WO2017181723A1 - 细纱断裂的检测方法和系统 - Google Patents

细纱断裂的检测方法和系统 Download PDF

Info

Publication number
WO2017181723A1
WO2017181723A1 PCT/CN2016/112887 CN2016112887W WO2017181723A1 WO 2017181723 A1 WO2017181723 A1 WO 2017181723A1 CN 2016112887 W CN2016112887 W CN 2016112887W WO 2017181723 A1 WO2017181723 A1 WO 2017181723A1
Authority
WO
WIPO (PCT)
Prior art keywords
spun yarn
image
probability
target
yarn
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.)
Ceased
Application number
PCT/CN2016/112887
Other languages
English (en)
French (fr)
Inventor
雷延强
罗忠辉
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.)
Guangzhou Shiyuan Electronics Thecnology Co Ltd
Original Assignee
Guangzhou Shiyuan Electronics Thecnology Co Ltd
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 Guangzhou Shiyuan Electronics Thecnology Co Ltd filed Critical Guangzhou Shiyuan Electronics Thecnology Co Ltd
Publication of WO2017181723A1 publication Critical patent/WO2017181723A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • D—TEXTILES; PAPER
    • D01—NATURAL OR MAN-MADE THREADS OR FIBRES; SPINNING
    • D01H—SPINNING OR TWISTING
    • D01H13/00—Other common constructional features, details or accessories
    • D01H13/32—Counting, measuring, recording or registering devices

