CN109886936A - A kind of low contrast defect inspection method and device - Google Patents

A kind of low contrast defect inspection method and device Download PDF

Info

Publication number
CN109886936A
CN109886936A CN201910081135.8A CN201910081135A CN109886936A CN 109886936 A CN109886936 A CN 109886936A CN 201910081135 A CN201910081135 A CN 201910081135A CN 109886936 A CN109886936 A CN 109886936A
Authority
CN
China
Prior art keywords
image
low
defect detection
intensity distribution
auxiliary
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201910081135.8A
Other languages
Chinese (zh)
Other versions
CN109886936B (en
Inventor
田劲东
李聪
田勇
李东
李晓宇
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shenzhen University
Original Assignee
Shenzhen University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shenzhen University filed Critical Shenzhen University
Priority to CN201910081135.8A priority Critical patent/CN109886936B/en
Publication of CN109886936A publication Critical patent/CN109886936A/en
Application granted granted Critical
Publication of CN109886936B publication Critical patent/CN109886936B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Landscapes

  • Image Analysis (AREA)
  • Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)
  • Image Processing (AREA)

Abstract

本发明公开了一种低对比度缺陷检测方法和装置,方法包括步骤:源图像采集、辅助图像生成和图像分类,本发明通过拟合算法生成待检测物体图像的辅助图像,并采用辅助图像和源图像训练深度学习模型,进行图像分类,克服了现有技术中使用人工目视检测方法检测效率低下的技术问题,实现了检测速度快,检测效果好的技术效果,尤其适用于低对比度、灰度变化小的缺陷检测。

The invention discloses a low-contrast defect detection method and device. The method includes the steps of: source image acquisition, auxiliary image generation and image classification. The invention generates an auxiliary image of an image of an object to be detected by a fitting algorithm, and adopts the auxiliary image and source image. Image training deep learning model, image classification, overcoming the technical problem of low detection efficiency using artificial visual detection method in the prior art, realizing the technical effect of fast detection speed and good detection effect, especially suitable for low contrast, grayscale Small variation defect detection.

Description

一种低对比度缺陷检测方法和装置A low-contrast defect detection method and device

技术领域technical field

本发明涉及深度学习技术领域,尤其是一种低对比度缺陷检测方法和装置。The invention relates to the technical field of deep learning, in particular to a low-contrast defect detection method and device.

背景技术Background technique

目前,对于手机摄像头模组检测来说水印缺陷是一类必须进行检测且检测难度大的缺陷。对于此类缺陷,产业中目前主要采用的是人工检测方法,即通过技术工人眼睛观察来判定是否存在水印缺陷。然而,这种人眼检测方法存在着诸多弊端,比如人眼容易疲劳,因而具有不稳定性,无法保证百分之百的检测正确率;此外,检测任务都是重复简单、枯燥、机械的动作,对人的精神是一种折磨,由于人眼的精度、速度上的限制,有些高速、高精度检测根本无法由人工来完成。因此,该领域亟需一种针对极低对比度、纹理特征少、背景不均匀的水印缺陷的自动化精准检测方法。At present, for the detection of mobile phone camera modules, watermark defects are a type of defects that must be detected and are difficult to detect. For such defects, artificial detection methods are currently mainly used in the industry, that is, the presence of watermark defects is determined by the eyes of skilled workers. However, this human eye detection method has many drawbacks. For example, the human eye is easily fatigued, so it is unstable and cannot guarantee a 100% detection accuracy; in addition, the detection tasks are repetitive, simple, boring, and mechanical actions. Due to the limitations of the accuracy and speed of the human eye, some high-speed and high-precision inspections cannot be done manually. Therefore, an automatic and accurate detection method for watermark defects with extremely low contrast, few texture features and uneven background is urgently needed in this field.

发明内容SUMMARY OF THE INVENTION

本发明旨在至少在一定程度上解决相关技术中的技术问题之一。为此,本发明的一个目的是提供一种适用于低对比度、灰度变化小的一种低对比度缺陷检测方法。The present invention aims to solve one of the technical problems in the related art at least to a certain extent. Therefore, an object of the present invention is to provide a low-contrast defect detection method suitable for low-contrast and small gray-scale variation.

为此,本发明的第二个目的是提供一种低对比度、灰度变化小的一种低对比度缺陷检测装置。Therefore, the second object of the present invention is to provide a low-contrast defect detection device with low contrast and small grayscale variation.

