Research on Wavelet Based Autofocus Evaluation in Micro-vision

This paper presents the construction of two kinds of focusing measure operators defined in wavelet domain. One mechanism is that the Discrete Wavelet Transform （DWT） coefficients in high frequency subbands of in-focused image are higher than those of defocused one. The other mechanism is that the autocorrelation of an in-focused image filtered through Continuous Wavelet Transform （CWT） gives a sharper profile than blurred one does. Wavelet base, scaling factor and form to get the sum of high frequency energy are the key factors in constructing the operator. Two new focus measure operators are defined through the autofocusing experiments on the micro-vision system of the workcell for micro-aligmnent. The performances of two operators can be quantificationally evaluated through the comparison with two spatial domain operators--Brenner Function （BF） and Squared Gradient Function （SGF）. The focus resolution of the optimized DWT-based operators is 14% higher than that of BF and its computational cost is 52% approximately lower than BF＇s. The focus resolution of the optimized CWT-based operators is 41% lower than that of SGF whereas its computational cost is approximately 36% lower than SGF＇s. It shows that the wavelet based autofocus measure functions can be practically used in micro-vision applications.

IMPROVEMENT OF PHASE GRADIENT AUTOFOCUS ALGORITHM

The phase error estimated by phase gradient autofocus(PGA) is not based on a finite order polynomial mode, so PGA has a good autofocus property for arbitrary order phase error and is fit for high resolution airborne SAR. But PGA has two shortcomings: first, it has a worse estimation property for fast changing phase error; second, there exists a section of linear phase in the phase error estimated by this algorithm. This paper introduces the idea of rank one phase estimate (ROPE) autofocus technique, and improves PGA. The improved PGA(IPGA) can successfully overcome both these shortcomings of PGA.

A Simple Model for On-Sensor Phase-Detection Autofocusing Algorithm

A simple model of the phase-detection autofocus device based on the partially masked sensor pixels is described. The cross-correlation function of the half-images registered by the masked pixels is proposed as a focus function. It is shown that—in such setting—focusing is equivalent to searching of the cross-correlation function maximum. Application of stochastic approximation algorithms to unimodal and non-unimodal focus functions is shortly discussed.

Image Reconstruction from Fan-Beam Projections without Back-Projection Weight in a 2-D Dynamic CT: Compensation of Time-Dependent Rotational, Uniform Scaling and Translational Deformations

In a dynamic CT, the acquired projections are corrupted due to strong dynamic nature of the object, for example: lungs, heart etc. In this paper, we present fan-beam reconstruction algorithm without position-dependent backprojection weight which compensates for the time-dependent translational, uniform scaling and rotational deformations occurring in the object of interest during the data acquisition process. We shall also compare the computational cost of the proposed reconstruction algorithm with the existing one which has position-dependent weight. To accomplish the objective listed above, we first formulate admissibility conditions on deformations that is required to exactly reconstruct the object from acquired sequential deformed projections and then derive the reconstruction algorithm to compensate the above listed deformations satisfying the admissibility conditions. For this, 2-D time-dependent deformation model is incorporated in the fan-beam FBP reconstruction algorithm with no backprojection weight, assuming the motion parameters being known. Finally the proposed reconstruction algorithm is evaluated with the motion corrupted projection data simulated on the computer.

Modified Approach to PGA Phase Averaging for ISAR Autofocus

A new approach to phase averaging in phase gradient algorithm (PGA) is proposed, which is based on the fundamental fact that the information of translational phase error is widely contained in every defocused scatterer in ISAR image. The new approach aims to choose strong scatterers for error phase averaging with a threshold rather than just simply to pick out the strongest point in each range cell, which is not necessarily real strong scatterers if in some range cells consists more than one strong scatterer and whereas in other range cells no scatterer at all. The results of processing real data are presented to confirm the validity of the proposed approach.

Rupture imaging of the 25 April 2015 M_W7.9 Nepal earthquake from back-projection of teleseismic P waves

The M7.9 Nepal earthquake of 25 April2015 had over 8, 500 fatalities and was the most destructive earthquake in Nepal since the Bihar-Nepal earthquake in 1934.In this study, we imaged the rupture process of this Nepal event by back-projecting the teleseismic P-wave energy recorded at the three regional networks in Alaska, Australia and Europe. The back-projection images of the three subarrays revealed that the Nepal earthquake propagated along the strike in a southeast direction over a distance of ～ 160–170 km with the duration of ～ 50–55 s. The rupture process was found to be a simple, unilateral event with a near constant velocity of 3.3 km/s.The beam power was mainly distributed in the geographic region just north of Kathmandu and the peak intensity for the source time function curve occurred at about 30 s. The earthquake was destructive due to its occurrence at shallow depth(～ 12–15 km) and the fact that the capital lies in a basin of soft sediment. Additionally, the resonance effect for the longer period waves that occurred in the Kathmandu valley led to destructive aggravation, impacting mainly the taller buildings.

