by using the multiple scan images and automatically adjusting the P value. Traditional AP algorithm and improved AP algorithm are compared by using the wood image feature extracted data sets. The result shows that the improved AP algorithm has better performance than the traditional algorithm.Thirdly, a wood defect recognition method based on a Self Organizing Feature map(SOM) clustering algorithm is proposed. This thesis uses color moment feature extraction method to extract the data set, and then use both SOM algorithm and K-means algorithm to recognize wood surface detect in aim to determine the clustering effect of SOM clustering method on wood defect recognition.