本发明所采用的技术方案是:The technical scheme adopted in the present invention is:

第一方面,本发明提供一种低对比度缺陷检测方法,包括如下步骤:In a first aspect, the present invention provides a low-contrast defect detection method, comprising the following steps:

源图像采集:采集待检测物体的图像,所述图像为源图像;Source image collection: collect the image of the object to be detected, and the image is the source image;

辅助图像生成:将所述源图像的强度分布进行拟合,得到待测物体的拟合强度分布函数,根据所述拟合强度分布函数生成辅助图像;Auxiliary image generation: fitting the intensity distribution of the source image to obtain a fitting intensity distribution function of the object to be measured, and generating an auxiliary image according to the fitting intensity distribution function;

图像分类:将所述源图像和所述辅助图像输入深度学习模型,并进行图像分类。Image classification: Input the source image and the auxiliary image into a deep learning model, and perform image classification.

进一步地,所述深度学习模型包括神经网络模型。Further, the deep learning model includes a neural network model.

进一步地,所述图像分类包括步骤:Further, the image classification includes the steps:

将所述源图像和所述辅助图像分别输入到具有激活功能的卷积模块中进行卷积运算提取图像的特征,将所述源图像和所述辅助图像的特征进行合并生成融合特征图像,将所述融合特征图像输入到所述神经网络模型的隐藏层中。The source image and the auxiliary image are respectively input into a convolution module with activation function to perform convolution operation to extract the features of the image, and the features of the source image and the auxiliary image are combined to generate a fusion feature image, and the The fused feature image is input into the hidden layer of the neural network model.

进一步地,还包括步骤:Further, it also includes the steps:

模型训练:采用所述源图像和所述辅助图像进行模型训练。Model training: Model training is performed using the source image and the auxiliary image.

进一步地,所述将所述源图像的强度分布进行拟合,得到待测物体的拟合强度分布函数包括采用高斯函数分布拟合法或多项式拟合法。Further, the fitting of the intensity distribution of the source image to obtain the fitted intensity distribution function of the object to be measured includes using a Gaussian function distribution fitting method or a polynomial fitting method.

进一步地,所述多项式拟合法包括采用二元四次多项式拟合所述强度分布。Further, the polynomial fitting method includes fitting the intensity distribution using a bivariate quartic polynomial.

进一步地,所述神经网络模型包括六个隐藏层和三个全连接层。Further, the neural network model includes six hidden layers and three fully connected layers.

进一步地,所述待检测物体包括玻璃。Further, the object to be detected includes glass.

第二方面,本发明提供一种低对比度缺陷检测装置,包括:源图像采集模块:用于采集待检测物体的图像;In a second aspect, the present invention provides a low-contrast defect detection device, comprising: a source image acquisition module: used to acquire an image of an object to be detected;

辅助图像生成模块,用于将所述源图像的强度分布进行拟合,得到待测物体的拟合强度分布函数,根据所述拟合强度分布函数生成辅助图像;an auxiliary image generation module, configured to fit the intensity distribution of the source image to obtain a fitted intensity distribution function of the object to be measured, and generate an auxiliary image according to the fitted intensity distribution function;

融合特征图像生成模块:用于将所述源图像和所述辅助图像进行融合,生成融合特征图像;Fusion feature image generation module: used to fuse the source image and the auxiliary image to generate a fusion feature image;

图像分类模块:用于将所述融合特征图像输入卷积神经网络训练,进行图像分类。Image classification module: used to input the fused feature image into the convolutional neural network for training to perform image classification.

进一步地,还包括:Further, it also includes:

模型训练模块,用于根据所述源图像和所述辅助图像生成的融合特征图像进行模型训练。The model training module is used for model training according to the fusion feature image generated from the source image and the auxiliary image.

本发明的有益效果是:The beneficial effects of the present invention are:

本发明通过拟合算法生成待检测物体图像的辅助图像,并采用辅助图像和源图像训练深度学习模型,进行图像分类,克服了现有技术中使用人工目视检测方法检测效率低下的技术问题,实现了检测速度快,检测效果好的技术效果,尤其适用于低对比度、灰度变化小的缺陷检测。The invention generates an auxiliary image of the image of the object to be detected by a fitting algorithm, and uses the auxiliary image and the source image to train a deep learning model to perform image classification, and overcomes the technical problem of low detection efficiency by using an artificial visual detection method in the prior art. The technical effect of fast detection speed and good detection effect is realized, and it is especially suitable for defect detection with low contrast and small grayscale change.