Implementation of Autofocus in Alignment System for Layered Imprint Fabrication

Autofocus method based on the analysis of image content information is investigated to reduce the alignment error resulting from mark positioning uncertainty due to defocus in microstructure layered fabrication process based on multilevel imprint lithography. The applicability of several autofocus functions to the alignment mark images is evaluated concerning their uniformity, sharpness near peak, reliability and measure computation efficiency and the most suitable one based on power spectrum in frequency domain (PSFD) is adopted. To solve the problem of too much computation amount needed in PSFD algorithm, the strategy of interested region detection and effective image reconstruction is proposed and the algorithm efficiency is improved. The test results show that the computation time is reduced from 0.316 s to 0.023 s under the same conditions while the other merits of the function are preserved, which indicates that the modified algorithm can meet the mark image autofocusing requirements in response time, accuracy and robustness. The alignment error due to defocus which is about 0.5 μm indicated by experimental results can be reduced or eliminated by the autofocusing implementation.

ADAPTIVE KURTOSIS OPTIMIZATION AUTOFOCUS ALGORITHM

This paper proposes an autofocus algorithm used for Synthetic Aperture Radar (SAR) images, called Adaptive Kurtosis Optimization Autofocus Algorithm (AKOAA). The AKOAA can reduce the differ-ence between initial value and real value in focusing by adaptively adjusting the initial value, therefore effec-tively improve the local extremum problem in the Contrast Optimization Autofocus Algorithm (COAA) and speed up the convergence velocity. The principle and realization method of AKOAA are thoroughly investi-gated, and experimental results using real L-band SAR data show that the focus speed of AKOAA is nearly doubled compared with that of the COAA, and the image contrast of AKOAA is improved as well.

面向高密度柔性集成电路封装基板的精密视觉检测算法

高密度柔性集成电路(IC)封装基板是电子组件的核心载体,其制造过程中的刻蚀、显影等工艺需要严格的品质控制.本文开发了一种融合金相显微镜与工业相机的成像视觉检测系统,用于柔性IC封装基板的图像采集与品质控制.针对显微成像视野小、易离焦等问题,提出了一种基于中值的三元模式局部纹理快速自动对焦算法.此外,针对高倍镜下柔性IC封装氧化缺陷分布不均匀问题,引入微分几何工具,利用曲率等几何特征提出了一种基于动态曲线演化的高精度表面氧化分布特征检测模型.通过定性与定量的实验研究,验证了两种算法在快速性和高精度方面的优越性能.

融合全局和局部信息的铁谱图像自动对焦算法

针对铁谱图像获取时人工对焦误差大、速度慢等问题,提出了一种融合全局信息和局部信息的铁谱图像自动对焦方法。此方法分为两个阶段:全局对焦阶段利用卷积神经网络(Convolutional Neural Networks,CNN)提取整幅图像的特征向量,并利用门控循环单元(Gate Recurrent Unit,GRU)融合对焦过程提取的特征,预测当前全局离焦距离,起到粗对焦的作用;局部对焦阶段提取磨粒的特征向量,利用GRU融合当前特征与前一轮对焦提取的特征,并依据最厚磨粒信息,预测当前磨粒离焦距离,起到精对焦的作用。同时,为了提高对焦准确率,提出了结合拉普拉斯梯度的对焦方向判定法。实验结果表明,此算法在测试集上的对焦误差为2.51μm,当景深为2.0μm时对焦准确率为80.1%,平均对焦时间为0.771 s。本文提出的自动对焦方法具有较好的性能,为铁谱图像自动准确采集提供了技术支持。

Rapid rupture characterization for the 2023 M_(S)6.2 Jishishan earthquake

On December 18, 2023, the M_(S)6.2 Jishishan earthquake occurred in the northeastern region of the QinghaiXizang Plateau, causing heavy casualties and property damage in Gansu and Qinghai Provinces. In this study,we integrate space imaging geodesy, finite fault inversion, and back-projection methods to decipher its rupture property, including fault geometry, coseismic slip distribution, rupture direction, and propagation speed. The results reveal that the seismogenic fault dips to the southwest at an angle of 29°. The major slip asperity is dominated by reverse slip and is concentrated within a depth range of 7–16 km, which explains the significant uplift near the epicenter observed by both the Sentinel-1 ascending and descending In SAR data. Moreover, the teleseismic array waveforms indicate a northwest propagating rupture with an overall slow rupture velocity of~1.91 km/s(AK array) or 1.01 km/s(AU array).

