中文摘要：

提出了一种基于小波矩不变量和保局投影（LPP）的特征提取方法,并应用于中厚板表面缺陷自动识别.首先对图像做三级小波变分解,将中厚板表面图像的细节分解到各个尺度的各个分量中并利用小波阈值收缩法降噪;然后对各分量的傅里叶幅值谱提取Hu不变矩作为原始特征向量,并利用LPP将该特征向量的维数从77维降到8维;最后利用AdaBoost分类器对样本进行分类识别.实验结果表明,本文提出的特征提取方法适用于中厚板表面缺陷分类,识别率达到91.60%.

英文摘要：

A feature extraction method based on wavelet moment invariant and locality preserving projection （LPP） was presented and applied to the automatic recognition of plate surface defects. 3-level wavelet decomposition was performed on the surface images, details of the plate surface images were decomposed into components on several scales, and then the noise scattered in detail components of all the scales was reduced by wavelet shrinkage. Moment invariants were extracted from amplitude spectra of all the components, and then the feature vector composed by all the moment invariants was reduced from 77-demension to 8-dimension via LPP. At last, an AdaBoost classifier based on decision trees was constructed to classify the samples. Experimental results demonstrated that the feature extraction method presented in this paper was applicable to the classification of plate surface defects, and the classification rate was 91.60% .