基于频域纹理消除的结构性纹理缺陷检测方法

针对金属加工表面等结构纹理表面图像缺陷检测问题,结构纹理的存在会对缺陷（比如划痕）检测带来干扰,该文开展在频率域中消除背景纹理的方法来进行缺陷检测的研究。首先基于傅里叶变换的图像复原技术,空间域图像中的结构性纹理对应傅里叶域中高能频率分量,使用最小二乘法直线拟合操作去除,并将这些能量设置为零,经傅里叶逆变换为空间域图像。在复原的图像中,原始图像中的结构纹理区域将变为近似的均匀灰度级,但其中缺陷部分将被保留下来。再使用统计过程控制来设置阈值的方法就能从复原图像中分离出缺陷。最后在一系列的结构性纹理图像上的实验证实了所提方法可行且有效。     

基于统计特征的轮胎纹理缺陷在线检测

根据轮胎X光纹理缺陷区域灰度及灰度分布异常的特点,研究了一种通过分析统计特征进行在线缺陷检测的方法。在轮胎X光纹理灰度分布模型基础上,采用正则化预处理去除背景噪声,然后进行图像分块,分别计算每块的灰度均值和方差,并采用双线性插值运算形成均值图像和方差图像,再通过二值化实现缺陷检测。实验表明,与人工检测方法进行对比,该方法误判率低,检测精度高,并且运算速度快,能满足在线检测要求。     

基于优化残差网络的复杂纹理表面缺陷检测

产品表面缺陷检测是工业自动化生产的重要环节,准确率是评价自动检测系统可靠性的主要指标。基于复杂纹理表面缺陷检测的特殊性以及对检测方法的实时性、通用性等要求,提出了优化骨干网络并使用迁移学习特征映射构建复杂纹理表面缺陷的检测方法。该方法通过优化残差网络模型并建立仿真数据集的方式进行迁移学习,以解决实际情况中复杂纹理表面产品数据集样本数量少、数据集制作困难、相似问题难以互相兼容等问题。实验结果表明,提出的方法可以准确地检测随机复杂纹理的人造木质板材表面缺陷,平均准确率可达99.6%。现有实验条件下单张人造木质板材的检测时间为305 ms,可以满足在线检测的实时性要求。研究结果可为基于深度学习的复杂纹理表面缺陷检测提供新的思路与理论参考。     

光流动态纹理在土地利用/覆盖变化检测研究中的应用

建议了一种基于光流动态纹理（optical flow dynamic texture）的高分辨率遥感影像变化检测新方法,用一种运动的关系描述地物变化,能够在多时相高分辨率遥感影像中自动获取土地利用和土地覆盖的变化信息。利用光流理论从原理上描述了地物渐变的过程,突破了以往遥感变化检测方法中认为地物发生突变的假设。该方法的流程简单,易于在目前的土地管理、城市规划等需要发现用地变化的系统和软件中使用。该方法考虑到了多时相遥感影像间的时间维度特征,为遥感变化检测提供了更加丰富的信息,进而改善了变化检测方法主要依赖空间维度信息的现状。以光流动态纹理作为变化的基本体现,结合光谱信息共同用于高分辨率遥感影像的支持向量机分类后变化检测,方法顾及了遥感影像时间维度的纹理,相较大多数空间纹理其数据量较小;纹理计算仅需设定一个参数,自动程度较高;可缓解行业中大量人工解译的现状。通过利用中国大庆市杜尔伯特蒙古自治县2011年和2012年QuickBird影像对该方法的有效性进行了评价。深入分析了不同的光流平滑系数α对该方法的影响,以及对地物变化描述效果的影响。实验结果显示,该方法效果理想,总体精度达到87.29%、Kappa系数达到0.850 7,其精度优于单纯利用光谱信息的分类后变化检测方法。     