附图说明Description of drawings

图1a是存在脏污测试不合格的手机摄像模组的照片;Figure 1a is a photo of a mobile phone camera module that has failed the contamination test;

图1b是测试合格的手机摄像模组的照片;Figure 1b is a photo of a mobile phone camera module that has passed the test;

图2是本发明中一种低对比度缺陷检测方法中一具体实施例的流程图;2 is a flowchart of a specific embodiment of a low-contrast defect detection method in the present invention;

图3a是本发明中一种低对比度缺陷检测方法一具体实施例中的被测手机摄像模组的源图像;3a is a source image of a camera module of a mobile phone under test in a specific embodiment of a low-contrast defect detection method in the present invention;

图3b是本发明中一种低对比度缺陷检测方法一种具体实施例中的被测手机摄像模组的源图像的强度分布图;3b is an intensity distribution diagram of a source image of a camera module of a mobile phone under test in a specific embodiment of a low-contrast defect detection method according to the present invention;

图4a是本发明中一种低对比度缺陷检测方法一种具体实施例中的被测手机摄像模组的辅助图像;4a is an auxiliary image of a camera module of a mobile phone under test in a specific embodiment of a low-contrast defect detection method in the present invention;

图4b是本发明中一种低对比度缺陷检测方法一种具体实施例中的被测手机摄像模组的辅助图像的强度分布图;4b is an intensity distribution diagram of an auxiliary image of a camera module of a mobile phone under test in a specific embodiment of a low-contrast defect detection method according to the present invention;

图5是传统神经网络中图像特征值提取示意图;5 is a schematic diagram of image feature value extraction in a traditional neural network;

图6是本发明中一种低对比度缺陷检测方法一种具体实施例中源图像和辅助图像的特征融合示意图;6 is a schematic diagram of feature fusion of a source image and an auxiliary image in a specific embodiment of a low-contrast defect detection method in the present invention;

图7是本发明中一种低对比度缺陷检测方法一种具体实施例中深度学习模型的结构示意图;7 is a schematic structural diagram of a deep learning model in a specific embodiment of a low-contrast defect detection method according to the present invention;

图8是四种不同深度学习网络模型识别低对比度缺陷的精度对比图;Figure 8 is a comparison chart of the accuracy of four different deep learning network models for identifying low-contrast defects;

图9是本发明中一种低对比度缺陷检测装置一种具体实施例的结构示意图;9 is a schematic structural diagram of a specific embodiment of a low-contrast defect detection device in the present invention;

图10是本发明中一种低对比度缺陷检测装置又一种具体实施例的结构示意图。FIG. 10 is a schematic structural diagram of yet another specific embodiment of a low-contrast defect detection device according to the present invention.

具体实施方式Detailed ways

需要说明的是,在不冲突的情况下,本申请中的实施例及实施例中的特征可以相互组合。It should be noted that the embodiments in the present application and the features of the embodiments may be combined with each other in the case of no conflict.

在缺陷检测领域中,水斑又称水印,这些缺陷通常出现在一些产品经过水洗后的表面,如液晶屏和相机组件上残留的水滴。如图1a和图1b所示,图1a为存在脏污测试不合格的手机摄像模组的照片,1为水印脏污,图1b为测试合格的手机摄像模组的照片,在传统的机器视觉方法中,此类缺陷称之为低对比度缺陷,目前这种缺陷的解决还远远不能令人满意。In the field of defect detection, water spots are also known as watermarks. These defects usually appear on the surface of some products after washing, such as water droplets remaining on LCD screens and camera components. As shown in Figure 1a and Figure 1b, Figure 1a is a photo of a mobile phone camera module that has failed the contamination test, 1 is a watermark dirty, and Figure 1b is a photo of a mobile phone camera module that has passed the test. In the method, such defects are called low-contrast defects, and the solution of such defects is far from satisfactory at present.

