Ultrasonic tomography of metal structures using the digitally focused antenna array method

Physical principles and algorithms for reconstructing images of the inner structure of an object made of a solid material are considered. These are based on the pulsed echo method of ultrasonic testing using multielement antenna arrays focused on each point of the visualized region of the object by spatiotemporal processing of signals from a combination sounding of the object by all possible pairs of the antenna array. Substantial improvement of the image during testing of a plane-parallel object is obtained by using signals that are multiply reflected from the object boundaries; the use of different algorithms of image reconstruction is expedient for different types of discontinuity flaws.     

Automated Acquisition and Processing of 3D GPR Data for Object Detection and Characterization

To improve image resolution of GPR data, proper data acquisition is the first key part of the whole imaging sequence. Obtainable down and cross range resolutions, insofar as they are related to acquisition, are determined by the frequency bandwidth of the GPR equipment, the temporal and spatial sampling rates and the accurate knowledge of the antenna positions. The key parameters for three-dimensional imaging algorithms are wave velocity distribution in the domain of interest, radiation (polarization and amplitude) characteristics of the antenna system and emitted waveform. To this end we have designed and manufactured a measurement frame with high horizontal position accuracy of the antennas, by using a servomotor that automatically moves the antennas over the whole area of the measurement frame for a fixed vertical position of the antennas. When acquisition is completed, the data needs to be processed such that object detection and characterization is most easily accomplished. In most cases, object detection can be carried out on the preprocessed data itself, but in some cases and certainly for characterization purposes improved image resolution is required. This is realized using imaging algorithms, preferably those that take into account the polarization and amplitude characteristics of the antennas.Now with Shell International Exploration and Production B.V.     

Obtaining flaw images that take the effect of multiple ultrasonic pulse reflections from the boundaries of a test object into account

The problem of obtaining flaw images from echo signals multiply reflected from the boundaries of a test object and modifications of the FT-SAFT and SAFT algorithms for obtaining flaw images in a test object with plane-parallel boundaries are considered. The reason that a transducer operating on transverse waves in the transceiver mode cannot yield information on flaw location depth is stated. It is shown that the detecting signals in the double-scanning mode, obtaining a set of partial images by the SAFT method, and combining them into a final image allow one to solve this problem. To obtain a combined image, the moduli of partial images were summed and their median was calculated. Because of difficulties in determining the velocity of sound and the thickness of a particular object, the coherent summation, which is potentially the most efficient combining technique among the considered ones, does not allow one to obtain a high-quality image. The results of a numerical simulation of the procedure of obtaining images of point defects are presented. The results of model experiments aimed at obtaining an image of a 1-mm-deep groove in a 20-mm-thick metal slab are presented as well. It is shown that measuring echo signals in the double-scanning mode and obtaining images by the DS-M-SAFT method make it possible to determine the flaw location depth.     

热轧螺纹钢在线图象检测技术的研究

本文介绍了利用计算机图象系统对具有复杂轮廓的热轧螺纹钢进行在线检测的技术。提出了用可控频闪光源照明和面阵ＣＣＤ成象方式获取高速运动体的清晰图象，设计了新的图象处理算法，能同时给出热轧螺纹钢的多个参数。实验证明测量具有高的精度，可有效实现螺纹钢的在线检测。     

Determining the kinematic parameters of moving test objects via the results of video recording

Algorithms for processing results of high-speed video recording of the processes of interaction between test objects and barriers are presented. The algorithms make it possible to determine the position of the test object relative to a barrier, as well as the kinematic parameters (the velocity and acceleration) of the points of the test object before, during, and after interaction with the barrier at initial test object collision velocities of up to 120 m/s. The efficiency of the developed algorithms is confirmed by the good coincidence of the results of videogram processing with the results of acceleration measurements by piezoelectric accelerometers.     