基于频域变换的MEMS声学薄膜缺陷检测算法

The micro-electro-mechanical system (MEMS) acoustic film has extremely high requirements for tape-out, storage and packaging environments, and its surface defects will affect the quality and performance of MEMS devices. The image defects detection is an effective non-contact optical detection means that can effectively improve the yield rate of MEMS production. However, the periodic structure texture of the MEMS devices surface will interfere with defect detection. A acoustic film defect detection algorithm based on frequency domain transformation was proposed. By calculating the gradient distribution of spectrogram and establishing the Boolean mask, the dominant frequency components corresponding to the periodic structure texture were eliminated. The residual spectrograms were subjected to a Fourier inversion to reconstruct the defect images. The reconstructed images were decomposed by single-layer Haar wavelet to obtain the low-frequency sub-band image and the defect information was extracted by simple threshold segmentation. The defect detection effects of different types of MEMS acoustic film were showed. The experimental results show that it is reasonable to set the zoom constant in the range of 0.7~1.0.     

Recognition algorithm of internal defect images of thermal battery北大核心CSCD

针对目前热电池内部装配缺陷检测效率低、准确度不高的问题,研究了一种可精准分割内部电池堆图像并能够准确识别缺陷种类的方法。首先采用水平、垂直积分投影法对目标电池堆边缘特征进行提取,利用局部自适应对比度增强算法对局部不清晰部分进行细节纹理增强;然后研究了缺陷结构的灰度特性,计算提取出缺陷特征参数;最后使用BP（back propagation）神经网络和CART（classification and regression tree）决策树对特征参数分类识别,并根据分类准确度进行权重分配,将加权融合后的结果作为检测的最终判据。实验结果表明:该方法对2 000个样本的检测准确度达98.9%,为热电池的X射线缺陷检测提供了有效的途径。     

The segmentation of timber defects based on color and the mathematical morphology

Wood has the texture of natural beauty and elegant color so that it can be widely used in the construction and furniture. So people put the special focus on the texture of wood and they are nearly particular about its color. According to the conventional segmentation method, the characteristics of the wood and the actual production process in a short time of the request, this paper presents a segmentation method of wood panel based on color difference and mathematical morphology. In the HSI space, it firstly focus on select morphological edge detection of H component and I component. Instead of considering the small pixel blocks, it uses median filter to clear it to retain accurate edge image. So edge detection is characterized by the H-component of the color model, then region growing is based on edge information, in order to overcome defects of pseudo edge. In this paper, it uses the boundary information to select seed points automatically, then it takes a regional model of the regional boundary for the region growing, and finally it splits out a defective portion of the timber well.     

基于YOLOv3框架的高分辨电镜图像原子峰位置检测

高分辨电镜图像中原子峰位置的检测具有十分重要的现实意义,通过精确定量化原子峰位置可以分析物质在微观尺度上的结构形变、电极化矢量分布等重要信息.近年来深度学习技术在图像目标检测领域取得了巨大突破,这一技术可用在高分辨电镜图像处理上,因为原子位置的检测可以看作是一个目标检测问题.本文利用先进的机器学习方法,通过制作高质量原子图像样本集,使用YOLOv3目标识别框架对原子图像进行自动检测,达到预期效果,实现了深度学习技术在高分辨电镜图像处理领域的应用.该方法的运用有望突破自动处理动态、大量电镜图片的瓶颈问题.     

Thermographic inspection of TBC coated gas turbine blades: Discrimination between coating over-thicknesses and adhesion defects

When performing in-field tests on thermal barrier coatings (TBCs) deposited on gas turbine blades, the main problem encountered is the difficulty to correctly interpret the experimental data. The coating thickness changes, which are not regarded as anomalies, are one of the most common sources of false alarms. They are very challenging to recognize because, unlike stains, eroded areas and optical property variations, they are not detectable in the visible band. Furthermore, their thermal signal is similar to the one produced by an adhesion defect. In this paper, a procedure to reliably discriminate thickness changes and real defects is proposed. The approach is based on the analysis of the apparent effusivity profile. Preliminary results are presented and discussed.     