如图2所示,图2示出了本发明一种低对比度缺陷检测方法一具体实施例的流程图,包括如下步骤:As shown in FIG. 2, FIG. 2 shows a flowchart of a specific embodiment of a low-contrast defect detection method of the present invention, which includes the following steps:

S1,源图像采集:采集待检测物体的图像,所述图像为源图像;本实施例中测试对象为手机的CMOS摄像模组(CMOS Camera Module),将待测试的手机摄像模组安装到测试版上,获取手机摄像模组图像,如图3a所示,并进行预处理,得到被测手机摄像模组的图像的强度分布图,如图3b所示;S1, source image collection: collect the image of the object to be detected, and the image is the source image; in the present embodiment, the test object is a CMOS camera module of a mobile phone, and the mobile phone camera module to be tested is installed in the test On the screen, obtain the image of the camera module of the mobile phone, as shown in Figure 3a, and perform preprocessing to obtain the intensity distribution map of the image of the camera module of the mobile phone under test, as shown in Figure 3b;

S2,辅助图像生成:将所述源图像的强度分布进行拟合,得到待测物体的拟合强度分布函数,根据所述拟合强度分布函数生成辅助图像;S2, auxiliary image generation: the intensity distribution of the source image is fitted to obtain a fitted intensity distribution function of the object to be measured, and an auxiliary image is generated according to the fitted intensity distribution function;

具体的,强度拟合的方法可采用二维高斯函数拟合和多项式拟合。Specifically, the method of intensity fitting may use two-dimensional Gaussian function fitting and polynomial fitting.

本实施例中,根据图3b中图像强度分布的特点,所述图像强度分布与高斯分布相似,可考虑用高斯分布拟合强度分布。由于镜头的阴影,强度分布不均匀,此外,透镜的光学特性,导致传感器图像区域的边缘接收的光强度比中心接收的光强度小,因此,中心和四个角之间的亮度不一致,透镜本身是凸透镜,根据凸透镜的原理,中心的感光度必须大于外围的感光度。In this embodiment, according to the characteristics of the image intensity distribution in FIG. 3b, the image intensity distribution is similar to the Gaussian distribution, and it may be considered to fit the intensity distribution with the Gaussian distribution. Due to the shadow of the lens, the intensity distribution is not uniform, and in addition, the optical characteristics of the lens, cause the edge of the sensor image area to receive less light intensity than the center, so the brightness between the center and the four corners is not uniform, and the lens itself It is a convex lens. According to the principle of convex lens, the sensitivity of the center must be greater than that of the periphery.

本实施例中采用采用二元四次多项式对图像强度分布进行拟合,与高斯分布相比和三次多项式拟合相比均方误差(RMSE)更小,与采用五次多项式拟合相比计算量更小,且可以避免过度拟合的问题。In this embodiment, the quadratic polynomial is used to fit the image intensity distribution. Compared with the Gaussian distribution and the cubic polynomial fitting, the mean square error (RMSE) is smaller. The amount is smaller and the problem of overfitting can be avoided.

具体的,设图像强度分布为f(x,y),x、y分别为像素坐标,p为拟合参数,将采样点的强度带入下式,采用如最小二乘法进行拟合计算,Specifically, let the image intensity distribution be f(x, y), x and y are the pixel coordinates respectively, p is the fitting parameter, the intensity of the sampling point is brought into the following formula, and the least squares method is used for fitting calculation,

f(x,y)=p00+p10·x+p01·y+p20·x2+p11·x·y+p02·y2+p30·x3+p21·x2·y+p12·x·y2+p03·y3+p40·x4+p31·x3·y+p22*x2*y2+p13·x·y3+p04·y4 f(x,y)=p 00 +p 10 x+p 01 y+p 20 x 2 +p 11 x y+p 02 y 2 +p 30 x 3 +p 21 x 2 y+p 12 x y 2 +p 03 y 3 +p 40 x 4 +p 31 x 3 y+p 22 *x 2 *y 2 +p 13 x y 3 +p 04 ·y 4

计算出的拟合强度的分布示意图如4b所示,根据拟合强度,进行图像还原生成辅助图像的示意图如图4a所示。A schematic diagram of the distribution of the calculated fitting strength is shown in Figure 4b, and a schematic diagram of image restoration to generate an auxiliary image according to the fitting strength is shown in Figure 4a.

S3,图像分类:将所述源图像和所述辅助图像输入深度学习模型,并进行图像分类。S3, image classification: input the source image and the auxiliary image into a deep learning model, and perform image classification.