动态电容层析成像图像重建算法

提出了融合ECT测量信息和被测对象动态演化信息的新型图像重建模型;基于Tikhonov正则化方法,建立一个同时考虑了ECT测量信息、被测对象动态演化信息、时间与空间约束的新型图像重建目标泛涵,将图像重建问题转化为最优化问题;提出了集成分裂Bregman迭代法优势的新型算法求解该目标泛涵。数值仿真结果表明,所提出的图像重建算法其图像重建质量均优于OIOR算法、STR算法及PLI算法;同时由于所提出的图像重建算法同时考虑了测量数据和重建模型的不精确性,其抵抗测量噪声的能力得以提高。     

Enhancing particle image tracking performance with a sequential Monte Carlo method: The bootstrap filter

One of the common problems of current pattern match and particle image tracking algorithms is the deployment of constant velocity assumption for particle motion between two frames, which would result in serious errors when high velocity gradient flows are measured. To address this issue, a new particle image tracking method—bootstrap filter tracking is proposed. In this new method, a simple nonlinear dynamic model which takes particle acceleration into account is employed and a sequential Monte Carlo method—bootstrap filter is used in conjunction with pattern match algorithm to strengthen the particle image tracking performance. By using the nonlinear system model and bootstrap filter, particle location at next time step can be predicted accurately and the new method is able to measure high velocity gradient flows with better performance than the traditional particle image tracking algorithms. This new method is validated by using numerically generated particle images. Its accuracy in terms of particle image density, out-of-plane displacement and displacement gradient is compared with the Kalman filter tracking (Takehara et al., 2000 [34]) and the Super-PIV (Keane et al., 1995 [30]) methods. The three algorithms are also compared by using a set of real turbulent jet images. The test results demonstrate that the bootstrap filter tracking method is superior than the Kalman filter tracking and the Super-PIV methods for measuring low density, high velocity gradient flows.     

Object selection under random geometric image transformations

The problem of object selection in the presence of random geometric transformations of the observed images due to the movement of the image sensor and random atmosphere fluctuations is solved. An analytical and simplified algorithms of for solution of the problem are developed. Results of statistical investigation of the algorithms are presented.     

Detection of a moving object without a priori information about the velocity vector and image and background intensities

Synthesis and analysis of the maximum likelihood algorithm of detecting an object extended in space are performed. The influence of the absence of a priori information on the velocity vector of object motion and image and background intensities on detection efficiency is studied.     

全向凝视型多目标识别实验研究

针对目前大视场凝视红外多目标处理技术的急需,提出了基于红外鱼眼全向凝视型多目标识别数据处理系统和威胁水平排列算法。给出了全向凝视型多目标识别系统组成、硬件电路和算法实现方法,通过采集序列红外图像,实现不同目标在航路、距离、方向、速度等参数的提取。综合目标运动速度、加速度、方向等因素,对多目标威胁等级排序。通过半实物仿真红外多目标源的配合实验,证明了识别系统的可行性和处理精度。     

基于深度卷积神经网络的目标检测研究综述

作为计算机视觉中的基本视觉识别问题,目标检测在过去的几十年中得到了广泛地研究。目标检测旨在给定图像中找到具有准确定位的特定对象,并为每个对象分配一个对应的标签。近年来,深度卷积神经网络DCNN(Deep Convolutional Neural Networks)凭借其特征学习和迁移学习的强大能力在图像分类方面取得了一系列突破,在目标检测方面,它越来越受到人们的重视。因此,如何将CNN应用于目标检测并获得更好的性能是一项重要的研究。首先回顾和介绍了几类经典的目标检测算法;然后将深度学习算法的产生过程作为切入点,以系统的方式全面概述了各种目标检测方法;最后针对目标检测和深度学习算法面临的重大挑战,讨论了一些未来的方向,以促进深度学习对目标检测的研究。     

基于无标定视觉伺服机器人的运动物体跟踪

介绍机器人视觉伺服的一般原理,然后选取图像的全局特征描述子-图像矩来描述图像的特征信息,推导出基于图像矩特征的图像雅可比矩阵,并按李雅普诺夫稳定性方法推导出模型无关的无定标视觉伺服控制律.最后通过运动物体轨迹跟踪的仿真试验,验证了摄像机固定和eye-in-hand两种结构下算法的正确性和有效性.     

