小波变换的木材纹理在线分选

Wavelet Transform of Wood Texture Online Classification

以180幅木材样本图片为对象,研究以小波变换方法提取特征参数,分析几种小波基的特点和性质,最终以对称性为依据,选择使用sym4小波对图像进行二级小波分解,可以得到一级水平细节HL1、垂直细节LH1、对角细节HH1,二级的近似LL2、水平细节HL2、垂直细节LH2、对角细节HH2共7个子图,提取整幅图像的熵和每个子图小波系数的均值及标准差作为特征参数。将木材纹理按照直纹、抛物线和乱纹3种纹理的分类标准,以BP神经网络作为分类器进行了木材纹理分类的验证,并与灰度共生矩阵的方法进行了对比。试验表明:采用小波变换的方法对木材纹理特征进行描述,不但提高了分类的准确率,重要的是缩短了运算时间,可以达到在线监测的要求。

In order to classify the wood surface texture online, choose 1$0 wood sample images to study the wavelet transform for extracting the characteristic parameters, analyze the characteristics of several wavelet bases. Ultimately according to the symmetry, choose to use sym4 wavelet for image secondary wavelet decomposition. So it can get level detail HL1, vertical detail LHI, diagonal detail HH1, secondary approximate detail LL2, level detail HL2 vertical detail LH2, diagonal details HH2, 7 subgraphs in all. Extract entropy of the whole image and the mean and standard deviation of wavelet coefficients of each subgraph as characteristic parameters of wood texture. Classify the wood texture as the straight grains, parabolic and disorder grains three kinds of texture classification according to their surface shapes. Choose BP neural network classifier to classify wood texture, and compare to the gray level co - occurrence matrix method. Experiments show that： based on wavelet transform method can be used to describe the characteristics of wood texture, and improve the classification accuracy, shorten the operation time. It can achieve the requirement of online monitoring.