A new approach for ME image restoration based on CNN

A new approach for image restoration by cellular neural network (CNN) is developed in this paper. Based on the statistical characteristics of Gibbs image model and the analysis of maximum entropy (ME) image restoration, a reasonable template for binary image restoration is proposed. To process multilevel image, a multi‐layer cellular neural network is employed and an extensive algorithm for multilevel image restoration is proposed. The results of computer simulation prove the effectiveness of this approach and show that we can get the effective template of CNN for some special image questions if we apply the statistical characteristics of Gibbs image model and analyse the physical meaning of the questions. Copyright © 1999 John Wiley & Sons, Ltd.     

CNN-based difference-controlled adaptive non-linear image filters

In this paper, we develop a common cellular neural network framework for various adaptive non-linear filters based on robust statistic and geometry-driven diffusion paradigms. The base models of both approaches are defined as difference-controlled non-linear CNN templates, while the self-adjusting property is ensured by simple analogic (analog and logic) CNN algorithms. Two adaptive strategies are shown for the order statistic class. When applied to the images distorted by impulse noise both give more visually pleasing results with lower-frequency weighted mean square error than the median base model. Generalizing a variational approach we derive the constrained anisotropic diffusion, where the output of the geometry-driven diffusion model is forced to stay close to a pre-defined morphological constraint. We propose a coarse-grid CNN approach that is capable of calculating an acceptable noise-level estimate (proportional to the variance of the Gaussian noise) and controlling the fine-grid anisotropic diffusion models. A combined geometrical–statistical approach has also been developed for filtering both the impulse and additive Gaussian noise while preserving the image structure. We briefly discuss how these methods can be embedded into a more complex algorithm performing edge detection and image segmentation. The design strategies are analysed primarily from VLSI implementation point of view; therefore all non-linear cell interactions of the CNN architecture are reduced to two fundamental non-linearities, to a sigmoid type and a radial basis function. The proposed non-linear characteristics can be approximated with simple piecewise-linear functions of the voltage difference of neighbouring cells. The simplification makes it possible to convert all space-invariant non-linear templates of this study to a standard instruction set of the CNN Universal Machine, where each instruction is coded by at most a dozen analog numbers. Examples and simulation results are given throughout the text using various intensity images. © 1998 John Wiley & Sons, Ltd.     

基于改进U-net和CNN的绝缘子自爆检测方法研究

针对绝缘子自爆故障人工检测效率低,成本高的问题,基于改进U-net和卷积神经网络（CNN）模型,提出一种可有效识别绝缘子自爆故障的双阶段目标检测算法。首先,在语义分割阶段使用改进U-net模型,通过翻倍提高图像分辨率的方法有效提高图像分割精度。其次,在图像分类阶段提出更适合所提问题且有效提高分类准确度的新型CNN模型。最后,使用无人机拍摄的绝缘子图片为实验数据进行实验。实验结果表明所提算法识别精度较高。     

蜣螂优化算法在 Canny 边缘检测算法中的应用

针对传统 Canny 边缘检测需要手动选取阈值以及不能有效提取边缘轮廓的问题,提出了一种基于改进的蜣螂优化算 法(DBO) 来优化 Canny 算子的边缘检测算法。首先通过快速引导滤波代替传统高斯滤波对图像进行保边去噪;其次用4方 向的 Sobel 模板来计算图像的梯度幅值和梯度方向;最后利用蜣螂优化算法优化的二维大津法自适应获取高低阈值。针对蜣 螂优化算法种群多样性不强问题,提出用 tent 映射初始化种群;为了提高算法跳出局部最优的能力,采用精英差分变异策略 对最优蜣螂个体进行变异扰动。实验结果表明,在边缘准确度和连接性上,该算法与传统 Canny 边缘检测算法对比有一定程 度的提升,能够有效提取图像的边缘轮廓,提高了Canny 边缘检测的边缘连接性,具有一定的实用性。     