在传统的深度学习方法中,目标图像是网络的唯一输入,然后用卷积核提取输入图像的特征,如图5所示,图像深度为256*256*3,经过5*5*3的卷积核进行卷积后得到图像的特征值x,然后通过完全连接输出提取的特征图像,如公式:节点i与节点j之间的权值为ωij,节点j的阈值为bj,每个节点的输出值为xj,每个节点的输出值基于上层所有节点、当前节点和上层节点的输出值。第一层上所有节点的权重和当前节点的阈值也由激活函数激活,如公式xj=f(Sj),其中,f为输出层激活函数,一般选用sigmoid函数。In the traditional deep learning method, the target image is the only input to the network, and then the features of the input image are extracted with convolution kernels, as shown in Figure 5, the image depth is 256*256*3, and after a volume of 5*5*3 The feature value x of the image is obtained after the convolution of the product kernel, and then the extracted feature image is output through the full connection, such as the formula: The weight between node i and node j is ωij, the threshold of node j is bj, the output value of each node is x j , and the output value of each node is based on the output values of all nodes in the upper layer, the current node and the upper node. The weights of all nodes on the first layer and the threshold of the current node are also activated by the activation function, such as the formula x j =f(S j ), where f is the activation function of the output layer, and the sigmoid function is generally used.

对于低对比度缺陷检测,有效特征较少,用卷积核直接提取是非常困难的。For low-contrast defect detection, there are few effective features, and it is very difficult to directly extract them with convolution kernels.

在本实施例中,深度学习模型的输入包括两幅图像,一幅是源图像,另一幅是由源图像生成的辅助图像,如图6所示,图6为本实施例中源图像和辅助图像的特征融合示意图,将图像深度均为256*256*3的源图像和辅助图像进行合并,再通过5*5*6的卷积核进行特征提取得到融合特征图。In this embodiment, the input of the deep learning model includes two images, one is the source image, and the other is the auxiliary image generated by the source image, as shown in FIG. 6 , in this embodiment, the source image and Schematic diagram of feature fusion of auxiliary images. The source image and the auxiliary image whose image depths are both 256*256*3 are merged, and then the fusion feature map is obtained by feature extraction with a 5*5*6 convolution kernel.

具体的,如图7所示,将辅助图像和源图像卷积、合并后得到的融合特征图输入到深度学习模型中,本实施例中为神经网络模型,隐藏层包括6个卷积层,每层后都包含一个激活函数,3个完全连接层,平均一次识别时间为0.03s。Specifically, as shown in FIG. 7 , the fusion feature map obtained by convolving and merging the auxiliary image and the source image is input into the deep learning model, which is a neural network model in this embodiment, and the hidden layer includes 6 convolution layers. There is an activation function after each layer, 3 fully connected layers, and the average recognition time is 0.03s.

利用Alexnet、Resnet和VGG三种具有代表性的深度学习网络模型深度学习网络与本实施例中采用的低对比度缺陷检测方法进行了比较,利用测试集每1000次训练迭代获得的不同网络的识别精度如图8所示,横坐标为迭代次数,纵坐标为精度,曲线1为Alexnet网络的分类精度曲线,曲线2为VGG网络的分类精度曲线,曲线3为Resnet网络的分类精度曲线,曲线4为本实施例中低对比度缺陷检测方法的分类精度曲线,从图8中可得知,在经过一定的迭代次数后,本发明采用的低对比度缺陷检测方法的精度最高,可达到95%的分类精度,和现有的深度学习的分类方便比较,可以很好地识别对比度很低的缺陷,具有更好的特征表达能力,更高的识别精度。Using three representative deep learning network models, Alexnet, Resnet and VGG, the deep learning network is compared with the low-contrast defect detection method adopted in this example, and the recognition accuracy of different networks obtained by each 1000 training iterations of the test set is used. As shown in Figure 8, the abscissa is the number of iterations, the ordinate is the accuracy, the curve 1 is the classification accuracy curve of the Alexnet network, the curve 2 is the classification accuracy curve of the VGG network, the curve 3 is the classification accuracy curve of the Resnet network, and the curve 4 is the classification accuracy curve of the Resnet network. The classification accuracy curve of the low-contrast defect detection method in this embodiment can be seen from FIG. 8 , after a certain number of iterations, the low-contrast defect detection method adopted in the present invention has the highest accuracy, which can reach a classification accuracy of 95% , compared with the existing deep learning classification, it can well identify defects with low contrast, and has better feature expression ability and higher recognition accuracy.