非编码标志点的几何形心提取的研究

在反求工程中,利用毛学测量得到的点云数据只能得到某个视角内的物体,为了得到完整的点云实体,需要对不同视角的点云数据进行拼合。在现有算法中,绝大多数算法都是基于曲面拓扑特征这一前提进行拼合的,这就使得那些曲面平滑,表面拓扑信息不明显的点曲面拼合很难得到较好的效果。之后对于特征不明显的曲面进行拼合,提出了一些基于添加外部特征的方法,标志球,带编码的标志点等。可是这些方法也存在着计算量大的缺陷。基于以上的现状,本文提出了一种基于非编码标志点的曲面拼合方法。直接从图僚中提取标志点的几何形心作为曲面局部匹配的对应关系,具有计算简单,精确度高的特点。     

基于投影特征的自适应图像匹配算法

针对传统图像匹配算法存在的计算量大、匹配速度缓慢的缺点,对基于相关函数法的投影法进行了改进,提出了一种基于投影特征的自适应图像匹配算法。得到投影序列后,选定初始阈值,在后续各点的计算中不断自适应更新,遵循超过当前阈值即停止匹配计算的原则,大大减少了对非匹配点的计算。实验结果表明,该算法抗干扰性强,匹配速度快;模板图像和待匹配图像大小越接近,该算法精度越高,速度优势越明显。     

基于混合图上随机游走的视觉注意显著目标检测模型

目前的视觉显著性检测算法,主要依赖像素间的对比,缺乏从全局角度对显著目标进行分析理解。根据生物视觉注意机制,显著目标通常是显眼、紧凑和完整的,提出一种基于混合图上随机游走的显著目标检测算法,将视觉显著性检测公式化为马尔科夫随机游走问题。首先将输入图像进行分块,利用颜色特征距离和方向的空间分布和方向熵对比分别确定无向图和有向图的边权重,进而得到混合图;然后通过全连通图搜索提取全局特性,突出全局较孤立的区域;同时通过k-regular图搜索提取局部特性,增强局部较均匀的区域;最后结合全局特性和局部特性得到输入图像的显著图,从而确定感兴趣区域位置。实验结果表明,相比于其他两种具有代表性的算法,所提算法检测结果更加准确、合理,证明该方法合理可行。     

Estimation of the velocity vector of a moving object with unknown intensities of its image and background

A maximum likelihood algorithm for estimating the velocity vector of a moving object from its image of unknown intensity against an applicative background whose intensity may be unknown a priori was designed and analyzed.     

凝视模式下的画幅相机两轴像移补偿

通过斜视画幅相机的几何模型,选择载机相关坐标系,建立了计算目标航迹速度到像面坐标系下像移速度的数学模型。通过坐标变换将航迹速度(目标在航迹坐标系下的速度)转换至机体坐标系下的速度矢量,最终至补偿坐标系下的运动矢量;计算视轴长度,得到了扫描镜补偿角速度。最后,阐述了像面旋转机构在像移补偿中的作用,并给出了具体位置角计算公式。提出的采用扫描镜和像面旋转机构相结合的方案实现了凝视工作模式下的画幅相机像移补偿,利用坐标变换计算出的相应量,可为将来画幅相机在该模式下的像移补偿工程应用提供必要的参考。     

最小交叉熵图像重建算法

CT技术通过扫描和图像重建算法,获取被检物场断层图像.由于具有非侵入性、可视化等特点,该技术在工业领域获得广泛应用.为了提高CT系统重建图像的分辨率,提出一种信息扩充策略,并以此为基础采用两种最小交叉熵算法--MAP和SMART,对多相流CT系统进行图像重建.与传统ART算法相比,最小交叉熵算法有效提高了重建图像的分辨率,减少重构图像伪影.仿真和实验结果表明,基于信息扩充的SMART算法不仅改进了重建图像质量,而且提高了实时性.     

微操作系统中微型物体上点的空间坐标的计算方法研究

为操作者提供微型物体的完整的、放大的三维信息,就可以使操作者像在他日常生活中那样精密地控制所使用的工具和被操作对象,解决了在显微镜下操作者的视野和观察方向受到限制以及缺少第三维信息的问题,而为了实时、无遮挡地生成物体的三维虚拟放大影像,需要获得被操作对象不同角度的图像信息并采用简洁、快速的算法以满足实时性要求.介绍在一个人类级微型物体虚拟遥操作系统中根据不同角度拍摄的图像计算物体上点的空间坐标时所采用的一些算法.     

一种超视场零件影像测量仪的研制

针对超过摄像机视场范围的待测零件,必须进行多次拍摄,然后对系列图像进行拼接以形成完整的零件图像。为了实现图像获取和拼接的自动化,设计了X-Y-Z三轴运动控制平台对定焦相机进行精确定位,讨论了仪器的配置、基于归一互相关测度的模板匹配算法的图像拼接技术以及算法优化策略。