Definitions

  • the invention relates to the field of textile spun yarn detection, and in particular to a method and a system for detecting spun yarn breakage.
  • the yield and quality of the spun yarn are the key factors affecting the textile, and whether the spun yarn breaks directly affects the yield and quality of the spun yarn.
  • the textile factory in the processing of the spun yarn, it is necessary to ensure that the spun yarn is not broken. If there is a break, the worker needs to be manually connected, that is, the worker needs to conduct a real-time inspection to artificially check whether the spun yarn is broken, and the worker's workload is extremely high. Big, and easy to miss judgment.
  • a method for detecting breakage of a spun yarn includes the following steps:
  • a detection system for spun yarn breakage comprising:
  • a calculating unit configured to perform forward calculation on the image containing the target spun yarn by using the trained convolutional neural network, obtaining a first probability that the target spun yarn belongs to the spun yarn breakage and the second spun yarn belongs to the second probability that the spun yarn is not broken;
  • the determining unit is configured to compare the first probability with the second probability, and if the first probability is greater than or equal to the second probability, determine that the target spun yarn breaks.
  • the image containing the target spun yarn is first obtained, and then the forward calculation is performed by using the trained convolutional neural network to obtain the first probability that the target spun yarn belongs to the spun yarn breakage and belongs to the spun yarn.
  • the second probability of breaking and finally determining whether the spun yarn is broken according to the magnitude of the first probability and the second probability.
  • this scheme saves the worker's inspection time and improves the efficiency.
  • the trained convolutional neural network is used. Judging whether the spun yarn is broken or not greatly improves the accuracy of identifying whether the spun yarn is broken.
  • FIG. 1 is a schematic flow chart of a method for detecting breakage of a spun yarn in one embodiment
  • FIG. 2 is a schematic diagram of a convolutional neural network after training in one of the embodiments
  • Figure 3 is an original image of a spun yarn taken in one of the embodiments
  • Figure 4 is a schematic view showing the positioning of the first area in one of the embodiments
  • Figure 5 is a schematic illustration of a first region after compensation in one of the embodiments
  • Figure 6 is a schematic view of an image of a target spun yarn taken in one of the embodiments
  • Figure 7 is a schematic view showing the structure of a detecting system for breaking a spun yarn in one embodiment
  • Figure 8 is a schematic view showing the structure of a detecting system for breaking a spun yarn in one of the embodiments.
  • the method for detecting the yarn breakage in this embodiment comprises the following steps:
  • Step S101 acquiring an image including the target spun yarn
  • Step S102 Perform a forward calculation on the image containing the target spun yarn by using the trained convolutional neural network, and obtain a first probability that the target spun yarn belongs to the spun yarn breakage and the second spun yarn belongs to the second probability that the spun yarn is not broken;
  • the trained convolutional neural network can operate on the image to be detected, and the target spun yarn therein Identification is made to obtain two probabilities that the spun yarn breaks and the spun yarn is not broken.
  • Step S103 Comparing the first probability with the second probability, and if the first probability is greater than or equal to the second probability, determining that the target spun yarn breaks.
  • the above steps S101, S102 and S103 are processes for performing online testing using the trained convolutional neural network.
  • the image containing the target spun yarn is first obtained, and then the forward calculation is performed by using the trained convolutional neural network to obtain the first probability that the target spun yarn belongs to the spun yarn breakage and the spun yarn is not broken.
  • the second probability finally determining whether the spun yarn is broken according to the magnitude of the first probability and the second probability.
  • this scheme saves the worker's inspection time and improves the efficiency.
  • the trained convolutional neural network is used. Judging whether the spun yarn is broken or not greatly improves the accuracy of identifying whether the spun yarn is broken.
  • the step of acquiring an image comprising the target spun yarn comprises the steps of:
  • the maximum inner truncated rectangle in the first region is determined, and an image in the largest inner truncated rectangle is intercepted in the captured image, and the image in the largest inner truncated rectangle is an image containing the target spun yarn.
  • the image actually taken on the target spun yarn contains a lot of background information, and the effective spun yarn area only occupies a small portion, so the background is set in the background of the target spun yarn, so as to locate the spun yarn, firstly when positioning Positioning the first region, and then determining the largest inner truncated rectangle, intercepting the image in the inner truncated rectangle can standardize the image containing the target spun yarn, and is convenient for the convolutional neural network after training.
  • the step of locating the first region in the captured image further comprises the following steps:
  • the gap of the first region is compensated by the morphological closing operation to obtain the compensated first region.
  • the first region after the first region is positioned in the captured image, the first region may not be a complete region due to the image quality and the image of the main body of the spun yarn, but is divided into a plurality of regions, such as an image of the spun yarn itself.
  • the pixel value is lower than the preset value, and the area of the spun yarn itself is not included in the first area.
  • the morphological closing operation is used for compensation, and the morphological closing operation can connect small gaps, intervals, etc. in the image, including the spun yarn itself.
  • a small area, after compensation, can get the complete first area, that is, the complete spun yarn candidate area.
  • the trained convolutional neural network is obtained by the following steps:
  • the convolutional neural network is trained according to the image sample set of the unbroken yarn and the image sample set of the broken spun yarn, and the trained convolutional neural network is obtained.
  • the convolutional neural network is trained based on the image sample set of the unbroken spun yarn and the image sample set of the broken spun yarn, so that the trained convolutional neural network can recognize whether the target spun yarn is broken.
  • the image sample set of the unbroken yarn and the acquisition method of the various images in the image sample set of the broken spun yarn are similar to the method of acquiring the image containing the target spun yarn, except that the spun yarn is known to be broken when the image sample set is acquired.
  • the convolutional neural network is trained in offline mode.
  • the trained convolutional neural network includes a convolution layer, an activation function layer, a pooling layer, a fully connected layer, and a probability calculation function layer;
  • the convolution operation is performed on the image data containing the target spun yarn through the convolution layer, the convolution operation result is nonlinearly transformed through the activation function layer, and the non-linear transformation result is pooled through the pooling layer, and the pool is connected through the fully connected layer.
  • the result of the operation is linearly transformed, and the linear transformation result is calculated by the probability calculation function layer to obtain the first probability and the second probability.
  • the trained convolutional neural network can obtain the first probability and the second with high precision through the cooperative operation of the convolution layer, the activation function layer, the pooling layer, the fully connected layer and the probability calculation function layer. Probability.
