-
实验图像预处理的具体步骤是:(1)使用SIFT算法获取图像变换参数;(2)利用图像变换参数和刚体变换模型(旋转和平移变换)截取实验图像中粒子喷溅目标区域;(3)使用BM3D算法对粒子喷溅目标区域图像进行去噪处理。
-
粒子识别采用基于FCM的图像分割方法实现,FCM通过迭代方法最小化目标函数来实现图像分割。FCM算法中的目标函数和约束条件分别为:
$$ J = \sum\limits_{i = 1}^c {\sum\limits_{j = 1}^n {\mu _{ij}^m{{\left\| {{x_j} - {c_i}} \right\|}^2}} } $$ (1) $$ \sum\limits_{i = 1}^c {{\mu _{ij}}} = 1,\;\;\;{\mu _{ij}} \in \left[ {0,1} \right] $$ (2) 计算得到迭代关系式为:
$$ {\mu _{ij}} = \frac{1}{{\sum\limits_{k = 1}^c {{{\left( {\dfrac{{\left\| {{x_j} - {c_i}} \right\|}}{{\left\| {{x_j} - {c_k}} \right\|}}} \right)}^{\frac{2}{{m - 1}}}}} }} $$ (3) $$ {c_i} = \frac{{\sum\limits_{j = 1}^n {\mu _{ij}^m{x_j}} }}{{\sum\limits_{j = 1}^n {\mu _{ij}^m} }} $$ (4) 式中:J为目标函数;c为聚类的分类数目;n为数据集中数据个数;
${\mu _{ij}}$ 为样本j属于类i的隶属度;${x_j}$ 为数据集j的位置;${c_i}$ 为类i的中心位置;m为样本的轻缓程度。粒子识别结果如图3所示,图3(a)和(b)分别为去噪处理后和去噪处理前的粒子识别结果,对比图中蓝色矩形框标记区域的粒子识别结果,发现两个区域实际存在的粒子数分别为3个和2个,但是去噪处理前图像中误识别的粒子数却达到13个和18个,说明去噪处理能够有效抑制图像噪声对粒子识别的影响。识别出喷溅粒子后,通过累加粒子像素数目和一阶矩原理计算粒子面积和形心。
-
基于Voronoi图的粒子匹配准确率和稳定性好,但是由于其苛刻的成功匹配条件,造成成功匹配的粒子数目较少,为更加充分利用粒子间的相对位置关系,文中提出一个两步的粒子匹配方法,第一步是基于分层Voronoi图的粒子匹配,其结果准确,但是成功匹配的粒子数目较少;第二步是基于弹性势能模型的粒子匹配,可以在第一步的基础上实现更多粒子的匹配。
-
成像系统中探测光捕获的两帧图像分别称为I1和I2。假设I1和I2中粒子数目分别为N1和N2,I1和I2中粒子集合分别表示为
${A^x}$ 和${A^y}$ 。粒子分层是指根据粒子面积按照由大到小顺序将${A^x}$ 和${A^y}$ 粒子集合分成$B_1^x,B_2^x, \cdots ,B_{n - 1}^x,B_n^x$ 和$B_1^y,B_2^y, \cdots ,B_{n - 1}^y,B_n^y$ 等若干个子集,子集间满足${B_{i - 1}} \subset {B_i}$ 。分层匹配就是将对应子集进行粒子匹配,前面子集正确匹配的粒子继续参与后面子集的特征描述但不参与粒子匹配,剩余粒子不仅需要参与后面子集的特征描述还要参与粒子匹配。基于分层Voronoi图的粒子匹配需要依次从$B_1^x$ 和$B_1^y$ 子集到$B_n^x$ 和$B_n^y$ 子集进行n次操作。(1)粒子筛选:筛选
$B_i^x$ 和$B_i^y$ ($i \in \left\{ {1,2, \cdots ,n - 1,n} \right\}$ )中粒子,其中属于$\left\{ {{s_k}} \right\}$ 的粒子参与特征描述但不参与粒子匹配,其余粒子同时参与特征描述和粒子匹配。(2)粒子特征描述:Voronoi图由一组连接两相邻点直线的垂直平分线构成的连续多边形组成。通过Voronoi图方法划分粒子喷溅目标区域图像,使图像中每个喷溅粒子都存在一个对应的区域,该区域称为Voronoi单元。
以8 000 ns延时粒子集为例,粒子集中粒子转化为对应特征曲线的过程如图4所示。具体来说,将原始图像(图4(a))转化为Voronoi图(图4(b)),Voronoi图中每个喷溅粒子及其单元分别用一个蓝点和一个封闭多边形表示。选择Voronoi图中一个单元作例,将所选Voronoi单元分解成一组位于直角坐标系中的紧凑拼接的三角形(Ⅰ,Ⅱ,