基于卷积神经网络的快速SAR自聚焦方法

自聚焦是SAR高分辨率成像的关键技术。然而,传统的SAR自聚焦方法均需要迭代多次,实时性差,不适合在轨处理。本文提出了一种基于卷积神经网络的在轨快速SAR自聚焦方法(CNN-AF),该方法采用卷积神经网络来学习失焦图像到聚焦图像的映射,主要用于校正方位向的相位误差,由于在测试阶段该方法无须迭代和调整参数,因此该方法实时性好,更加适用于在轨处理。在真实SAR数据上的试验结果表明,本文方法具有较高的聚焦质量和聚焦速度。

基于物方差动共焦轴向测量引导的显微镜自动对焦方法

针对传统自动聚焦方法效率低,难以适应工业产品微观形貌检测高精度、高效率需求的问题,提出一种基于物方差动共焦轴向测量引导的显微镜自动对焦方法(DCAFM)。在获取测量系统中离焦量和轴向光强差之间的标定曲线后,控制运动机构使物镜与样本的轴向距离以等步距的方式采集图像。通过焦前与焦后位置光强差快速确定当前轴向离焦量,控制轴向运动机构,实现快速自动对焦。搭建显微镜自动对焦实验平台,对样品进行测试。实验结果表明:文中方法的图像对焦速度和灵敏度优于DFT,EOG等自动对焦方法,系统的轴向分辨率优于0.4μm,有效地提高了显微镜的对焦速度和精度。

面向大跨度场景的全景深快速聚焦成像法

非结构化场景在特种行业中作用十分关键,保证其内表面健康检测对装置的安全运行至关重要。但其内表面景深跨度大,极易引起常规相机成像时的景深不足或景深误匹配,致使聚焦模糊,影响表面缺陷检测效果。因此,本文提出了基于极值搜索算法的聚焦跟踪曲线快速构造法,并实现了相机的快速高清成像。首先,建立了聚焦跟踪曲线的基本构造方法;其次,提出了极值搜索算法,简化曲线构造过程;最后,设计基于A/B标定板的单/多聚焦跟踪曲线构造方式,并分析其对成像清晰度的影响。结果表明,本算法在准确聚焦的同时,有效减少了图像采集次数且将聚焦速度提高了34.8%。基于多聚焦跟踪曲线的全景深成像,其图像客观评价指标均值提高了10.35%。可实现100~1 000 mm景深跨度下0.5 mm裂纹的清晰分辨。因此,该方法能为常规相机的非结构化场景中高清全景深成像提供解决方案。

一种改进选点及加窗处理的相位梯度自聚焦算法

相位梯度自聚焦(Phase Gradient Autofocus,PGA)算法被广泛应用于补偿合成孔径雷达(Synthetic Aperture Radar,SAR)图像的相位误差。PGA算法的处理流程中,选点和加窗这两步操作对于算法性能的影响非常大。传统的PGA算法常因选点质量不佳或窗宽估计有误而导致聚焦效果变差,收敛速度变慢。本文提出一种基于能量-信杂比最大准则的强点选择方法和一种自适应窗宽估计方法。利用能量和信杂比两个维度从图像数据中筛选出理想的孤立强散射点,结合并改进两种传统加窗方法自适应地估计窗宽,实现算法稳定性和收敛速度的提升。仿真和实测数据处理结果证实了本文算法的有效性。

刀具损伤视觉检测系统自动对准对焦研究

为解决数控机床刀具无拆卸条件下,基于机械臂的刀具损伤视觉检测系统对准对焦调节耗时长、计算分析方法鲁棒性差等难题,提出了一种融合YOLOv5网络智能感兴趣区域(region of interest, ROI)的机器人视觉系统自动对准对焦方法。首先,利用ROI模型检测并定位刀具中心,通过九点标定法计算机械臂末端对准坐标;然后自适应筛选ROI对焦窗口,采用改进的Laplacian函数计算清晰度评价值以确定最佳刀具图像。在实际设备上开展实验后结果表明,所提方法比一般方法灵敏度至少提高1.63倍,平均中心点误差为3.76像素,有效提升了刀具损伤视觉检测系统的准确度和灵活性。