本实施例通过源图像生成辅助图像,通过卷积核的方法将源图像和辅助图像融合,获得了新的特征值,由于辅助图像中提供了一些参考信息,因此本实例中的模型结构具有更好的表达能力,具有识别灰度变化小的能力,尤其适用于玻璃等透明材料的检测In this embodiment, the auxiliary image is generated from the source image, and the source image and the auxiliary image are fused by the method of convolution kernel to obtain a new feature value. Since some reference information is provided in the auxiliary image, the model structure in this example has more Good expression ability, with the ability to identify small grayscale changes, especially suitable for the detection of transparent materials such as glass

本实施例中还包括模型训练步骤,模型训练的原理和上述图像识别过程相同,在此不再赘述。This embodiment also includes a model training step, and the principle of the model training is the same as the above-mentioned image recognition process, which is not repeated here.

图9为本发明中一种低对比度缺陷检测装置的一种具体实施例的结构示意图,源图像采集模块:用于采集待检测物体的图像;9 is a schematic structural diagram of a specific embodiment of a low-contrast defect detection device in the present invention, a source image acquisition module: used to acquire an image of an object to be detected;

辅助图像生成模块,用于将所述源图像的强度分布进行拟合,得到待测物体的拟合强度分布函数,根据所述拟合强度分布函数生成辅助图像;an auxiliary image generation module, configured to fit the intensity distribution of the source image to obtain a fitted intensity distribution function of the object to be measured, and generate an auxiliary image according to the fitted intensity distribution function;

融合特征图像生成模块:用于将所述源图像和所述辅助图像进行融合,生成融合特征图像;Fusion feature image generation module: used to fuse the source image and the auxiliary image to generate a fusion feature image;

图像分类模块:用于将所述融合特征图像输入卷积神经网络训练,进行图像分类。Image classification module: used to input the fused feature image into the convolutional neural network for training to perform image classification.

进一步的,如图10所示,还包括:Further, as shown in Figure 10, it also includes:

模型训练模块,用于根据所述源图像和所述辅助图像生成的融合特征图像进行模型训练。The model training module is used for model training according to the fusion feature image generated from the source image and the auxiliary image.

以上是对本发明的较佳实施进行了具体说明,但本发明创造并不限于所述实施例,熟悉本领域的技术人员在不违背本发明精神的前提下还可做出种种的等同变形或替换,这些等同的变形或替换均包含在本申请权利要求所限定的范围内。The above is a specific description of the preferred implementation of the present invention, but the present invention is not limited to the described embodiments, and those skilled in the art can also make various equivalent deformations or replacements on the premise that does not violate the spirit of the present invention , these equivalent modifications or substitutions are all included within the scope defined by the claims of the present application.

Claims (10)