$\cdots $ ,Ⅴ),坐标系中心$\left( {{x_0},{y_0}} \right)$ 与单元内原始粒子的形心位置一致,如图4(c)所示。对于标记为I的三角形,其极径r作为极角$\alpha $ 函数的计算公式为:$$ r\left( \alpha \right) = \frac{h}{{\left| {\sin \left( {\alpha + {\theta _1} - {\alpha _1}} \right) + 1 - {\rm{logical}}\left( {{\theta _1} - {{90}^ \circ }} \right)} \right|}} $$ (5) $$ {\rm{logical}}\left( x \right) = \left\{ \begin{array}{l} 1\;\;\;\left( {x \ne 0} \right)\\ 0\;\;\;\left( {x = 0} \right) \end{array} \right. $$ (6) $$ h = \frac{{\left| {A{x_0} + B{y_0} + C} \right|}}{{\sqrt {{A^2} + {B^2}} }}\;\;\;\left\{ \begin{array}{l} A = {y_2} - {y_1}\\ B = {x_1} - {x_2}\\ C = {x_2}{y_1} - {x_1}{y_2} \end{array} \right. $$ (7) 对该Voronoi单元剩余的三角形(Ⅱ,Ⅲ,
$\cdots $ ,Ⅴ)进行相同操作,该Voronoi单元对应的喷溅粒子转化为特征曲线$r\left( \alpha \right)$ ,如图4(d)所示。特征描述指利用
$B_i^x$ 和$B_i^y$ 子集生成Voronoi图,并计算子集中粒子的特征曲线$r\left( \alpha \right)$ ,利用特征曲线进行粒子匹配。(3)粒子匹配与检验:确定候选粒子时设置筛选条件,即两个待匹配粒子的面积差应该小于两个粒子中面积较小粒子的面积。利用筛选条件确定
$B_i^x$ 中随机粒子${x_i}$ 的粒子候选集$\left\{ {{y_{ij}}} \right\}$ 。特征曲线间的相似系数定义为:
$$ {C_r} = \frac{{{\mathop{\rm cov}} \left( {{r_1},{r_2}} \right)}}{{\sqrt {{\mathop{\rm cov}} \left( {{r_1}} \right){\mathop{\rm cov}} \left( {{r_2}} \right)} }} $$ (8) 式中:‘cov’代表两个向量的协方差;
${r_1}$ 和${r_2}$ 分别代表两个粒子对应的特征曲线。依次计算粒子${x_i}$ 与候选粒子集$\left\{ {{y_{ij}}} \right\}$ 中粒子特征曲线间的相似系数,选取相似系数最大的候选粒子${y_i}$ 作为${x_i}$ 的匹配粒子。设置相似系数阈值过滤错误匹配,判定相似系数${C_r} \geqslant 0.6$ 时粒子匹配有效。$B_i^x$ 子集中的所有粒子重复上述匹配过程,得到粒子匹配的初步结果。利用粒子间相对位置关系检验粒子匹配的初步结果,其中正确匹配的粒子存入$\left\{ {{s_k}} \right\}$ ,并将$\left\{ {{s_k}} \right\}$ 返回粒子筛选环节。接着对$B_{i + 1}^x$ 和$B_{i + 1}^y$ 执行上述操作,直至完成对$B_n^x$ 和$B_n^y$ 的操作,每对子集正确匹配的粒子相继存入$\left\{ {{s_k}} \right\}$ ,最终正确匹配的粒子全部存储在$\left\{ {{s_k}} \right\}$ 中,粒子匹配的第一步完成,结果如图5所示。 -
(1)粒子筛选:筛选
${A^x}$ 和${A^y}$ 粒子集合中粒子,属于$\left\{ {{s_k}} \right\}$ 的粒子参与特征描述但不参与粒子匹配,剩余粒子${R^x}$ 和${R^y}$ 同时参与特征描述和粒子匹配。(2)粒子特征描述:PTV启发式1假设粒子的帧间位移是有限的[21],故四倍放大第一步中得到的粒子位移的最大值作为约束半径,同时加以面积筛选条件对粒子进行筛选,确定
${R^x}$ 中的随机粒子${x_i}$ 的候选粒子集$\left\{ {{y_{ij}}} \right\}$ 。基于弹簧模型[20]的粒子匹配以粒子簇为基本单位,假设簇内粒子由不可见的弹性弹簧连接。基于弹性势能模型的粒子匹配方法选取正确匹配粒子集
$\left\{ {{s_k}} \right\}$ 中距离待匹配粒子P最近的五个粒子作为参考粒子,计算待匹配粒子P与参考粒子间位置变化产生的弹性势能,公式为:$$ U = \frac{1}{2}k\Delta {x^2} $$ (9) $$ k = a\frac{1}{s} $$ (10) 式中:k为弹性系数;
$\Delta x$ 为形变量;a为常数;s为弹簧长度。${R^x}$ 和${R^y}$ 的特征描述便是确定${R^x}$ 中粒子在I1和I2中的参考粒子,计算${R^x}$ 中粒子候选粒子的弹性势能。(3)粒子匹配与检验:粒子匹配就是比较候选粒子集