小波变换在图像检测中的应用

提出了一种基于小波变换的图像边缘检测算法,该算法具有计算复杂度小,运算速度快等特点,仿真结果表明该算法在信／噪比和边缘检测定位精度方面优于传统的边缘检测方法,获取了较满意的图像边缘检测效果。     

基于TensorFlow的高压输电线路异物识别

针对传统异物识别准确率较低的问题,提出一种基于TensorFlow的深度卷积神经网络的异物识别模型。将巡检图像进行图像灰度化和尺寸压缩等预处理,并采用三维块匹配滤波(BM3D)算法进行图像去噪得到实验所需的训练数据。提出基于TensorFlow的深度卷积神经网络框架,通过使用框架中的TensorBoard模块设计深度卷积神经网络模型结构与优选模型参数,并针对ReLU激活函数与特征权重进行理论分析。实验结果表明,经过15次迭代训练后,深度卷积神经网络比传统的支持向量机(SVM)、极限学习机(ELM)和BP神经网络算法具有更强的巡检图像识别能力;与经典的LeNet-5和VGGNet模型以及相关文献中的模型相比,所提模型更具有优越性。     

基于多尺度Gabor滤波器的彩色图像边缘检测

为了更好实现对彩色图像进行边缘提取,并保证算法具有更好的稳定性,文章使用多尺度Gabor滤波器的方法,来提取彩色图像边缘和轮廓.构建了3个尺度、16个方向的Gabor滤波器.首先将彩色图像灰度化,利用多尺度Gabor虚部滤波器提取图像灰度变化信息.通过非极大值抑制,并通过高低阈值获取边缘像素点及其候选边缘,最后利用局部边缘连接获取图像边缘轮廓.并将本算法与常用边缘检测算法进行实验性能比较,实验结果表明:提出的算法既能获得较高的定位准确度,又具有很好的噪声鲁棒性,该算法与常用的Roberts等一系列算法相比,检测效果更好,稳定性更强.     

改进的自适应Canny边缘检测算法

传统Canny边缘检测容易受到环境噪声和光照条件变化等外界因素的影响,导致检测效果不佳,而且需要人工选择参数,算法的灵活性较低。为了提高传统算法的检测效果,使其更好的应用到图像识别等领域中,通过最大类间方差法来获得图像梯度量的最佳分割阈值,动态确定边缘检测的高低阈值参数,能够有效解决传统算法的不足。将改进后的算法用于家用仪表计度轮图像的边缘检测中,与传统算法相比,既去除了噪声干扰,又保留了完整的图像边缘,在取得了良好的检测效果的基础上提高了算法效率。     

Analysis of time-varying cellular neural networks for quadratic global optimization

The algorithm for quadratic global optimization performed by a cellular neural network (CNN) with a slowly varying slope of the output characteristic (see References 1 and 2) is analysed. It is shown that the only CNN which finds the global minimum of a quadratic function for any values of the input parameters is the network composed by only two cells. If the dimension is higher than two, even the CNN described by the simplest one-dimensional space-invariant template Â=[A₁, A₀, A₁], fails to find the global minimum in a subset of the parameter space. Extensive simulations show that the CNN described by the above three-element template works correctly within several parameter ranges; however, if the parameters are chosen according to a random algorithm, the error rate increases with the number of cells. © 1998 John Wiley & Sons, Ltd.     

基于提升小波变换的指针式仪表图像边缘检测

为解决工业中对指针式仪表的自动检定,针对自动检定系统中所涉及的图像边缘检测技术难点,分析了现有传统算法的不足之处,提出一种有效的基于提升小波变换的边缘检测算法。该算法利用提升小波变换尺度的低通作用,避免受高频噪声影响,在传统边缘检测算法的基础上,提取低频轮廓。利用小波系数的方向性,结合方向性边缘检测算子,获得高频边缘信息,最后利用小波重构获得准确清晰的图像边缘,为后续Hough变换准确的提取指针提供了有利保障。实验结果表明,该算法得到的图像边缘较传统Canny算法更清晰准确,无虚假边缘,应用于仪表自动检测系统中是可行的。     