1.一种低对比度缺陷检测方法,其特征在于,包括如下步骤:1. a low-contrast defect detection method, is characterized in that, comprises the steps: 源图像采集:采集待检测物体的图像,所述图像为源图像;Source image collection: collect the image of the object to be detected, and the image is the source image; 辅助图像生成:将所述源图像的强度分布进行拟合,得到待测物体的拟合强度分布函数,根据所述拟合强度分布函数生成辅助图像;Auxiliary image generation: fitting the intensity distribution of the source image to obtain a fitting intensity distribution function of the object to be measured, and generating an auxiliary image according to the fitting intensity distribution function; 图像分类:将所述源图像和所述辅助图像输入深度学习模型,并进行图像分类。Image classification: Input the source image and the auxiliary image into a deep learning model, and perform image classification. 2.根据权利要求1所述的一种低对比度缺陷检测方法,其特征在于,所述深度学习模型包括神经网络模型。2 . The low-contrast defect detection method according to claim 1 , wherein the deep learning model comprises a neural network model. 3 . 3.根据权利要求2所述的一种低对比度缺陷检测方法,其特征在于,所述图像分类包括步骤:3. A kind of low-contrast defect detection method according to claim 2, is characterized in that, described image classification comprises the step: 将所述源图像和所述辅助图像分别输入到具有激活功能的卷积模块中进行卷积运算提取图像的特征,将所述源图像和所述辅助图像的特征进行合并生成融合特征图像,将所述融合特征图像输入到所述神经网络模型的隐藏层中。The source image and the auxiliary image are respectively input into a convolution module with activation function to perform convolution operation to extract the features of the image, and the features of the source image and the auxiliary image are combined to generate a fusion feature image, and the The fused feature image is input into the hidden layer of the neural network model. 4.根据权利要求1所述的一种低对比度缺陷检测方法,其特征在于,还包括步骤:4. a kind of low-contrast defect detection method according to claim 1, is characterized in that, also comprises the step: 模型训练:采用所述源图像和所述辅助图像进行模型训练。Model training: Model training is performed using the source image and the auxiliary image. 5.根据权利要求1所述的一种低对比度缺陷检测方法,其特征在于,所述将所述源图像的强度分布进行拟合,得到待测物体的拟合强度分布函数包括采用高斯函数分布拟合法或多项式拟合法。5 . The low-contrast defect detection method according to claim 1 , wherein the fitting of the intensity distribution of the source image to obtain the fitted intensity distribution function of the object to be tested comprises using a Gaussian function distribution. 6 . fitting method or polynomial fitting method. 6.根据权利要求5所述的一种低对比度缺陷检测方法,其特征在于,所述多项式拟合法包括采用二元四次多项式拟合所述强度分布。6 . The low-contrast defect detection method according to claim 5 , wherein the polynomial fitting method comprises using a quadratic polynomial to fit the intensity distribution. 7 . 7.根据权利要求2所述的一种低对比度缺陷检测方法,其特征在于,所述神经网络模型包括六个隐藏层和三个全连接层。7. A low-contrast defect detection method according to claim 2, wherein the neural network model comprises six hidden layers and three fully connected layers. 8.根据权利要求1至7任一项所述的一种低对比度缺陷检测方法,其特征在于,所述待检测物体包括玻璃。8 . The low-contrast defect detection method according to claim 1 , wherein the object to be detected comprises glass. 9 . 9.一种低对比度缺陷检测装置,其特征在于,包括:9. A low-contrast defect detection device, comprising: 源图像采集模块:用于采集待检测物体的图像;Source image acquisition module: used to acquire the image of the object to be detected; 辅助图像生成模块,用于将所述源图像的强度分布进行拟合,得到待测物体的拟合强度分布函数,根据所述拟合强度分布函数生成辅助图像;an auxiliary image generation module, configured to fit the intensity distribution of the source image to obtain a fitted intensity distribution function of the object to be measured, and generate an auxiliary image according to the fitted intensity distribution function; 融合特征图像生成模块:用于将所述源图像和所述辅助图像进行融合,生成融合特征图像;Fusion feature image generation module: used to fuse the source image and the auxiliary image to generate a fusion feature image; 图像分类模块:用于将所述融合特征图像输入卷积神经网络训练,进行图像分类。Image classification module: used to input the fused feature image into the convolutional neural network for training to perform image classification. 10.根据权利要求9所述的一种低对比度缺陷检测装置,其特征在于,还包括:10. The low-contrast defect detection device according to claim 9, further comprising: 模型训练模块,用于根据所述源图像和所述辅助图像生成的融合特征图像进行模型训练。The model training module is used for model training according to the fusion feature image generated from the source image and the auxiliary image.
CN201910081135.8A 2019-01-28 2019-01-28 A low-contrast defect detection method and device Active CN109886936B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910081135.8A CN109886936B (en) 2019-01-28 2019-01-28 A low-contrast defect detection method and device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910081135.8A CN109886936B (en) 2019-01-28 2019-01-28 A low-contrast defect detection method and device

Publications (2)

Publication Number Publication Date
CN109886936A true CN109886936A (en) 2019-06-14
CN109886936B CN109886936B (en) 2023-05-16

Family

ID=66927107

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910081135.8A Active CN109886936B (en) 2019-01-28 2019-01-28 A low-contrast defect detection method and device

Country Status (1)

Country Link
CN (1) CN109886936B (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110487803A (en) * 2019-08-20 2019-11-22 Oppo(重庆)智能科技有限公司 The defect inspection method and device of infrared light-emitting component
CN111445507A (en) * 2020-04-16 2020-07-24 北京深测科技有限公司 Data processing method for non-visual field imaging
CN114463327A (en) * 2022-04-08 2022-05-10 深圳市睿阳精视科技有限公司 Multi-shooting imaging detection equipment and method for watermark defect of electronic product lining package