随机样本选择的合成孔径雷达距离空变相位梯度自聚焦算法

针对距离依赖的相位梯度自聚焦(phase gradient autofocus,PGA)算法中样本选择的问题,本文提出了一种新的基于随机样本选择的距离依赖PGA(range-dependent PGA,RDPGA)算法。不同于传统算法利用固定门限对特显点样本进行硬剔除的选择方式,该算法利用样本的信杂比(signal to clutter ratio,SCR)构造了样本选择概率密度函数,在每次PGA迭代估计过程中,利用该概率密度函数对样本进行随机选择。随机样本选择方法不仅通过增加距离依赖样本的丰富性保证了RDPGA的估计精度,同时还保证了高质量样本在模型参数估计中提供较高贡献,在保持高效性的同时进一步提升了算法的稳健性。实测数据处理结果表明所提算法具有较高的估计精度和稳健性。

基于自动聚焦的近眼显示虚像距离测试装置研究

目的 虚拟现实(Virtual Reality,VR)和增强现实(Augmented Reality,AR)能够将虚拟图像渲染并投影到人眼,通过控制近眼显示器的屈光度,可以调节虚像距离,从而适应不同视力的人员观看。但由于虚像距离跨度范围较大,很难实现连续准确的测量,为解决这一问题,本文提出基于自动聚焦的近眼显示虚像距离测试装置。方法 提出一种可准确测定VR/AR设备虚拟图像距离的方法。装置由可调焦镜头、步进电机、固件工装及自适应程序构成,并通过步进电机配合图像清晰度分析自适应程序自动调节可调焦镜头的焦距。分析该装置的检查精度和测试范围。结果 相比传统方法(全局扫描的方法),本文使用爬山搜索法是通过对比前后的拉普拉斯值来确定电机的移动方向以及移动步长,基于爬山算法的自动聚焦方法聚焦时间缩短,效率提高,且测试装置性能更佳,该装置自动完成整个测量过程仅需要4 min。实验的精度为0.001 m,该商业VR眼镜在+2D到-4D的度数下,虚像距离的变化范围为0~13.830 m;且根据距离标定曲线可以得到,运用该装置能够测得虚拟图像距离在17 m以上。结论 本文所提出的基于自动聚焦的近眼显示虚像距离测试方法及装置能够对虚拟图像的深度距离进行高精度、大行程、自动化测试,可准确标定近眼显示设备的虚拟图像深度距离,有效提升测试性能。

Rapid assessment of the September 5,2022 M_(S)6.8 Luding earthquake in Sichuan,China

At 12:52,September 5,2022,an M_(S)6.8 earthquake occurred in Luding,Sichuan.The earthquake caused serious casualties and property loss,and was determined to have an epicenter intensity of Ⅸ degree.In this study,we used three earthquake intensity rapid assessment methods(i.e.WFM,BPM and ASM) to evaluate the intensity of this earthquake.Then,we comparatively analyzed the three methods based on strong ground motion observation data and actual intensity maps.The results show that:(1) The earthquake is associated with a southeast-oriented single-sided rupture.The WFM method can only evaluate earthquakes with two-sided ruptures,which has some limitations;(2) The intensity of BPM and ASM was overestimated on the southwest and north sides of the epicenter,but other high-intensity zones were similar to the intensities measured by actual surveys;(3) The residuals of the three intensity assessment methods were all between-0.5 and 1.Although a small number of stations were underestimated,the overall residuals were good,and the residuals gradually approached 0 with the increase of distance;(4) The number of towns and villages evaluated by the three methods in the earthquake area was almost all lower than the field survey results.One exception is the area of Ⅷ degree,where the BPM and ASM were higher than the survey results;(5) The area of the earthquake area evaluated by the three methods was low in Ⅵ and Ⅶ degree,moderate in Ⅷ degree,and low in Ⅸ degree(the area from ASM is similar to the area measured by actual survey).Overall,ASM is applicable to this earthquake intensity assessment.

高平面分辨率穿透成像雷达的研究与实现

穿透成像雷达利用电磁波对介质的穿透特性,对非金属介质内的不连续点进行高分辨成像。为实现该雷达系统的毫米级高平面分辨率、高探测效率和高便携适用性,设计了高平面分辨率穿透成像雷达系统。采用连续波体制和快速扫描空间采样方案以确保小型化和高成像性能,实现了一体化雷达射频前端;提出了参数未知条件下的自聚焦成像处理等数据处理方法,研制了穿透成像雷达系统样机,质量仅为2.5 kg,可单人手持操作;开展了成像分辨率和穿透能力实验测试,验证了方案的可行性和有效性。