$\left\{ {{y_{ij}}} \right\}$ 中粒子的弹性势能,选取弹性势能最小的候选粒子${y_i}\left( {{y_i} \in {R^y}} \right)$ 作为${x_i}\left( {{x_i} \in {R^x}} \right)$ 的匹配粒子。${R^x}$ 集合中的所有粒子重复上述匹配过程,得到粒子匹配的初步结果。利用粒子间相对位置关系检验粒子匹配的初步结果,其中正确匹配的粒子存入$\left\{ {{s_k}} \right\}$ ,粒子匹配的最终结果如图6所示。 -
实现粒子匹配后,利用粒子的形心坐标估算粒子瞬时速度,公式为:
$$ {\overrightarrow v _{est}} = \Delta s/\Delta t {\rm{ = }}\sqrt {{{\left( {{y_2} - {y_1}} \right)}^2} + {{\left( {{x_2} - {x_1}} \right)}^2}} /\Delta t $$ (11) 式中:
$\Delta s$ 为粒子间位移;$\Delta t$ 为I1和I2之间的时间间隔;$\left( {{x_2},{y_2}} \right)$ 和$\left( {{x_1},{y_1}} \right)$ 分别为粒子在I2和I1中的形心坐标。粒子的等效直径定义为与喷溅粒子面积相等的圆的直径,计算公式为:$$ d = \sqrt {4n{a^2}/\pi } = \sqrt {2\left( {{n_1} + {n_2}} \right){a^2}/\text{π} } $$ (12) 式中:n为
粒子在图像中所占平均像素数目;a为单个像素代表的实际尺寸; ${n_1}$ 和${n_2}$ 分别为粒子在I1和I2中所占像素数目。
Automatic acquisition of dynamic characteristics of fused silicon particle ejection induced by laser
-
摘要: 粒子喷溅是熔石英后表面激光损伤过程的重要现象,其实验观测及动力学参数获取一直存在困难。为实现后表面损伤粒子喷溅动力学特征获取,搭建了时间分辨双帧阴影成像系统,并开发了基于分层Voronoi图和弹性势能模型的粒子识别与匹配算法。结合图像配准和去噪步骤,该方法能够克服双帧阴影成像技术固有的背景噪声严重和原点失配的缺点,在6 μs及之后的延迟图像中能实现100%正确检测与匹配。与人工处理对比,该方法能以像素精度同时获取粒子的尺寸、速度大小及飞行方向,且数据处理时间缩短了约20倍。Abstract: Particle ejection is an important phenomenon in laser damage process of fused silicon exit surface, and it has been difficult to obtain the experimental observation and dynamic parameters. A time-resolved two-frame shadowgraphy system was built to acquire the dynamic characteristics of exit surface particle ejection. A particle recognition and matching algorithm based on layered Voronoi diagram and elastic potential model was developed. Combining the steps of image registration and denoising, this method can overcome the inherent shortcomings of serious background noise and mismatch of origin in two-frame shadowgraphy technology, and can achieve 100% correct detection and matching in the image of 6 μs delayed and later. Compared with manual processing, this method can simultaneously obtain the size, velocity and flight direction of particles with pixel accuracy, and the data processing time is reduced by about 20 times.