  • the convolutional neural network has a very strong expression ability, which can effectively solve the problem of low precision in multi-classification tasks, and can achieve very high precision.
  • the step of acquiring the image containing the target spun yarn further comprises the following steps:
  • the size of the image containing the target spun yarn is normalized.
  • the convolutional neural network can facilitate the forward calculation processing of the image data, and the recognition process of the spun yarn breakage is accelerated.
  • the method for detecting the breakage of the spun yarn includes three processes of image acquisition, yarn positioning, and convolutional neural network identification, as follows:
  • the slide rail and the camera are set at the position of the spun yarn, and the real-time image of the photographed spun yarn is captured by the camera to provide original image data for the next analysis.
  • the original image is shown in Figure 3.
  • the area a is the imaging area of the camera; the area b is the effective imaging area of the spun yarn, and the b area corresponds to the real area.
  • a pure white backing film can be posted for highlighting the image of the yarn; c is the image of the yarn.
  • the yarn is positioned.
  • the original image contains a lot of background information, and the effective spun yarn area is only a small part, so the spun yarn area positioning is required.
  • T is the preset threshold
  • I(x, y) represents the image pixel amplitude at (x, y)
  • M(x, y) the region where the spun yarn is located
  • the spun yarn region (first region) may be divided into a plurality of regions due to the image quality and the influence of the main body of the spun yarn, as in the b1 and b2 regions in FIG. Therefore, the morphological closing operation is used for compensation, and the morphological closing operation can connect small gaps, intervals, etc. in the image to obtain a complete spun yarn candidate region (the first region after compensation), as shown in the white background in FIG. region.
  • the largest inner truncated rectangle is intercepted as the real spun yarn area, and the corresponding original image data is intercepted on the original image, as shown in FIG. 6, and the intercepted image can also be normalized to the uniform. Size, which can be used as input to the next convolutional neural network.
  • the image of the spun yarn obtained by positioning the spun yarn is input into a convolutional neural network, and forward calculation is performed to identify the probability that the spun yarn image belongs to both fractured and unbroken conditions (the convolutional neural network needs to be trained in advance). If the probability of fracture is greater than or equal to the probability of unbreaking, it indicates that the spun yarn is broken; otherwise, it is not broken.
  • the present invention uses a convolutional neural network as a classifier for determining whether a spun yarn is broken, and only needs to be trained once in practice. It mainly includes two steps of data collection and convolutional neural network training.
  • the main content is to collect a large number of positive samples (unbroken yarn images) and negative samples (fractured spun yarn images).
  • the process of collecting positive and negative samples is similar to the process of obtaining a spun yarn image to be determined whether or not to break.
  • Convolutional neural network training Using the collected positive and negative samples, the convolutional neural network is trained to obtain the trained convolutional neural network.
  • the convolutional neural network used in the present invention may include five Convolutional Layers, and the output of each convolutional layer is subjected to nonlinear transformation by a ReLU (Rectified Linear Units) activation function layer, and then through the pooling layer. (Pooling Layer) pools the nonlinear transformation result, and then connects the result of the pooling operation with a Fully-Connected Layer. Finally, the result of the connection operation is performed by the Softmax function layer (probability calculation function layer). Classification, output classification probability (probability of fracture and non-fracture).
  • the convolutional neural network does not limit the number of layers of each portion, and the spun yarn can be identified, and a deep convolution network having more than five layers can be used.
  • the solution of the invention does not require a large number of modifications to the existing spinning yarn production line, and only needs to set up the slide rail, the camera and the ordinary PC with the convolutional neural network, the implementation cost is low, and compared with the manual inspection In a manner, the invention greatly saves the inspection time of the worker and improves the efficiency.
  • the trained convolutional neural network is used to determine whether the spun yarn is broken, greatly Improves the accuracy of identifying whether the spun yarn is broken.
  • the present invention also provides a spun yarn breakage detecting system, and an embodiment of the spun yarn breakage detecting system of the present invention will be described in detail below.
  • the detection system of the spun yarn breakage in this embodiment includes an acquisition unit 210, a calculation unit 220, and a determination unit 230;
  • An acquiring unit 210 configured to acquire an image that includes the target electronic component
  • the calculating unit 220 is configured to perform forward calculation on the image containing the target spun yarn by using the trained convolutional neural network to obtain a first probability that the target spun yarn belongs to the spun yarn breakage and the second spun yarn belongs to the second probability that the spun yarn is not broken;
  • the determining unit 230 is configured to compare the first probability with the second probability, and if the first probability is greater than or equal to the second probability, determine that the target spun yarn breaks.
  • the acquiring unit 210 acquires a captured image of the target spun yarn, wherein a background is set in the background of the target spun yarn, and the first region is positioned in the photographed image, and the first region is an image pixel image in the region where the back bottom is located.
  • the area whose value is greater than the preset value determines the maximum inner truncated rectangle in the first area, and intercepts the image in the largest inner truncated rectangle in the captured image, and the image in the largest inner truncated rectangle is the image containing the target spun yarn.
  • the obtaining unit 210 is further configured to compensate the gap of the first region by using a morphological closing operation to obtain the compensated first region.
  • the yarn breakage detection system further includes an establishing unit 240 and a training unit 250;
  • the establishing unit 240 is configured to establish an image sample set of the unbroken spun yarn and an image sample set of the broken spun yarn;
  • the training unit 250 is configured to train the convolutional neural network according to the image sample set of the unbroken spun yarn and the image sample set of the broken spun yarn to obtain a trained convolutional neural network.
  • the trained convolutional neural network comprises a convolutional layer, an activation function layer, a pooling layer, a fully connected layer, and a probability calculation function layer;
  • the calculating unit 220 convolutes the image containing the target spun yarn through the convolution layer, performs nonlinear transformation on the convolution operation result through the activation function layer, and performs pooling operation on the non-linear transformation result through the pooling layer, through the fully connected layer
  • the pooling operation result is linearly transformed, and the linear transformation result is calculated by the probability calculation function layer to obtain the first probability and the second probability.
  • the yarn breaking detection system of the present invention corresponds to the yarn breaking method of the present invention in one-to-one correspondence, and the technical features and advantageous effects of the embodiment of the yarn breaking method described above are applicable to the embodiment of the yarn breaking detection system. in.