一种基于数学形态学的灰度图像边缘检测方法

图像边缘包含图像的绝大部分有用信息,因而图像识别、匹配等具有很大的研究价值。传统的边缘检测算法去噪能力差、边缘定位能力低,而一般数学形态学算法提取的边缘清晰度不高,且不够平滑,难以满足图像处理的需要。采用序列多尺度结构元素分别对图像进行滤波,再灰度累加取均值,有效滤除噪声的基础上,尽量突出有效的边缘信息,去掉无效的细节信息。通过仿真验证,边缘定位精度高且平滑连贯,效果良好。     

Data driven approach for fault detection and diagnosis of turbine in thermal power plant using Independent Component Analysis (ICA)

In this paper, a statistical signal processing technique, known as Independent Component Analysis (ICA) for fault detection and diagnosis in the real Turbine system (V94.2 model) is suggested. The information of one of MAPNA’s power plants turbine system is utilized at first. In order to reduce the dimensionality of the data set, to identify the essential variables and to choose the most useful variables, PCA approach is applied. Then, the fault sources are diagnosed by ICA technique. The results indicate that suggested approach can distinguish main factors of abnormality, among many diverse parts of a typical turbine system. The presented results will show that suggested approach can avoid false alarms and fault misdiagnosis due to changes in operation conditions and model uncertainty. The presented results show the validity and effectiveness of ICA approach for faults detection and diagnosis in noisy states.     

基于计算机视觉的墙地砖表面缺陷检测

提出了一种基于计算机视觉的墙地砖表面缺陷识别方法,利用改进的分块图像相减匹配和相关算法,对目标图像和模板图像进行分块处理,将对应子区域图像与数据库中相应模块图像进行相减匹配检测,利用相关系数来判断目标图像质量是否满足要求.实验结果表明使用本文方法能够准确地实现墙地砖表面缺陷检测,检测出墙地砖表面平均缺陷数目为4.5个,平均识别率达95.74％,优于直方图统计特征法和灰度共生矩阵法,具有实时、可靠、高效的特点,对墙地砖表面检测具有较大的实用价值.     

An improved algorithm for image crack detection based on percolation model

In a complex background, because of uneven illumination, concrete bubbles, shadows of various shapes, and other noise, traditional crack detection methods based on image processing cannot accurately detect cracks, especially unclear cracks. A crack detection method based on the percolation model fully considers the features of cracks including the characteristics of brightness and length, and therefore can accurately detect cracks in the image. But this method is time consuming, and some noisy areas are detected as crack regions. In order to solve the problems above, we propose an improved algorithm for image crack detection, which includes an accelerated algorithm and a new denoising method based on the percolation model. The accelerated algorithm decreases the number of iterations of percolation processing to reduce the computing time, and the denoising method is based on the characteristic of brightness and the length feature of cracks to remove the noisy regions. Experimental results show that the proposed algorithm can accurately and efficiently detect cracks in the image. © 2014 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.     

一种基于Hough变换的直角角点检测算法

直角角点检测在广泛的图像处理与机器视觉应用领域中具有重要性.提出了一个新的直角角点检测方法.原始图像经边缘检测后得到二值化边缘,将直角角点检测问题分解为一系列的模板匹配问题.应用了Hough变换来高效地解决模板匹配的问题,并在Hough变换中结合了边缘方向信息以减少检测出的虚假直角角点数量.在一个实际的图像数据库上进行了提出的方法与其他数种角点检测算法的对比实验.实验结果表明,与其他方法相比,提出的方法具有最高的直角角点检出率,而其整体性能(包括运行时间、检出率、虚假角点检测率)也具有优势.     