-
-
[1] Negres R A, Cross D A, Liao Z M, et al. Growth model for laser-induced damage on the exit surface of fused silica under UV, ns laser irradiation [J]. Optics Express, 2014, 22(4): 3824−3844. doi: 10.1364/OE.22.003824 [2] Sun L, Ni R, Xia H, et al. Theoretical and experimental research on laser-induced damage of fused silica optics due to stimulated Brillouin scattering [J]. IEEE Photonics Journal, 2018, 10(5): 1−15. [3] Cormont P, Combis P, Gallais L, et al. Removal of scratches on fused silica optics by using a CO(2)laser [J]. Optics Express, 2013, 21(23): 28272−28289. doi: 10.1364/OE.21.028272 [4] Doualle T, Gallais L, Cormont P, et al. Effect of annealing on the laser induced damage of polished and CO2 laser-processed fused silica surfaces [J]. Journal of Applied Physics, 2016, 119(21): 213106. doi: 10.1063/1.4953146 [5] Demange P, Negres R A, Raman R N, et al. Role of phase instabilities in the early response of bulk fused silica during laser-induced breakdown [J]. Physical Review B, 2011, 84(5): 054118. doi: 10.1103/PhysRevB.84.054118 [6] Crisp M D, Boling N L, Dube G. Importance of Fresnel reflections in laser surface damage of transparent dielectrics [J]. Applied Physics Letters, 1972, 21(8): 364−366. doi: 10.1063/1.1654414 [7] Demos S G, Negres R A, Raman R N, et al. Material response during nanosecond laser induced breakdown inside of the exit surface of fused silica [J]. Laser & Photonics Reviews, 2013, 7(3): 444−452. [8] Demos S G, Raman R N, Negres R A. Time-resolved imaging of processes associated with exit-surface damage growth in fused silica following exposure to nanosecond laser pulses [J]. Optics Express, 2013, 21(4): 4875−4888. doi: 10.1364/OE.21.004875 [9] Davis S J, Raman R N, Heaven M C, et al. Time-resolved imaging of material response following laser-induced breakdown in the bulk and surface of fused silica[C]// Proc SPIE, 2010, 7581: 75810D. [10] Raman R N. Time-resolved microscope system to image material response following localized laser energy deposition: exit surface damage in fused silica as a case example [J]. Optical Engineering, 2011, 50(1): 013602. doi: 10.1117/1.3526689 [11] Raman R N, Negres R A, Demos S G. Kinetics of ejected particles during breakdown in fused silica by nanosecond laser pulses [J]. Applied Physics Letters, 2011, 98(5): 051901. doi: 10.1063/1.3549193 [12] Demos S G, Negres R A, Raman R, et al. Comparison of material response in fused silica and KDP following exit surface laser- induced breakdown[C]//Proceedings of SPIE - The International Society for Optical Engineering, 2013, 8885: 88850W. [13] Demos S G, Negres R A, Rubenchik A. Dynamics of the plume containing nanometric-sized particles ejected into the atmospheric air following laser-induced breakdown on the exit surface of a CaF2 optical window [J]. Applied Physics Letters, 2014, 104: 031603. doi: 10.1063/1.4862815 [14] Shen C, Cheng X A, Xu Z, et al. Observation of particle ejection behavior following laser-induced breakdown on the rear surface of a sodium chloride optical window [J]. Optical Engineering, 2016, 56(1): 011009. doi: 10.1117/1.OE.56.1.011009 [15] Demos S G, Negres R A, Raman R N, et al. Relaxation dynamics of nanosecond laser superheated material in dielectrics [J]. Optica, 2015, 2(8): 765−772. doi: 10.1364/OPTICA.2.000765 [16] 沈超. 高功率纳秒激光诱导熔石英元件损伤动力学研究[D]. 长沙: 国防科学技术大学, 2016. Shen Chao. Investigation of dynamics of high power nanosecond laser-induced damage of fused silica optical windows[D]. Changsha: National University of Defense Technology, 2016. (in Chinese) [17] Lowe D G. Distinctive Image Features from Scale-Invariant Keypoints [J]. International Journal of Computer Vision, 2004, 60(2): 91−110. doi: 10.1023/B:VISI.0000029664.99615.94 [18] Dabov K, Foi A, Katkovnik V, et al. Image denoising by sparse 3-D transform-domain collaborative filtering [J]. IEEE Transactions on Image Processing, 2007, 16(8): 2080−2095. doi: 10.1109/TIP.2007.901238 [19] Bezdek J C. Pattern recognition with fuzzy objective function algorithms [J]. Advanced Applications in Pattern Recognition, 1981, 22(1171): 203−239. [20] Okamoto K, Hassan Y A, Schmidl W D. New tracking algorithm for particle image velocimetry [J]. Experiments in Fluids, 1995, 19(5): 342−347. doi: 10.1007/BF00203419 [21] Zhang Y, Wang Y, Yang B, et al. A particle tracking velocimetry algorithm based on the Voronoi diagram [J]. Measurement Science and Technology, 2015, 26(7): 075302. doi: 10.1088/0957-0233/26/7/075302 [22] Diaz R, Chambonneau M, Courchinoux R, et al. Influence of longitudinal mode beating on laser-induced damage in fused silica [J]. Optics Letters, 2014, 39(3): 674−677. doi: 10.1364/OL.39.000674 [23] Raman R N, Elhadj S, Negres R A, et al. Characterization of ejected fused silica particles following surface breakdown with nanosecond pulses [J]. Optics Express, 2012, 20(25): 27708−27724. doi: 10.1364/OE.20.027708