Landscapes

  • Engineering & Computer Science (AREA)
  • Mechanical Engineering (AREA)
  • Textile Engineering (AREA)
  • Treatment Of Fiber Materials (AREA)
  • Investigating Strength Of Materials By Application Of Mechanical Stress (AREA)
  • Image Analysis (AREA)

Abstract

一种细纱断裂的检测方法和系统，先获取包含目标细纱的图像，再利用训练后的卷积神经网络对其作前向计算，得到目标细纱属于细纱断裂的第一概率和属于细纱未断裂的第二概率，最后根据第一概率和第二概率的大小来确定细纱是否断裂。相比人工巡查的方式，大量节约了工人的巡查时间，提高了效率，同时，相比简单的图像分析技术，采用训练后的卷积神经网络来判断细纱是否断裂，极大地提高了识别细纱是否断裂的精确程度。

Description

细纱断裂的检测方法和系统 技术领域
本发明涉及纺织细纱检测领域，特别是涉及一种细纱断裂的检测方法和系统。
背景技术
在纺织工序中，细纱的产量和质量是影响纺织的关键因素，而细纱是否断裂直接影响到细纱的产量和质量。目前纺织厂内，在细纱的加工过程中，需要保证细纱无断裂，若有断裂，需要工人进行手工连接，也就是工人需要实时巡查，人为地查看细纱是否断裂，这种方式工人的工作量极大，并且容易漏判。
现在还有一种方式来检测细纱断裂，对细纱进行拍摄获得图像，并利用图像分析技术对图像进行二值化操作，并简单计算二值图像中细纱对应的像素点数，在点数小于阈值时，判定细纱断裂，但是由于细纱是柔性的，成像多样化，这种方式的误报率较高。
发明内容
基于此，有必要针对检测细纱断裂的工作量较大，检测精度不高的问题，提供一种细纱断裂的检测方法和系统。
一种细纱断裂的检测方法，包括以下步骤：
获取包含目标细纱的图像；
利用训练后的卷积神经网络对包含目标细纱的图像作前向计算，获得目标细纱属于细纱断裂的第一概率和目标细纱属于细纱未断裂的第二概率；
比较第一概率和第二概率，若第一概率大于或等于第二概率，则判定目标细纱断裂。
一种细纱断裂的检测系统，包括：
获取单元，用于获取包含目标细纱的图像；
计算单元，用于利用训练后的卷积神经网络对包含目标细纱的图像作前向计算，获得目标细纱属于细纱断裂的第一概率和目标细纱属于细纱未断裂的第二概率；
判断单元，用于比较第一概率和第二概率，若第一概率大于或等于第二概率，则判定目标细纱断裂。
根据上述细纱断裂的检测方法和系统，其是先获取包含目标细纱的图像，再利用训练后的卷积神经网络对其作前向计算，得到目标细纱属于细纱断裂的第一概率和属于细纱未断裂的第二概率，最后根据第一概率和第二概率的大小来确定细纱是否断裂。此方案相比人工巡查的方式，大量节约了工人的巡查时间，提高了效率，同时，相比简单的图像分析技术，由于卷积神经网络的强大表达能力，采用训练后的卷积神经网络来判断细纱是否断裂，极大地提高了识别细纱是否断裂的精确程度。
附图说明
图1是其中一个实施例中细纱断裂的检测方法的流程示意图；
图2是其中一个实施例中训练后的卷积神经网络的示意图；
图3是其中一个实施例中拍摄细纱的原始图像；
图4是其中一个实施例中第一区域的定位示意图；
图5是其中一个实施例中补偿后的第一区域的示意图；
图6是其中一个实施例中截取包含目标细纱图像的示意图；
图7是其中一个实施例中细纱断裂的检测系统的结构示意图；
图8是其中一个实施例中细纱断裂的检测系统的结构示意图。
具体实施方式
为使本发明的目的、技术方案及优点更加清楚明白，以下结合附图及实施例，对本发明进行进一步的详细说明。应当理解，此处所描述的具体实施方式仅仅用以解释本发明，并不限定本发明的保护范围。
参见图1所示，为本发明的细纱断裂的检测方法实施例。如图1所示，该实施例中的细纱断裂的检测方法包括以下步骤：
步骤S101：获取包含目标细纱的图像；
步骤S102：利用训练后的卷积神经网络对包含目标细纱的图像作前向计算，获得目标细纱属于细纱断裂的第一概率和目标细纱属于细纱未断裂的第二概率；
在本步骤中，训练后的卷积神经网络可以对待检测图像进行操作，对其中的目标细纱 进行识别，获得细纱断裂和细纱未断裂的两种概率。
步骤S103：比较第一概率和第二概率，若第一概率大于或等于第二概率，则判定目标细纱断裂。
上述步骤S101、S102和S103是使用训练后的卷积神经网络进行在线测试的过程。
