基于SVM的多核学习飞秒激光烧蚀光斑图像分类

Classification of ablation spot by femtosecond laser based on multi-kernel learning SVM

在利用飞秒激光加工单晶硅材料的过程中,会出现等离子体发光现象。随着飞秒激光烧蚀功率的变化,烧蚀过程中单晶硅材料表面溢出的等离子体光斑轮廓特征也大不相同。针对不同烧蚀功率下的光斑图像在分类过程中准确率不高的问题,提出了一种基于SVM的多核学习方法。首先,选取大量不同烧蚀功率下的光斑图像,对其进行预处理后提取光斑边缘轮廓信息,使用Hu不变矩和傅里叶描述子分别对图像轮廓特征进行描述。其次,建立光斑图像样本库,选取最合适的复合核函数以及核参数对分类模型进行训练。最后,使用高斯核函数与Sigmoid核函数的复合函数对光斑图像进行分类识别,实验研究表明:基于SVM的多核学习有效提高了光斑烧蚀功率的分类准确率。

In the process of processing a single crystal silicon material by femtosecond laser,the plasma luminescence phenomenon occurs. With the change of femtosecond laser ablation power,the contour characteristics of plasma spot on the surface of single crystal silicon material during ablation are also very different. In this paper,a multi-kernel learning method based on SVM is proposed for the problem that the accuracy of spot image under different ablation power is not high. Firstly,a large number of spot images with different ablation powers are selected,and the edge contour information of the spot is extracted after preprocessing,and the contour features of the image are described by Hu invariant moment and Fourier descriptor respectively. Secondly,the light spot image sample library is established,and the most suitable composite kernel function and kernel parameters are selected to train the classification model. Finally,the Gaussian kernel function and the Sigmoid kernel function are used to classify and identify the spot image. Experimental results show that multi-kernel learning based on SVM effectively improves the classification accuracy of spot ablation power.