基于纹理分割的遥感图像的变化探测

提出一种基于纹理分割的遥感图像变化探测算法,利用纹理来反应不同地表形态的影响。对于两幅同一地点、不同时刻的图像,先将前一时刻的遥感图像分割成一系列代表不同地表形态的纹理,假设前一时刻图像的灰度分布在这些分割所得的纹理区域上可以用高斯分布描述,则后一时刻的图像在这些纹理区域上远离高斯分布的像素点便是异常的变化点。由于算法是在分割以后的纹理上分析像素值的统计特性,而不是在滑动的窗口中,因此不受变化大小和形状的限制。实验结果进一步验证了算法具有很好的探测性能。     

基于图谱理论的纹理图像分析

图谱理论作为目前模式识别研究的热门方法之一,广泛应用于聚类与分割,但对其进行图像分析,尤其是复杂的纹理图像分析未见报道。将图谱理论引入到图像分析领域,并结合窗口纹理分析方法构造新的纹理分析算法。首先对图像灰度级进行窗口划分得到不同灰度级下的子图像,然后以各子图像作为图的顶点,子图像间的相似度作为图的边,将原图像解析为一幅带权无向图。利用图谱理论的相关思想对该无向图进行分析,可以从中获得纹理粗糙度等特征,从而完成对原始图像的分析。通过对Brodatz图像库的检索实验证明,该方法优于传统纹理分析算法。     

A Straightforward Immunoassay Applicable to a Wide Range of Antibodies Based on Surface Enhanced Fluorescence

A straightforward immunoassay based on surface enhanced fluorescence (SEF) has been demonstrated using a fluorescent immune substrate and antibody functionalized-silver nanoparticles. Unlike the conventional SEF-based immunoassay, which usually uses the dye-labeled antibodies and the metallic nanostructured-substrates, the presented immune system does not need the antibodies to be labeled with dye molecules. Thus, this immunoassay can be easily applied to the detection of a wide range of target antigens, which is of great importance for its practical application. The experimental results show that this immunoassay has a good specificity as well as the capacity of quantitative detection. Basically, the surface density of the immuno-adsorbed silver nanoparticles increases with the increased amount of target antigens, resulting in a fluorescence enhancement up to around 7 fold. The dose-responsive performance of the immunoassay has been investigated and the limit of detection (LOD) is 1 ng/mL. Due to its simple preparation method and the wide range of detectable antigens, this presented immunoassay is expected to be helpful for extending the SEF-based application.     

基于Gabor小波变换的ICA火灾图像纹理识别算法

针对火灾图像纹理识别问题,提出了基于Gabor小波变换的ICA火灾图像纹理识别算法,并根据火灾图像纹理识别特点进行了优化。首先用不同尺度和方向的Gabor滤波器对待识别图像滤波,得到其特征图像,然后将特征图像转化成特征向量作为ICA的输入,得到基矢量子空间,再将测试图像经过Gabor滤波器的特征向量投影到ICA子空间中得到系数向量作为目标识别特征,最后用支持向量机进行识别。通过与Gabor滤波器法和ICA方法的对比实验,表明该算法可以在火灾纹理图像的识别率上比传统方法提高5%以上,为火灾图像识别提供了一种新思路。     

Excitation of defect modes from the extended photonicband-gap structures of 1D photonic lattices

This paper stuides numerically the model equation in a one dimensional defective photonic lattice by modifying the potential function to a periodic function. It is found that defect modes (DMs) can be regarded as Bloch modes which are excited from the extended photonic band-gap structure at Bloch wave-numbers with k x = 0. The DMs for both positive and negative defects are considered in this method.     

Fractal dimension-bound spatio-temporal analysis of digital mammograms

A new Fractal Dimension-based diagnosis technique for the change detection and time-series analysis of masses in the temporal digital mammogram is presented in this paper. As the digital mammograms are confirmed as a reliable source for the prognosis of breast cancer, the demand for the development of precise computer aided detection techniques is constantly on the increase. This formed the basis for the development of this method using Fractal geometry, which is an efficient mathematical approach that deals with self-similar and irregular geometric objects called fractals. This work comprises of the detection of spatial masses using Fractal Hurst bound enhancement and segmentation of those temporal masses using Fractal Thresholding. The consultant radiologist’s assessment of mass lesions forms the baseline for comparison and validation of the detected masses. Further, this research work performs temporal analysis of mass lesions, detected from the mammograms of the current and the respective prior view using the principle of Fractal Dimension. The precision of Fractal-dimension based temporal texture analysis of malignant masses of digital mammograms subsequently attributes to their characterization.     

纵向带状裂隙形貌的逆时偏移超声成像

传统的工业超声成像方法通常只能确定缺陷的位置与横向尺寸,无法获得缺陷的形貌信息.一些特殊的缺陷,如纵向裂纹,是典型的例子.基于多阵元技术,开展了固体介质缺陷逆时偏移超声成像方法的数值与实验研究.针对铝块平底纵裂纹及内部纵裂纹两种传统方法无法有效成像的缺陷,首先开展了单分量逆时偏移成像方法研究,给出了基于数值仿真的逆时偏移成像结果以及基于多阵元超声成像实验系统实验测试的逆时偏移成像结果.进一步开展了基于多分量位移检测与转换横波分离的逆时偏移成像方法研究,并提出了基于新型多分量激光干涉仪进行检测的思路.数值仿真结果证实了多分量逆时偏移图像重建结果可以克服单分量方式的缺点,得到明显优于单分量检测时的图像.     

Novel quantitative non-destructive testing method for composite structures

A novel QNDT (quantitative non-destructive testing) method is developed that is combined with a phase-shifting shearing speckle and thermograph, and, it aims at the detection of faults such as cracks, voids, delamination and weak areas. The technique is immune to ambient noise and is suitable for measuring the in situ environment. Some different depth defects that would produce deformation differing from other positions could be found with shearing speckle when the sample is loaded, however, a thermograph based on the thermal resistance effect of a defect can detect only those varisized defects embedded deeply in the composite structure by measuring the surface temperature distribution. The resolution is examined for artificial delaminated defects in carbon-fiber composite structures using a phase-shifting shearing speckle and thermograph. The experimental results have demonstrated that the technique is effective for revealing defects in composite structures.     

一种基于规则带特征的图案布匹检测新方法

为了对具有纹理图案的纺织品缺陷进行准确有效的检测,提出了一种新的基于规则带( regular band,RB)特征的纹理图案布匹瑕疵检测方法.首先利用规则带分析提取图像的各子带纹理特征;然后对各子带图像特征求梯度,并通过融合各子带梯度来获得纹理梯度,使其在纹理梯度中能有效地突出纹理区域的边界;最后在此基础上,结合分水岭...     

改进的Faster-RCNN算法在聚乙烯管接头内部缺陷检测中的应用

超声相控阵技术是目前聚乙烯管道热熔接头内部缺陷检测的一种主流方法。提出了基于注意力机制的改进Faster-RCNN目标检测网络用于超声相控阵D扫图聚乙烯管接头内部缺陷检测。针对聚乙烯管道热熔接头内部超声相控阵D扫图小缺陷较多、特征信息容易丢失的问题,将残差网络(ResNet50)与特征金字塔网络(FPN)相结合作为骨干网络,并引入卷积注意力模块(CBAM)自适应细化特征。将SSD网络框架和Faster-RCNN网络框架用于模型训练和测试,使用VGG16、ResNet50、ResNet50+FPN、ACBM+ResNet50+FPN作为骨干网络依次对超声相控阵聚乙烯管道热熔对接接头内部缺陷样本进行训练对比。结果表明,改进的Faster-RCNN网络模型在聚乙烯管接头内部缺陷检测和分类方面有明显改进,对小缺陷的检测性能有了显著的提高。     

Moving target detection approach based on spatio-temporal salient perception

The differences in texture and motion between man-made object and natural scene are the key features for human biological visual system to detect moving object in scenery. The paper proposed a moving target detection approach based on spatio-temporal perception, which is a crucial function of the visual attention mechanism. The spatial feature including edge, orientation, texture and contrast of the image are extracted, and then the corresponding spatial salient map are constructed by fusing the features through difference of Gaussian (DOG) function, which can suppress the common and enhance the difference of local region. Then, the global motion, local motion and relative motion between continuous images are extracted by means of pyramid multi-resolution, and the moving salient map is constructed after the motion difference between moving target and background is confirmed. Finally, the spatio-temporal salient map is constructed by fusing the spatial salient map and the moving salient map through competition strategy, and the moving target could be detected by searching the maximum in the spatio-temporal salient map. Some experiments are included and the results show that the method can accurately detect the moving target in complex background.