本实施方式的细纱断裂的检测方法，是先获取包含目标细纱的图像，再利用训练后的卷积神经网络对其作前向计算，得到目标细纱属于细纱断裂的第一概率和属于细纱未断裂的第二概率，最后根据第一概率和第二概率的大小来确定细纱是否断裂。此方案相比人工巡查的方式，大量节约了工人的巡查时间，提高了效率，同时，相比简单的图像分析技术，由于卷积神经网络的强大表达能力，采用训练后的卷积神经网络来判断细纱是否断裂，极大地提高了识别细纱是否断裂的精确程度。
在其中一个实施例中，获取包含目标细纱的图像的步骤包括以下步骤：
获取目标细纱的拍摄图像，其中，在目标细纱的背景中设置背底；
在拍摄图像中定位第一区域，其中，第一区域为背底所在区域中图像像素幅值大于预设值的区域；
确定第一区域中的最大内截矩形，在拍摄图像中截取最大内截矩形内的图像，最大内截矩形内的图像为包含目标细纱的图像。
在本实施例中，对目标细纱实际拍摄的图像中包含许多背景信息，有效的细纱区域只占少部分，因此在目标细纱的背景中设置背底，以便于对细纱进行定位，在定位时先定位第一区域，再确定其中的最大内截矩形，截取内截矩形内的图像可以使包含目标细纱的图像标准化，便于训练后的卷积神经网络识别。
在其中一个实施例中，在拍摄图像中定位第一区域的步骤之后还包括以下步骤：
采用形态学闭操作对第一区域的缝隙进行补偿，获得补偿后的第一区域。
在本实施例中，在拍摄图像中定位第一区域后，由于成像质量和细纱主体成像的影响，第一区域可能不是完整的一个区域，而是被划分成多个区域，如细纱本身的图像像素值低于预设值，第一区域中就不包括细纱本身的区域，此时采用形态学闭操作来进行补偿，形态学闭操作可以连接图像中细小的缝隙、间隔等，包括细纱本身这种细小的区域，补偿后就可得到完整的第一区域，也就是完整的细纱候选区域。
在其中一个实施例中，训练后的卷积神经网络通过以下步骤获得：
建立未断裂细纱的图像样本集和断裂细纱的图像样本集；
根据未断裂细纱的图像样本集和断裂细纱的图像样本集对卷积神经网络进行训练，获得训练后的卷积神经网络。
在本实施例中，卷积神经网络是根据未断裂细纱的图像样本集和断裂细纱的图像样本集来训练的，如此可以使训练后的卷积神经网络能够识别目标细纱是否断裂。
优选的，未断裂细纱的图像样本集和断裂细纱的图像样本集中的各种图像的获取方法与获取包含目标细纱的图像的方法相类似，只是获取图像样本集时已知细纱是否断裂。卷积神经网络是在离线模式下训练的。
在其中一个实施例中，如图2所示，训练后的卷积神经网络包括卷积层、激活函数层、池化层、全连接层和概率计算函数层；
通过卷积层对包含目标细纱的图像数据进行卷积运算，通过激活函数层对卷积运算结果进行非线性变换，通过池化层对非线性变换结果进行池化操作，通过全连接层对池化操作结果进行线性变换，通过概率计算函数层对线性变换结果进行计算，获得第一概率和第二概率。
在本实施例中，训练后的卷积神经网络通过卷积层、激活函数层、池化层、全连接层和概率计算函数层的协同操作，可以获得精准程度高的第一概率和第二概率。卷积神经网络的表达能力非常强，可有效解决在多分类任务中的精度不高问题，可达到非常高的精准度。
在其中一个实施例中，获取包含目标细纱的图像的步骤之后还包括以下步骤：
对包含目标细纱的图像的大小进行归一化处理。
在本实施例中，进行归一化处理后，可便于卷积神经网络对图像数据的前向计算处理，加快细纱断裂的识别过程。
在一个具体的实施例中，细纱断裂的检测方法包括图像获取、细纱定位、卷积神经网络识别三个过程，具体如下：
第一，图像获取。
在细纱所在位置架设滑轨、摄像头，通过摄像头来抓取拍摄细纱的实时图像，为下一步分析提供原始图像数据。原始的图像示意图如图3。
图3中a区域是摄像头的成像区域；b区域是细纱有效的成像区域，b区域对应的实 际场景中可张贴纯白色背底底片，用于突出细纱成像；c是细纱成像。
第二，细纱定位。
原始图像包含许多背景信息，有效的细纱区域只占少部分，因此需要进行细纱区域定位。
对原始图像采取阈值化操作：
Figure PCTCN2016112887-appb-000001
其中T为预设的阈值，I(x,y)表示(x,y)处的图像像素幅值，根据M(x,y)可粗略得到细纱所在的区域(即第一区域)，如图4所示。
细纱区域(第一区域)由于成像质量和细纱主体成像影响，可能被划分为多个区域，如图4中的b1、b2区域。故接下来采用形态学闭操作进行补偿，形态学闭操作可连接图像中细小的缝隙，间隔等，可得到完整的细纱候选区域(补偿后的第一区域)，如图5中的白底色区域。
在细纱候选区域中，截取最大的内截矩形，作为真正的细纱区域，并在原始图像上，截取相应的原始图像数据，如图6所示，同时还可以将截取的图像归一化到统一尺寸，可作为下一步卷积神经网络的输入。
第三，卷积神经网络识别。
将细纱定位后得到的细纱图像输入到卷积神经网络中，并作前向计算，识别出该细纱图像属于断裂与未断裂两种情况的概率(卷积神经网络需要事先训练)。若断裂概率大于等于未断裂概率，表明该细纱断裂；反之，则未断裂。
本发明采用卷积神经网络作为判定细纱是否断裂的分类器，在实际应用中仅需事先训练一次。主要包括数据收集与卷积神经网络训练两个步骤。
数据收集：主要内容是收集大量的正样本(未断裂细纱图像)与负样本(断裂细纱图像)，正负样本的采集过程同获取待确定断裂与否的细纱图像的步骤相类似。
卷积神经网络训练：利用收集的正负样本，对卷积神经网络进行训练，得到训练后的卷积神经网络。
本发明采用的卷积神经网络可以包括五个卷积层(Convolutional Layer)，每个卷积层的输出都经过ReLU(Rectified Linear Units)激活函数层进行非线性变换，再经过池化层 (Pooling Layer)对非线性变换结果进行池化，再接一个全连接层(Fully-Connected Layer)对池化操作结果进行连接操作，最后通过Softmax函数层(概率计算函数层)对连接操作结果进行分类，输出分类概率(断裂与未断裂的概率)。在本发明中，卷积神经网络对各部分的层数不作限制，可对细纱进行识别即可，层数超过五层的深度卷积网络也可以。
本发明的方案不需要对现有的细纱生产线进行大量的改造，只需架设滑轨、摄像头和具备卷积神经网络的普通PC机即可，方案实施的成本低廉，而且，相比人工巡查的方式，本发明大量节约了工人的巡查时间，提高了效率，相比简单的图像分析技术，由于卷积神经网络的强大表达能力，采用训练后的卷积神经网络来判断细纱是否断裂，极大地提高了识别细纱是否断裂的精确程度。
根据上述细纱断裂的检测方法，本发明还提供一种细纱断裂的检测系统，以下就本发明的细纱断裂的检测系统的实施例进行详细说明。
参见图7所示，为本发明的细纱断裂的检测系统的实施例。该实施例中的细纱断裂的检测系统，包括获取单元210，计算单元220，判断单元230；
获取单元210，用于获取包含目标电子元件的图像；
计算单元220，用于利用训练后的卷积神经网络对包含目标细纱的图像作前向计算，获得目标细纱属于细纱断裂的第一概率和目标细纱属于细纱未断裂的第二概率；
判断单元230，用于比较第一概率和第二概率，若第一概率大于或等于第二概率，则判定目标细纱断裂。
在其中一个实施例中，获取单元210获取目标细纱的拍摄图像，其中，在目标细纱的背景中设置背底，在拍摄图像中定位第一区域，第一区域为背底所在区域中图像像素幅值大于预设值的区域，确定第一区域中的最大内截矩形，在拍摄图像中截取最大内截矩形内的图像，最大内截矩形内的图像为包含目标细纱的图像。
在其中一个实施例中，获取单元210还用于采用形态学闭操作对第一区域的缝隙进行补偿，获得补偿后的第一区域。
在其中一个实施例中，如图8所示，细纱断裂的检测系统还包括建立单元240和训练单元250；
建立单元240用于建立未断裂细纱的图像样本集和断裂细纱的图像样本集；
训练单元250用于根据未断裂细纱的图像样本集和断裂细纱的图像样本集对卷积神经网络进行训练，获得训练后的卷积神经网络。
在其中一个实施例中，训练后的卷积神经网络包括卷积层、激活函数层、池化层、全连接层和概率计算函数层；
计算单元220通过卷积层对包含目标细纱的图像进行卷积运算，通过激活函数层对卷积运算结果进行非线性变换，通过池化层对非线性变换结果进行池化操作，通过全连接层对池化操作结果进行线性变换，通过概率计算函数层对线性变换结果进行计算，获得第一概率和第二概率。
本发明的细纱断裂的检测系统与本发明的细纱断裂的检测方法一一对应，在上述细纱断裂的检测方法的实施例阐述的技术特征及其有益效果均适用于细纱断裂的检测系统的实施例中。
以上所述实施例的各技术特征可以进行任意的组合，为使描述简洁，未对上述实施例中的各个技术特征所有可能的组合都进行描述，然而，只要这些技术特征的组合不存在矛盾，都应当认为是本说明书记载的范围。
以上所述实施例仅表达了本发明的几种实施方式，其描述较为具体和详细，但并不能因此而理解为对发明专利范围的限制。应当指出的是，对于本领域的普通技术人员来说，在不脱离本发明构思的前提下，还可以做出若干变形和改进，这些都属于本发明的保护范围。因此，本发明专利的保护范围应以所附权利要求为准。

Claims (10)

  1. 一种细纱断裂的检测方法，其特征在于，包括以下步骤：
    获取包含目标细纱的图像；
    利用训练后的卷积神经网络对所述包含目标细纱的图像作前向计算，获得所述目标细纱属于细纱断裂的第一概率和所述目标细纱属于细纱未断裂的第二概率；
    比较所述第一概率和所述第二概率，若所述第一概率大于或等于所述第二概率，则判定所述目标细纱断裂。
  2. 根据权利要求1所述的细纱断裂的检测方法，其特征在于，所述获取包含目标细纱的图像的步骤包括以下步骤：
    获取所述目标细纱的拍摄图像，其中，在所述目标细纱的背景中设置背底；
    在所述拍摄图像中定位第一区域，其中，所述第一区域为所述背底所在区域中图像像素幅值大于预设值的区域；
    确定所述第一区域中的最大内截矩形，在所述拍摄图像中截取所述最大内截矩形内的图像，所述最大内截矩形内的图像为所述包含目标细纱的图像。
  3. 根据权利要求2所述的细纱断裂的检测方法，其特征在于，所述在所述拍摄图像中定位第一区域的步骤之后还包括以下步骤：
    采用形态学闭操作对所述第一区域的缝隙进行补偿，获得补偿后的第一区域。
  4. 根据权利要求1所述的细纱断裂的检测方法，其特征在于，所述训练后的卷积神经网络通过以下步骤获得：
    建立未断裂细纱的图像样本集和断裂细纱的图像样本集；
    根据所述未断裂细纱的图像样本集和所述断裂细纱的图像样本集对所述卷积神经网络进行训练，获得所述训练后的卷积神经网络。
  5. 根据权利要求1所述的细纱断裂的检测方法，其特征在于，所述训练后的卷积神经网络包括卷积层、激活函数层、池化层、全连接层和概率计算函数层；
    所述利用训练后的卷积神经网络对所述包含目标细纱的图像作前向计算的步骤包括以下步骤：
    通过所述卷积层对所述包含目标细纱的图像进行卷积运算，通过所述激活函数层对卷积运算结果进行非线性变换，通过所述池化层对非线性变换结果进行池化操作，通过所述 全连接层对池化操作结果进行线性变换，通过概率计算函数层对线性变换结果进行计算，获得所述第一概率和所述第二概率。
  6. 一种细纱断裂的检测系统，其特征在于，包括：
    获取单元，用于获取包含目标细纱的图像；
    计算单元，用于利用训练后的卷积神经网络对所述包含目标细纱的图像作前向计算，获得所述目标细纱属于细纱断裂的第一概率和所述目标细纱属于细纱未断裂的第二概率；
    判断单元，用于比较所述第一概率和所述第二概率，若所述第一概率大于或等于所述第二概率，则判定所述目标细纱断裂。
  7. 根据权利要求6所述的细纱断裂的检测系统，其特征在于，所述获取单元获取所述目标细纱的拍摄图像，其中，在所述目标细纱的背景中设置背底，在所述拍摄图像中定位第一区域，所述第一区域为所述背底所在区域中图像像素幅值大于预设值的区域，确定所述第一区域中的最大内截矩形，在所述拍摄图像中截取所述最大内截矩形内的图像，所述最大内截矩形内的图像为所述包含目标细纱的图像。
  8. 根据权利要求7所述的细纱断裂的检测系统，其特征在于，所述获取单元还用于采用形态学闭操作对所述第一区域的缝隙进行补偿，获得补偿后的第一区域。
  9. 根据权利要求6所述的细纱断裂的检测系统，其特征在于，还包括建立单元和训练单元；
    所述建立单元用于建立未断裂细纱的图像样本集和断裂细纱的图像样本集；
    所述训练单元用于根据所述未断裂细纱的图像样本集和所述断裂细纱的图像样本集对所述卷积神经网络进行训练，获得所述训练后的卷积神经网络。
  10. 根据权利要求6所述的细纱断裂的检测系统，其特征在于，所述训练后的卷积神经网络包括卷积层、激活函数层、池化层、全连接层和概率计算函数层；
    所述计算单元通过所述卷积层对所述包含目标细纱的图像进行卷积运算，通过所述激活函数层对卷积运算结果进行非线性变换，通过所述池化层对非线性变换结果进行池化操作，通过所述全连接层对池化操作结果进行线性变换，通过概率计算函数层对线性变换结果进行计算，获得所述第一概率和所述第二概率。
PCT/CN2016/112887 2016-04-20 2016-12-29 细纱断裂的检测方法和系统 Ceased WO2017181723A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201610251096.8 2016-04-20
CN201610251096.8A CN105821538B (zh) 2016-04-20 2016-04-20 细纱断裂的检测方法和系统

Publications (1)

Publication Number Publication Date
WO2017181723A1 true WO2017181723A1 (zh) 2017-10-26

Family

ID=56527156

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2016/112887 Ceased WO2017181723A1 (zh) 2016-04-20 2016-12-29 细纱断裂的检测方法和系统

Country Status (2)

Country Link
CN (1) CN105821538B (zh)
WO (1) WO2017181723A1 (zh)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116815365A (zh) * 2023-08-28 2023-09-29 江苏恒力化纤股份有限公司 一种环锭纺细纱机断纱自动检测方法
CN117779255A (zh) * 2023-12-27 2024-03-29 无锡千帆竞桨科技有限公司 一种基于图像分析的环锭纺断纱检测方法及系统
CN119090886A (zh) * 2024-11-08 2024-12-06 山东国泰大成科技有限公司 一种碳纤维断丝智能检测方法及系统

Families Citing this family (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105821538B (zh) * 2016-04-20 2018-07-17 广州视源电子科技股份有限公司 细纱断裂的检测方法和系统
CN107123117B (zh) * 2017-04-26 2020-10-20 广东工业大学 一种基于深度学习的ic引脚焊点质量检测方法及装置
CN108428247B (zh) * 2018-02-27 2020-06-12 广州视源电子科技股份有限公司 焊锡点方向的检测方法和系统
CN110175659B (zh) * 2018-12-04 2022-04-19 深圳码隆科技有限公司 纺丝机故障监测方法
CN111080611A (zh) * 2019-12-12 2020-04-28 哈尔滨市科佳通用机电股份有限公司 一种铁路货车摇枕弹簧折断故障图像识别方法
CN111235709A (zh) * 2020-03-18 2020-06-05 东华大学 一种基于机器视觉的环锭纺细纱条干在线检测系统
CN118373278B (zh) * 2024-04-18 2024-10-18 重庆市特种设备检测研究院(重庆市特种设备事故应急调查处理中心) 一种电梯曳引系统中曳引绳的监测方法、系统及存储介质
CN119399416A (zh) * 2024-11-06 2025-02-07 湖北天门纺织机械股份有限公司 一种基于视觉识别并条机喂入条断缺条的检测方法

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103544506A (zh) * 2013-10-12 2014-01-29 Tcl集团股份有限公司 一种基于卷积神经网络的图像分类方法和装置
CN104850858A (zh) * 2015-05-15 2015-08-19 华中科技大学 一种注塑制品缺陷检测识别方法
CN105095919A (zh) * 2015-09-08 2015-11-25 北京百度网讯科技有限公司 图像识别方法和装置
CN105096279A (zh) * 2015-09-23 2015-11-25 成都融创智谷科技有限公司 一种基于卷积神经网络的数字图像处理方法
JP2016033806A (ja) * 2014-05-15 2016-03-10 株式会社リコー 画像処理装置、画像処理方法及びプログラム
CN105821538A (zh) * 2016-04-20 2016-08-03 广州视源电子科技股份有限公司 细纱断裂的检测方法和系统

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH0782620A (ja) * 1993-09-10 1995-03-28 Murata Mach Ltd 延伸仮撚機
CN104573731B (zh) * 2015-02-06 2018-03-23 厦门大学 基于卷积神经网络的快速目标检测方法
CN105469400B (zh) * 2015-11-23 2019-02-26 广州视源电子科技股份有限公司 电子元件极性方向的快速识别、标注的方法和系统
CN105386174B (zh) * 2015-12-03 2017-10-27 江南大学 一种基于图像技术的赛络纺断股检测装置及检测算法

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103544506A (zh) * 2013-10-12 2014-01-29 Tcl集团股份有限公司 一种基于卷积神经网络的图像分类方法和装置
JP2016033806A (ja) * 2014-05-15 2016-03-10 株式会社リコー 画像処理装置、画像処理方法及びプログラム
CN104850858A (zh) * 2015-05-15 2015-08-19 华中科技大学 一种注塑制品缺陷检测识别方法
CN105095919A (zh) * 2015-09-08 2015-11-25 北京百度网讯科技有限公司 图像识别方法和装置
CN105096279A (zh) * 2015-09-23 2015-11-25 成都融创智谷科技有限公司 一种基于卷积神经网络的数字图像处理方法
CN105821538A (zh) * 2016-04-20 2016-08-03 广州视源电子科技股份有限公司 细纱断裂的检测方法和系统

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116815365A (zh) * 2023-08-28 2023-09-29 江苏恒力化纤股份有限公司 一种环锭纺细纱机断纱自动检测方法
CN116815365B (zh) * 2023-08-28 2023-11-24 江苏恒力化纤股份有限公司 一种环锭纺细纱机断纱自动检测方法
CN117779255A (zh) * 2023-12-27 2024-03-29 无锡千帆竞桨科技有限公司 一种基于图像分析的环锭纺断纱检测方法及系统
CN119090886A (zh) * 2024-11-08 2024-12-06 山东国泰大成科技有限公司 一种碳纤维断丝智能检测方法及系统
CN119090886B (zh) * 2024-11-08 2025-02-11 山东国泰大成科技有限公司 一种碳纤维断丝智能检测方法及系统

Also Published As

Publication number Publication date
CN105821538A (zh) 2016-08-03
CN105821538B (zh) 2018-07-17

Similar Documents

Publication Publication Date Title
WO2017181723A1 (zh) 细纱断裂的检测方法和系统
CN104237121B (zh) 一种精子活力性能检测方法
CN103984951B (zh) 一种磁粉检测缺陷自动识别方法及系统
TWI707137B (zh) 智能產線監測系統及監測方法
CN109087286A (zh) 一种基于计算机图像处理和模式识别的检测方法及应用
CN109946304A (zh) 基于特征匹配的零件表面缺陷在线检测系统及检测方法
CN111257329A (zh) 一种智能手机摄像头缺陷检测方法以及检测系统
CN112419261B (zh) 具有异常点去除功能的视觉采集方法及装置
CN104424633B (zh) 一种视频对比度异常检测方法及装置
CN114723668A (zh) 一种基于面阵相机的阴极铜质量检测方法及系统
CN114332058A (zh) 基于神经网络的血清质量识别方法、装置、设备及介质
CN111126112B (zh) 候选区域确定方法和装置
CN112967224A (zh) 一种基于人工智能的电子电路板检测系统、方法及介质
CN115482462A (zh) 基于深度学习的混凝土裂缝识别系统及方法
CN112784835A (zh) 圆形印章的真实性识别方法、装置、电子设备及存储介质
Fu et al. Research on image-based detection and recognition technologies for cracks on rail surface
CN107607205A (zh) 线束颜色顺序检测系统和方法
CN112800975A (zh) 一种基于图像处理的安检通道内行为识别方法
CN118430025A (zh) 安监领域智能多模态检测方法、装置及计算机设备
CN105631458A (zh) 一种电子元件样本标注方法及装置
CN112329664B (zh) 一种原核期胚胎原核数量的评估方法
CN114078271B (zh) 阈值确定方法、目标人员识别方法、装置、设备及介质
FR3126253B1 (fr) Procédé pour normaliser la variabilité d’une image, application de ce procédé à la détection d’anomalie et système d’inspection visuelle implémentant cette détection
CN111783775A (zh) 一种图像采集方法、装置、设备和计算机可读存储介质
CN112733722A (zh) 姿态识别方法、装置、系统及计算机可读存储介质

Legal Events

Date Code Title Description
NENP Non-entry into the national phase

Ref country code: DE

121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 16899302

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 16899302

Country of ref document: EP

Kind code of ref document: A1