基于阶段聚焦损失和并行增广策略的 遥感图像场景分类

随着深度学习的不断普及,卷积神经网络已成为遥感图像场景分类的主要手段,然而当前的研究主要集中于多网络主干的信息融合以及注意力机制等领域,在提高分类精度的同时也带来更高的计算复杂度。针对上述问题,分别从改进卷积神经网络损失函数和设计新的样本训练策略两个角度出发,在不增加计算复杂度的前提下,提升卷积神经网络的分类性能。首先,在对传统交叉熵和Focal loss损失函数进行分析的基础上,提出一种阶段聚焦损失函数,该损失函数可以在训练阶段对卷积网络进行有侧重的性能挖掘。其次,设计了一种并行样本训练策略,将采用Gridmask算法增广后的样本图像和原始样本图像,分为两路输入卷积网络进行并行训练,进一步提升卷积网络的分类性能。实验结果表明,所提出的算法分别在AID和NWPU-RESISC45两个大规模数据库上取得了96.72%和93.95%的检测精度,可以显著提升遥感图像场景分类的性能。     

多尺度卷积神经网络的图像边缘检测

在受图像拍摄条件、图像内容自身复杂性、图像内容与背景接近程度等多种因素的影响,图像的边缘线检测容易发生漏检、误检。因模型自身设计缺陷或训练样本中边缘像素点与非边缘像素点的不平衡原因,多数算法的图像边缘检测结果普遍存在线条粗、质量较低的问题。提出一种多尺度卷积神经网络模型,由三个分别接受一幅图像的不同尺度输入的子网络结构组成,分别在不同尺度视觉下学习图像的边缘知识。然后按尺度从粗到细对各尺度提取的知识特征进行融合,实现边缘轮廓检测。模型充分利用多尺度技术在图像处理领域的优势,同时引入了自注意力机制以提升卷积特征内部关联性的捕获能力。本文提出了一个新的损失函数,由交叉熵损失函数和L1范数组成,避免训练样本非均衡性对训练模型的影响。使用指标ODS、OIS、AP度量图像边缘检测的质量。在BIPED数据集上测试,三个指标的得分分别为0.845,0.856,0.886。在BSDS500数据集上测试,算法在F-measure指标上得分为0.826。实验结果表明,与其它学习型的算法相比,算法输出图像边缘结果漏检率更低、且质量更高。     

基于模糊集的图像边缘检测算法

边缘检测一直是图像处理中一个热门课题,本文在对经典模糊边缘检测算法思想进行分析的基础上,提出了一种边缘检测算法,该算法通过阈值分区间定义一种模糊隶属函数,可将数字图像转化在等效的模糊特征平面上,然后选用适当的增强算子对图像进行模糊增强,增强后的图像经逆变换后进行边缘提取即可得到所需图像的边缘特征。应用这种算法计算量小,运算速度较快,提取的边缘细致,是一种很有实用价值的边缘提取算法。     

基于图像处理的麻花钻角度测量

为了解决当前麻花钻头修磨设备需要人工干预,而导致修磨质量不高,且修模效率低的问题。采用图像处理相关技术,实现钻头初始角度的自动识别。这对实现麻花钻头修磨的自动化具有较高的工程应用价值。首先运用差影法、Canny边缘检测算法、霍夫变换等算法对拍摄图像进行处理,提取钻头直线刃边缘,然后运用直线拟合技术得到钻头主切削刃直线的直线方程,进而得到钻头的初始角度。经实际测试,该方法可达到工程应用的要求。     

基于压缩感知的电力设备视频图像去噪方法研究

针对电力视频监控图像中存在的噪声,结合压缩感知理论,采用基于过完备字典的稀疏表示方法进行去噪。使用噪声图像训练过完备字典,其中过完备字典的更新使用K-SVD算法,求解稀疏系数使用OMP算法,且根据算法的特点引入了Dice匹配准则来改进正交匹配追踪算法用于求解稀疏系数,最后重构去噪后的图像。Matlab仿真实验表明,对添加了不同标准差的高斯噪声的图像,文中方法具有良好的去噪效果,与目前常用的小波函数相比,能更好的降低图像中的高斯白噪声,并且在字典训练过程中直接使用视频拍摄的带噪声图像,即使没有原始的无噪声图像依然能够完成去噪任务。