摘要
针对惯性约束聚变(ICF)光学元件损伤问题,提出了一种基于随机抽样一致性(RANSAC)及最小二乘支持向量机(LSSVM)回归的高精度检测方法。建立了损伤区域总灰度与实际尺寸的回归模型,通过该回归模型对待检测损伤区域的尺寸进行预测,得到损伤区域的高精度尺寸。为剔除回归模型建立过程中离群样本点的影响,采用RANSAC方法对训练样本进行优化选择。针对抽样组中样本数对检测精度及检测效率的影响进行了相关实验,确定了抽样组中样本数的合适区间。RANSAC-LSSVM方法可通过改变误差评价函数得到不同评价体系下的最优回归模型。实验证明,在传统像素级检测方法的基础上,该方法将损伤尺寸检测的平均相对误差降低了近90%。
A high precision method, vector machine （LSSVM） regression, based on random sampling consensus is advanced to improve the inspection （RANSAC） and least squares support for inertial confinement fusion （ICF） optics damage. In present study, the measurement precision of damaged areas is improved significantly, via analyzing the regression model of grey value and size. RANSAC is adopted for optimizing the training samples to avoid the effect of outliers. In addition, special experiment is proceeded to assess the effect of sample size on the measurement precision and the efficiency, and the appropriate interval of sample size is obtained as well. RANSAC-LSSVM can get optimal regression models which is adopted in different evaluative systems, by using different error evaluation functions. The results show that RANSAC-LSSVM reduces the mean relative error by 90% approximately, in measuring of damaged areas, compared with traditional pixel-level detection.
出处
《中国激光》
EI
CAS
CSCD
北大核心
2013年第8期160-164,共5页
Chinese Journal of Lasers
基金
国家自然科学基金(51275120
61275096)
航空科学基金(20120177004)
中央高校基本科研业务费专项资金(HIT.NSRIF.2013012)
关键词
图像处理
光学元件损伤
在线检测
回归
最小二乘支持向量机
随机抽样一致性
imaging processing
optics damage
online inspection
regression
least squares support vector machine
random sampling consensus