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2017223560A1 (en) * 2016-06-24 2017-12-28 Rensselaer Polytechnic Institute Tomographic image reconstruction via machine learning
CN108009628A (en) * 2017-10-30 2018-05-08 杭州电子科技大学 A kind of method for detecting abnormality based on generation confrontation network
CN108053454A (en) * 2017-12-04 2018-05-18 华中科技大学 A kind of graph structure data creation method that confrontation network is generated based on depth convolution
US20180232601A1 (en) * 2017-02-16 2018-08-16 Mitsubishi Electric Research Laboratories, Inc. Deep Active Learning Method for Civil Infrastructure Defect Detection

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2017223560A1 (en) * 2016-06-24 2017-12-28 Rensselaer Polytechnic Institute Tomographic image reconstruction via machine learning
US20180232601A1 (en) * 2017-02-16 2018-08-16 Mitsubishi Electric Research Laboratories, Inc. Deep Active Learning Method for Civil Infrastructure Defect Detection
CN108009628A (en) * 2017-10-30 2018-05-08 杭州电子科技大学 A kind of method for detecting abnormality based on generation confrontation network
CN108053454A (en) * 2017-12-04 2018-05-18 华中科技大学 A kind of graph structure data creation method that confrontation network is generated based on depth convolution

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
JINDONG TIAN .ETC: "Dynamic Phase Measurement Based on Two-Step Spatial Carrier-Frequency Phase-Shifting Interferometry", 《IEEE PHOTONICS JOURNAL》 *

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110487803A (en) * 2019-08-20 2019-11-22 Oppo(重庆)智能科技有限公司 The defect inspection method and device of infrared light-emitting component
CN111445507A (en) * 2020-04-16 2020-07-24 北京深测科技有限公司 Data processing method for non-visual field imaging
CN111445507B (en) * 2020-04-16 2023-07-18 北京深测科技有限公司 Data processing method for non-visual field imaging
CN114463327A (en) * 2022-04-08 2022-05-10 深圳市睿阳精视科技有限公司 Multi-shooting imaging detection equipment and method for watermark defect of electronic product lining package

Also Published As

Publication number Publication date
CN109886936B (en) 2023-05-16

Similar Documents

Publication Publication Date Title
Ren et al. State of the art in defect detection based on machine vision
CN104093016B (en) A kind of dirty detection method of camera module and system
CN107123111B (en) Deep residual error network construction method for mobile phone screen defect detection
CN111932511B (en) A method and system for quality detection of electronic components based on deep learning
WO2024002187A1 (en) Defect detection method, defect detection device, and storage medium
CN106650721A (en) Industrial character identification method based on convolution neural network
CN109886936B (en) A low-contrast defect detection method and device
CN114926407A (en) Steel surface defect detection system based on deep learning
CN111523540A (en) Deep learning-based metal surface defect detection method
KR20200075141A (en) Apparatus and method for testing solarcell module
CN115861170A (en) Surface defect detection method based on improved YOLO V4 algorithm
CN114719749A (en) Method and system for metal surface crack detection and real size measurement based on machine vision
CN115272256A (en) Sub-pixel level sensing optical fiber path Gaussian extraction method and system
CN114486916A (en) Defect detection method of mobile phone glass cover based on machine vision
CN112200790A (en) Cloth defect detection method, device and medium
CN113970560A (en) A three-dimensional defect detection method based on multi-sensor fusion
CN114580559A (en) Speed measuring method based on monocular vision system
CN114494134A (en) Industrial defect detection system based on component point cloud registration detection
CN117333476A (en) A microservice-based visual quality inspection method and experimental system for silicon steel sheets
CN108960413A (en) A kind of depth convolutional neural networks method applied to screw surface defects detection
CN114612468B (en) Equipment external defect detection method based on positive sample
Shang et al. LaserNet: a method of laser stripe center extraction under non-ideal conditions
CN110426395A (en) A kind of solar energy EL cell silicon chip surface inspecting method and device
CN114881998A (en) Workpiece surface defect detection method and system based on deep learning
CN114066810A (en) Method and device for detecting concave-convex point defects of packaging box

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant