WO2017088552A1 - 电子元件极性的识别方法和系统、标注方法和系统 - Google Patents

电子元件极性的识别方法和系统、标注方法和系统 Download PDF

Info

Publication number
WO2017088552A1
WO2017088552A1 PCT/CN2016/098227 CN2016098227W WO2017088552A1 WO 2017088552 A1 WO2017088552 A1 WO 2017088552A1 CN 2016098227 W CN2016098227 W CN 2016098227W WO 2017088552 A1 WO2017088552 A1 WO 2017088552A1
Authority
WO
WIPO (PCT)
Prior art keywords
electronic component
image
convolutional neural
polarity direction
neural network
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/CN2016/098227
Other languages
English (en)
French (fr)
Inventor
杨铭
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 WO2017088552A1 publication Critical patent/WO2017088552A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0004—Industrial image inspection
    • G06T7/0006—Industrial image inspection using a design-rule based approach
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/09—Supervised learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/096—Transfer learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30108—Industrial image inspection

Definitions

  • the present invention relates to the field of automatic optical detection, and in particular to a method and system for identifying the polarity of electronic components, an annotation method and system.
  • a method for identifying a polarity direction of an electronic component includes the following steps:
  • the probability distribution of various polar direction categories of the target electronic components belonging to various electronic components is obtained;
  • the polarity direction category with the highest probability is selected as the polarity direction category of the target electronic component.
  • the image including the electronic component is first acquired, and then the trained convolutional neural network is used for forward calculation, and the probability distribution of the polarity direction category of the electronic component is obtained, and the polarity with the highest probability is selected.
  • the direction category is used as the polarity direction category of electronic components.
  • the trained convolutional neural network is obtained by the following steps:
  • the pre-trained convolutional neural network is adjusted and further trained to obtain the trained convolutional neural network.
  • the step of establishing a sample set of various polarity directions of various types of electronic components includes the following steps:
  • the precise position of the components, the corresponding electronic components are adjusted according to the precise position of the corresponding electronic components, so that the corresponding electronic components are located at the center of the image of various electronic components, and image sample sets of various polar directions of various electronic components are obtained.
  • the step of pre-training the convolutional neural network through the preset external data set comprises the steps of:
  • the convolutional neural network is pre-trained with an external data set, so that the convolutional neural network learns the general image features of each level, and obtains the initial parameter values of the pre-trained convolutional neural network.
  • the pre-trained convolutional neural network is performed according to the image sample set
  • the steps to adjust and further train include the following steps:
  • the number of nodes of the last layer of the pre-trained convolutional neural network in the initial parameter values is adjusted to the number of categories of various polarity directions of various electronic components, and the adjusted convolutional neural network is used. For further training.
  • a method for marking the polarity direction of an electronic component includes the following steps:
  • the polarity information of the target electronic component is marked in the panel file, and the panel file is used to store various attribute information of the electronic component.
  • the above-mentioned method for labeling the polarity direction of the electronic component can be applied to the AOI board production, and the polarity direction of the components on the PCB board is automatically and accurately marked, thereby improving the automation level of the panel production and improving the efficiency and accuracy of the panel production.
  • An identification system for the polarity direction of an electronic component comprising the following units:
  • An acquiring unit configured to acquire an image including a target electronic component
  • a calculation unit for performing forward calculation on the image using the trained convolutional neural network to obtain a probability distribution of various polarity direction categories of the target electronic components belonging to various electronic components;
  • the selection unit is configured to select the polarity direction category with the highest probability as the polarity direction category of the target electronic component.
  • the image including the electronic component is first acquired, and then the trained convolutional neural network is used for forward calculation, and the probability distribution of the polarity direction category of the electronic component is obtained, and the polarity with the highest probability is selected.
  • the direction category is used as the polarity direction category of electronic components.
  • the identification system of the polarity direction of the electronic component further includes an establishing unit, a pre-training unit, and an adjusting unit;
  • the pre-training unit is configured to pre-train the convolutional neural network by a preset external data set, wherein the external data set includes a plurality of pre-labeled natural images of the plurality of categories;
  • the adjusting unit is configured to adjust and further train the pre-trained convolutional neural network according to the image sample set to obtain a trained convolutional neural network.
  • An annotation system for the polarity direction of an electronic component comprising an annotation unit and an identification system for the polarity direction of the electronic component, wherein the labeling unit is configured to mark the polarity direction information of the target electronic component in the panel file, and the panel file is used for saving Various attribute information of electronic components.
  • the above-mentioned electronic component polarity direction labeling system can be applied to AOI board production, and the polarity direction of the components on the PCB board is automatically and accurately marked, thereby improving the automation level of the panel production and improving the efficiency and accuracy of the panel production.
  • FIG. 1 is a flow chart showing a method for identifying a polarity direction of an electronic component in one embodiment
  • FIG. 2 is a partial flow chart showing a method for identifying a polarity direction of an electronic component in one embodiment
  • FIG. 3 is a schematic structural view of an identification system of a polarity direction of an electronic component in one embodiment
  • FIG. 4 is a schematic structural view of an identification system of a polarity direction of an electronic component in one embodiment
  • FIG. 5 is a schematic structural diagram of an identification system of a polarity direction of an electronic component in one embodiment
  • Figure 6 is a partial structural schematic view of an identification system for the polarity direction of an electronic component in one embodiment
  • FIG. 7 is a schematic structural view of an identification system of a polarity direction of an electronic component in one embodiment
  • Figure 8 is a block diagram showing the structure of the polarity direction of the electronic component in one embodiment.
  • the method for identifying the polarity direction of an electronic component in this embodiment includes the following steps:
  • Step S101 Acquire an image including the target electronic component
  • the image containing the electronic component may be a PCB that needs to identify the polarity of the electronic component.
  • the board image can also be other images that need to identify the polarity of the electronic components.
  • Step S102 Perform forward calculation on the image by using the trained convolutional neural network to obtain a probability distribution of various polarity direction categories of the target electronic components belonging to various electronic components;
  • various polarity direction categories refer to various polarity directions of various electronic components, including various electronic components, and the obtained probability distribution can be applied to various electronic components; the convolutional neural network after training The image can be manipulated to identify the polarity direction of the target electronic component therein to obtain a probability distribution.
  • Step S103 selecting the polarity direction category with the highest probability as the polarity direction category of the target electronic component
  • the above steps S101, S102 and S103 are processes for performing online testing using the trained convolutional neural network.
  • the method for identifying the polarity direction of the electronic component is to first acquire an image including the target electronic component, and then perform forward calculation using the trained convolutional neural network to obtain the target electronic component belonging to various electronic components.
  • the probability distribution of various polar direction categories selects the polarity direction category with the highest probability as the polarity direction category of the target electronic component.
  • This scheme uses a convolutional neural network to automatically and accurately identify the polarity of electronic components.
  • the direction, and does not target a specific electronic component structure, is applicable to various electronic components with polarity, and realizes polarity direction recognition of electronic components across categories, and has wide applicability.
  • the target electronic components in the image are adjusted by the template image of the target electronic component, so that the trained convolutional neural network can more easily identify the target electronic component.
  • the convolutional neural network is trained in an offline mode, as shown in Figure 2, the steps are as follows:
  • Step S104 establishing image sample sets of various polarity directions of various electronic components
  • Step S105 Pre-training the convolutional neural network by using a preset external data set, where the external data set includes a plurality of pre-labeled natural images of a plurality of categories;
  • Step S106 According to the image sample set, the pre-trained convolutional neural network is adjusted and further trained to obtain a trained convolutional neural network.
  • the convolutional neural network can be offline trained to process the polarity direction information of the electronic components in the image, so that the trained convolutional neural network can be used in the online test.
  • step S104 includes the following steps:
  • the precise position of the components, the corresponding electronic components are adjusted according to the precise position of the corresponding electronic components, so that the corresponding electronic components are located at the center of the image of various electronic components, and image sample sets of various polar directions of various electronic components are obtained.
  • the image sample sets obtained by the above methods are directly taken from the PCB card image, and are images of actual electronic components, which are more instructive.
  • the camera can be set on the PCB board production line, and the PCB board images of different models can be collected in batches, and the board tracking technology is used to avoid repeatedly shooting a certain PCB board, so that each type of PCB board includes a plurality of image samples, each image sample corresponding to a certain PCB card of a certain model; when collecting the PCB card image, the corresponding PCB template image needs to be saved;
  • the position of the PCB card image in multiple image samples may be offset. It is necessary to use the corresponding template PCB template map as a reference, perform image registration for each image sample, and use the electronic component position information of the model PCB card (from The panel file or manual labeling automatically intercepts the electronic component image and automatically labels the electronic component image according to the electronic component category information (from the panel file or manual labeling);
  • the precise position of the electronic components in the image sample is further obtained by matching the electronic components in the PCB template image, and the electronic component images are aligned and adjusted to ensure that the electronic components are located at the center of the image, and image samples of various polar directions of the electronic components are obtained. set.
  • step S105 includes the following steps:
  • the convolutional neural network is pre-trained with an external data set, so that the convolutional neural network learns the general image features of each level, and obtains the initial parameter values of the pre-trained convolutional neural network.
  • the ImageNet data set is used as an external data set for pre-training.
  • the present invention refers to ImageNet, an external dataset pre-training convolutional neural network.
  • ImageNet an external dataset pre-training convolutional neural network.
  • the ImageNet dataset is not an electronic component dataset, it contains more than 22,000 categories of 15 million labeled natural images for pre-training. The general image features of each level are learned, and the initial parameters of the convolutional neural network are obtained.
  • the convolutional neural network has a very strong expression ability, which can effectively solve the problem of low precision in multi-classification tasks. Even in the cross-class electronic component polarity direction recognition task, a very high accuracy can be achieved.
  • step S106 includes the following steps:
  • the number of nodes of the last layer of the pre-trained convolutional neural network in the initial parameter values is adjusted to the number of categories of various polarity directions of various electronic components, and the adjusted convolutional neural network is used. For further training.
  • the adjusted convolutional neural network is similar in structure to the unadjusted convolutional neural network, the difference being that the last layer, originally 1000 nodes, assumes that all the polarities of all electronic components have a total of N classes, then Change this layer to N nodes.
  • the convolutional neural network is further trained to obtain a trained convolutional neural network. Network.
  • the following steps are further included:
  • the above steps are a process of detecting the polarity direction, and it is possible to judge whether or not the electronic component is correctly mounted.
  • the image of the target electronic component acquired by the online test is a PCB card image
  • the corresponding PCB template is matched and matched to obtain the precise position of the target electronic component, and the target electronic component in the image is aligned and adjusted. To ensure that the target electronic component is at the center of the image.
  • the corresponding PCB template is used for the matching, and the PCB template may be a PCB template corresponding to the image sample set image, or may be a PCB template different from the PCB template corresponding to the image sample set image, if In the former, the PCB template corresponding to the image sample set image has been saved during offline training, and can be directly used. In the latter case, in addition to obtaining the PCB board image, the online PCB template needs to obtain the corresponding PCB template from the outside to achieve the subsequent matching match.
  • the implementation of the solution can identify the polarity direction of all electronic components in the acquired PCB card image.
  • step S101 and step S104 each comprise the following steps:
  • the size of the adjusted image is normalized.
  • step S101 after the normalization process, the forward calculation processing of the image by the convolutional neural network can be facilitated, and the identification process of the electronic component is accelerated.
  • step S104 after the normalization process, the further training of the convolutional neural network can be facilitated, so that the convolutional neural network can more accurately learn the multi-layer features of the image samples.
  • the method for identifying the polarity direction of the electronic component of the present invention uses a convolutional neural network to automatically and accurately identify the polarity direction of the electronic component, and is not specific to a specific electronic component structure, and is applicable to various polar-polarized electrons.
  • Component as long as the image sample set used to train the convolutional neural network covers multiple species
  • the electronic components of the class can realize the polarity direction identification of the electronic components across the category, and the applicability is wider; whether it is the offline training phase or the online test phase, the manual intervention is minimized, and the cost is not dependent on the special hardware mechanism.
  • the implementation of this method does not need to collect a large number of image samples of the polarity direction of the electronic components in advance, and therefore is also suitable for the scene where the image acquisition of electronic components is difficult.
  • the invention also provides a method for marking the polarity direction of an electronic component, comprising the following steps:
  • the polarity information of the target electronic component is marked in the panel file, and the panel file is used to store various attribute information of the electronic component.
  • the above-mentioned method for labeling the polarity direction of the electronic component can be applied to the AOI board production, and the polarity direction of the components on the PCB board is automatically and accurately marked, thereby improving the automation level of the panel production and improving the efficiency and accuracy of the panel production.
  • the present invention also provides an identification system for the polarity direction of the electronic component.
  • an embodiment of the identification system for the polarity direction of the electronic component of the present invention will be described in detail.
  • the identification system of the polarity direction of the electronic component in this embodiment includes the acquisition unit 200, the calculation unit 210, and the selection unit 220 in FIG. 3;
  • An obtaining unit 200 configured to acquire an image including a target electronic component
  • the calculating unit 210 is configured to perform forward calculation on the image by using the trained convolutional neural network to obtain a probability distribution of various polarity direction categories of the target electronic components belonging to various electronic components;
  • the selecting unit 220 is configured to select a polarity direction category with the highest probability as the polarity direction category of the target electronic component.
  • the identification system of the polarity direction of the electronic component further includes a pre-processing unit 230;
  • the pre-processing unit 230 is configured to acquire a template image of the target electronic component after acquiring the image of the target electronic component, and match the target electronic component in the image with the template image of the target electronic component to obtain the target electronic component in the image.
  • the precise position of the target electronic component in the image is adjusted according to the precise position, so that the target electronic component in the image is located at the center of the image, and the adjusted image is used for forward calculation by the trained convolutional neural network.
  • the identification system of the polarity direction of the electronic component further includes an establishing unit 240, a pre-training unit 250, and an adjusting unit 260;
  • the establishing unit 240 is configured to establish an image sample set of various polarity directions of various electronic components
  • the pre-training unit 250 is configured to pre-train the convolutional neural network by a preset external data set, wherein the external data set includes a plurality of pre-labeled natural images of the plurality of categories;
  • the adjusting unit 260 is configured to adjust and further train the pre-trained convolutional neural network according to the image sample set to obtain a trained convolutional neural network.
  • the establishing unit 240 includes a registration unit 241 and an intercept processing unit 242;
  • the registration unit 241 is configured to obtain a PCB card image and a PCB template image, and perform position registration on the PCB card image by using the PCB template image as a reference;
  • the intercepting processing unit 242 is configured to intercept the electronic component image on the PCB card image after the position registration, and match various electronic components in the electronic component images by various electronic components in the PCB template image to obtain various electronic components.
  • the precise position of the corresponding electronic components in the image, the corresponding electronic components are adjusted according to the precise position of the corresponding electronic components, so that the corresponding electronic components are located at the center of the image of various electronic components, and image samples of various polar directions of various electronic components are obtained. set.
  • the pre-training unit 250 is configured to pre-train the convolutional neural network with an external data set, so that the convolutional neural network learns the general image features of each level, and obtains the initial of the pre-trained convolutional neural network. Parameter value.
  • the adjusting unit 260 is configured to adjust the number of nodes of the last layer of the pre-trained convolutional neural network in the initial parameter values to various polar directions of various electronic components according to the image sample set. The number of categories and further training of the adjusted convolutional neural network.
  • the identification system of the polarity direction of the electronic component further includes The alarm unit 270 is configured to provide error alarm information if the polarity direction category with the highest probability of selection is different from the preset polarity direction category of the electronic component.
  • the identification system of the polarity direction of the electronic component of the present invention corresponds one-to-one with the method of identifying the polarity direction of the electronic component of the present invention, and the technical features and advantageous effects of the embodiment of the method for identifying the polarity direction of the electronic component are applicable.
  • the identification system for the polarity of the electronic components corresponds one-to-one with the method of identifying the polarity direction of the electronic component of the present invention, and the technical features and advantageous effects of the embodiment of the method for identifying the polarity direction of the electronic component are applicable.
  • the present invention also provides an annotation system for the polarity direction of an electronic component, as shown in FIG. 8, comprising an identification unit 280 and the above-described identification system for the polarity direction of the electronic component, and the labeling unit 280 is configured to be based on the polarity direction of the electronic component.
  • the polar direction of the target electronic component determined by the identification method, the polarity direction information of the electronic component is marked in the panel file, and the panel file is used to store various attribute information of the electronic component.
  • the above-mentioned electronic component polarity direction labeling system can be applied to AOI board production, and the polarity direction of the components on the PCB board is automatically and accurately marked, thereby improving the automation level of the panel production and improving the efficiency and accuracy of the panel production.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Quality & Reliability (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)

Abstract

一种电子元件极性方向的识别方法和系统、标注方法和系统。该识别方法包括先获取包含电子元件的图像（S101），再利用训练后的卷积神经网络对其作前向计算，得到电子元件的极性方向类别的概率分布（S102），选择其中概率最大的极性方向类别作为电子元件的极性方向类别（S103）。该方法使用了卷积神经网络，能自动精准地识别电子元件的极性方向，而且并不针对特定的电子元件结构，适用于各种带极性的电子元件，实现跨类别的电子元件的极性方向识别，适用性较广。根据确定的目标电子元件的极性方向，可以应用于板式制作，对电子元件进行极性方向标注，从而提升板式制作的自动化水平，改善板式制作的效率与准确性。

Description

电子元件极性的识别方法和系统、标注方法和系统 技术领域
本发明涉及自动光学检测领域，特别是涉及一种电子元件极性的识别方法和系统、标注方法和系统。
背景技术
目前，对于PCB板，需要在对其上的电子元件进行极性方向判断。现在对于电子元件极性方向判断，主要有两种方法：一是基于元件结构特征的极性方向判断，这种方法是基于特定的元件结构特征，通常只针对某类元件设计，适用性受限，另外，部分结构特征在不同环境下(如光照不同、拍摄角度不同、噪声干扰等)不稳定，导致判断准确率较低；二是基于正反模板匹配的极性方向判断，只要针对每种元件都设定一个模板，但这样可能会因多类别模板混用而导致准确率降低，而且元件只要外观稍有不同就得引入新的模板，可扩展性受限，计算时间也会因模板库的增大而增加。
发明内容
基于此，有必要针对电子元件极性方向判断准确率低、适用性有限的问题，提供一种电子元件极性方向的识别方法和系统、标注方法和系统。
一种电子元件极性方向的识别方法，包括以下步骤：
获取包含目标电子元件的图像；
使用训练后的卷积神经网络对图像作前向计算，得到目标电子元件属于各类电子元件的各种极性方向类别的概率分布；
选取概率最大的极性方向类别作为目标电子元件的极性方向类别。
根据上述识别方法，其是先获取包括电子元件的图像，再利用训练后的卷积神经网络对其作前向计算，得到电子元件的极性方向类别的概率分布，选择其中概率最大的极性方向类别作为电子元件的极性方向类别，此方案中使用了卷积神经网络，可以自动精准地识别电子元件的极性方向，而且并不针对特定 的电子元件结构，适用于各种带极性的电子元件，实现跨类别的电子元件的极性方向识别，适用性较广。
在其中一个实施例中，在获取包含目标电子元件的图像的步骤之后，包括以下步骤：
获取目标电子元件的模板图，以目标电子元件的模板图对图像中的目标电子元件进行匹配，获得图像中的目标电子元件的精确位置，根据精确位置对图像中的目标电子元件进行调整，使图像中的目标电子元件位于图像的中心，调整后的图像供训练后的卷积神经网络作前向计算。
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：
建立各类电子元件的各种极性方向的图像样本集；
通过预设的外部数据集预训练卷积神经网络，其中外部数据集包含多个类别的多张已预先标注的自然图像；
根据图像样本集，对预训练后的卷积神经网络进行调整并作进一步训练，得到训练后的卷积神经网络。
在其中一个实施例中，建立各类电子元件的各种极性方向的样本集的步骤包括以下步骤：
获取PCB板卡图像和PCB模板图，并以PCB模板图为参考，对PCB板卡图像进行位置配准；
截取位置配准后PCB板卡图像上的各类电子元件图像，以PCB模板图中的各类电子元件对各类电子元件图像中相应的电子元件进行匹配，获得各类电子元件图像中相应电子元件的精确位置，根据相应电子元件的精确位置对相应电子元件进行调整，使相应电子元件位于各类电子元件图像的中心，获得各类电子元件的各种极性方向的图像样本集。
在其中一个实施例中，通过预设的外部数据集预训练卷积神经网络的步骤包括以下步骤：
用外部数据集对卷积神经网络进行预训练，使卷积神经网络学习各层次的通用图像特征，获得预训练后的卷积神经网络的初始参数值。
在其中一个实施例中，根据图像样本集，对预训练后的卷积神经网络进行 调整并作进一步训练的步骤包括以下步骤：
根据图像样本集，将初始参数值中的预训练后的卷积神经网络的最后一层节点数调整为各种电子元件的各种极性方向的类别数，并对调整后的卷积神经网络作进一步训练。
一种电子元件极性方向的标注方法，包括以下步骤：
根据上述电子元件极性方向的识别方法确定的目标电子元件的极性方向，在板式文件中标注目标电子元件的极性方向信息，板式文件用于保存电子元件的各种属性信息。
上述电子元件极性方向的标注方法可以应用于AOI板式制作，自动、精准地对PCB板卡上的元件进行极性方向标注，从而提升板式制作的自动化水平，改善板式制作的效率与准确性。
一种电子元件极性方向的识别系统，包括以下单元：
获取单元，用于获取包含目标电子元件的图像；
计算单元，用于使用训练后的卷积神经网络对图像作前向计算，得到目标电子元件属于各类电子元件的各种极性方向类别的概率分布；
选取单元，用于选取概率最大的极性方向类别作为目标电子元件的极性方向类别。
根据上述识别系统，其是先获取包括电子元件的图像，再利用训练后的卷积神经网络对其作前向计算，得到电子元件的极性方向类别的概率分布，选择其中概率最大的极性方向类别作为电子元件的极性方向类别，此方案中使用了卷积神经网络，可以自动精准地识别电子元件的极性方向，而且并不针对特定的电子元件结构，适用于各种带极性的电子元件，实现跨类别的电子元件的极性方向识别，适用性较广。
在其中一个实施例中，电子元件极性方向的识别系统还包括建立单元、预训练单元以及调整单元；
建立单元用于建立各类电子元件的各种极性方向的图像样本集；
预训练单元用于通过预设的外部数据集预训练卷积神经网络，其中外部数据集包含多个类别的多张已预先标注的自然图像；
调整单元用于根据图像样本集，对预训练后的卷积神经网络进行调整并作进一步训练，得到训练后的卷积神经网络。
一种电子元件极性方向的标注系统，包括标注单元和上述电子元件极性方向的识别系统，其中，标注单元用于在板式文件中标注目标电子元件的极性方向信息，板式文件用于保存电子元件的各种属性信息。
上述电子元件极性方向的标注系统可以应用于AOI板式制作，自动、精准地对PCB板卡上的元件进行极性方向标注，从而提升板式制作的自动化水平，改善板式制作的效率与准确性。
附图说明
图1是其中一个实施例中电子元件极性方向的识别方法的流程示意图；
图2是其中一个实施例中电子元件极性方向的识别方法的部分流程示意图；
图3是其中一个实施例中电子元件极性方向的识别系统的结构示意图；
图4是其中一个实施例中电子元件极性方向的识别系统的结构示意图；
图5是其中一个实施例中电子元件极性方向的识别系统的结构示意图；
图6是其中一个实施例中电子元件极性方向的识别系统的部分结构示意图；
图7是其中一个实施例中电子元件极性方向的识别系统的结构示意图；
图8是其中一个实施例中电子元件极性方向的标注系统的结构示意图。
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，为本发明的电子元件极性方向的识别方法实施例。如图1所示，该实施例中的电子元件极性方向的识别方法包括以下步骤：
步骤S101：获取包含目标电子元件的图像；
在本步骤中，包含电子元件的图像可以是需要识别电子元件极性的PCB 板卡图像，也可以是其他需要识别电子元件极性的图像。
步骤S102：使用训练后的卷积神经网络对图像作前向计算，得到目标电子元件属于各类电子元件的各种极性方向类别的概率分布；
在本步骤中，各种极性方向类别是指各种电子元件的各种极性方向，包括了多种电子元件，获取的概率分布可以适用于各种电子元件；训练后的卷积神经网络可以对图像进行操作，对其中的目标电子元件的极性方向进行识别，获得概率分布。
步骤S103：选取概率最大的极性方向类别作为目标电子元件的极性方向类别；
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。
本实施方式所述的电子元件极性方向的识别方法，是先获取包含目标电子元件的图像，再利用训练后的卷积神经网络对其作前向计算，得到目标电子元件属于各类电子元件的各种极性方向类别的概率分布，选择其中概率最大的极性方向类别作为目标电子元件的极性方向类别，此方案中使用了卷积神经网络，可以自动精准地识别电子元件的极性方向，而且并不针对特定的电子元件结构，适用于各种带极性的电子元件，实现跨类别的电子元件的极性方向识别，适用性较广。
在其中一个实施例中，在获取包含目标电子元件的图像的步骤之后，包括以下步骤：
获取目标电子元件的模板图，以目标电子元件的模板图对图像中的目标电子元件进行匹配，获得图像中的目标电子元件的精确位置，根据精确位置对图像中的目标电子元件进行调整，使目标电子元件位于图像的中心。
用目标电子元件的模板图对图像中的目标电子元件进行调整，使得训练后的卷积神经网络更加容易识别目标电子元件。
在其中一个实施例中，卷积神经网络是在离线模式下训练的，如图2所示，步骤如下：
步骤S104：建立各类电子元件的各种极性方向的图像样本集；
步骤S105：通过预设的外部数据集预训练卷积神经网络，其中外部数据集包含多个类别的多张已预先标注的自然图像；
步骤S106：根据图像样本集，对预训练后的卷积神经网络进行调整并作进一步训练，得到训练后的卷积神经网络。
通过上述步骤可以对卷积神经网络进行离线训练，使其能够处理图像中电子元件的极性方向信息，从而可以在在线测试中使用该训练后的卷积神经网络。
在其中一个实施例中，步骤S104包括以下步骤：
获得PCB板卡图像和PCB模板图，并以PCB模板图为参考，对PCB板卡图像进行位置配准；
截取位置配准后PCB板卡图像上的各类电子元件图像，以PCB模板图中的各类电子元件对各类电子元件图像中相应的电子元件进行匹配，获得各类电子元件图像中相应电子元件的精确位置，根据相应电子元件的精确位置对相应电子元件进行调整，使相应电子元件位于各类电子元件图像的中心，获得各类电子元件的各种极性方向的图像样本集。
通过上述方式获取的图像样本集均是直接从PCB板卡图像上截取的，都是实际电子元件的图像，更具有指导性。
优选的，可以在PCB板卡生产线上架设摄像头，并批量采集不同型号的PCB板卡图像，并以板卡跟踪技术避免重复拍摄某一张PCB板卡，这样每种型号的PCB板卡均包含多个图像样本，每个图像样本对应某一型号的某张PCB板卡；在采集PCB板卡图像时需要保存相应的PCB模板图；
多个图像样本中的PCB板卡图像位置可能偏移，需要以相应型号的PCB模板图作为参考，对每个图像样本进行图像配准，并利用该型号PCB板卡的电子元件位置信息(来自板式文件或人工标注)自动截取电子元件图片，并根据电子元件类别信息(来自板式文件或人工标注)对电子元件图片进行自动标注；
以PCB模板图中的电子元件匹配进一步求得图像样本中电子元件的精确位置，对电子元件图像进行对齐调整，以保证电子元件位于图像中心位置，获得电子元件的各种极性方向的图像样本集。
在其中一个实施例中，步骤S105包括以下步骤：
用外部数据集对卷积神经网络进行预训练，使卷积神经网络学习各层次的通用图像特征，获得预训练后的卷积神经网络的初始参数值。
优选的，将ImageNet数据集作为外部数据集用于预训练。
目前，现有的电子元件极性方向的识别方法往往利用的是电子元件的颜色信息等低层图像特征，这种传统方法对于许多应用场景都过于简单，鲁棒性较低，适用范围也受限，效果较差；而本方案中使用卷积神经网络学习多层特征表示方式(包括低层、中层、高层图像特征，而不仅仅是低层图像特征)，并综合这样的多层特征去识别电子元件极性方向，从而大大提升识别精度，也扩展了适用范围。
在实际场景中，建立的各类电子元件的各种极性方向的图像样本集相对于表达能力较强的卷积神经网络来说仍然较少。因此，本发明引用ImageNet这一外部数据集预训练卷积神经网络，虽然ImageNet数据集并非电子元件数据集，但其包含超过22000个类别的1500万张带标注的自然图像，用于预训练可学习出各层次的通用图像特征，得到较好的卷积神经网络初始参数值。
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，即便在跨类别的电子元件极性方向识别任务中，亦可达到非常高的准确率。
在其中一个实施例中，步骤S106包括以下步骤：
根据图像样本集，将初始参数值中的预训练后的卷积神经网络的最后一层节点数调整为各种电子元件的各种极性方向的类别数，并对调整后的卷积神经网络作进一步训练。
优选的，经过调整的卷积神经网络与未调整的卷积神经网络结构类似，不同点在于最后一层，原先是1000个节点，假设所有电子元件的各种极性方向共N类，那么现将该层改为N个节点。
该步骤针对得到的图像样本集，基于前序步骤所得到的卷积神经网络初始参数值，调整卷积神经网络的最后一层节点数，并根据图像样本集，采用迁移学习策略，对调整后的卷积神经网络作进一步训练，得到训练后的卷积神经网 络。
在其中一个实施例中，在选取概率最大的极性方向类别作为所述电子元件的极性方向类别的步骤之后，还包括以下步骤：
若选取的概率最大的极性方向类别与预设的电子元件极性方向类别不同，则给出错误报警信息。
上述步骤是对极性方向的检测过程，可以对电子元件是否安装正确进行判断。
在其中一个实施例中，在线测试获取的包含目标电子元件的图像为PCB板卡图像，以相应的PCB模板来对照匹配，获得目标电子元件的精确位置，对图像中的目标电子元件进行对齐调整，以保证目标电子元件位于图像的中心位置。
在本实施例中，对照匹配时使用的是相应的PCB模板，该PCB模板可以是图像样品集中图像对应的PCB模板，也可以是与图像样品集中图像对应的PCB模板不同的PCB模板，如果是前者，在离线训练时已经保存了与图像样品集中图像对应的PCB模板，可以直接获取使用。如果是后者，在线测试时除了获取PCB板卡图像时还需要从外界获取相应的PCB模板，以实现之后的对照匹配。实施本方案可以对获取的PCB板卡图像中的所有电子元件进行极性方向的识别。
在其中一个实施例中，步骤S101和步骤S104均包括以下步骤：
对调整后的图像的大小进行归一化处理。
在步骤S101中，进行归一化处理后，可便于卷积神经网络对图像的前向计算处理，加快电子元件的识别过程。
在步骤S104中，进行归一化处理后，可便于对卷积神经网络的进一步训练，使卷积神经网络对图像样本的多层特征的学习更为准确。
本发明的电子元件极性方向的识别方法，使用了卷积神经网络，可以自动精准地识别电子元件的极性方向，而且并不针对特定的电子元件结构，适用于各种带极性的电子元件，只要训练卷积神经网络用的图像样本集覆盖了多个种 类的电子元件，就可以实现跨类别的电子元件的极性方向识别，适用性较广；无论是离线训练阶段还是在线测试阶段，均尽可能减少人工干预，并且不依赖于特殊硬件机构，成本较低；由于引用外部数据集并采用迁移学习策略，本方明的实施并不需要预先收集海量的电子元件极性方向的图像样本，因此也适用于电子元件图像样本获取较为困难的场景下。
本发明还提供一种电子元件极性方向的标注方法，包括以下步骤：
根据上述电子元件极性方向的识别方法确定的目标电子元件的极性方向，在板式文件中标注目标电子元件的极性方向信息，板式文件用于保存电子元件的各种属性信息。
上述电子元件极性方向的标注方法可以应用于AOI板式制作，自动、精准地对PCB板卡上的元件进行极性方向标注，从而提升板式制作的自动化水平，改善板式制作的效率与准确性。
根据上述电子元件极性方向的识别方法，本发明还提供一种电子元件极性方向的识别系统，以下就本发明的电子元件极性方向的识别系统的实施例进行详细说明。
参见图3所示，为本发明的电子元件极性方向的识别系统的实施例。该实施例中的电子元件极性方向的识别系统，包括图3中的获取单元200，计算单元210，选取单元220；
获取单元200，用于获取包含目标电子元件的图像；
计算单元210，用于使用训练后的卷积神经网络对图像作前向计算，得到目标电子元件属于各类电子元件的各种极性方向类别的概率分布；
选取单元220，用于选取概率最大的极性方向类别作为目标电子元件的极性方向类别。
在其中一个实施例中，如图4所示，电子元件极性方向的识别系统还包括预处理单元230；
预处理单元230用于在获取单元获取包含目标电子元件的图像后，获取目标电子元件的模板图，以目标电子元件的模板图对图像中的目标电子元件进行匹配，获得图像中的目标电子元件的精确位置，根据精确位置对图像中的目标电子元件进行调整，使图像中的目标电子元件位于图像的中心，调整后的图像供训练后的卷积神经网络作前向计算。
在其中一个实施例中，如图5所示，电子元件极性方向的识别系统还包括建立单元240、预训练单元250以及调整单元260；
建立单元240用于建立各类电子元件的各种极性方向的图像样本集；
预训练单元250用于通过预设的外部数据集预训练卷积神经网络，其中外部数据集包含多个类别的多张已预先标注的自然图像；
调整单元260用于根据图像样本集，对预训练后的卷积神经网络进行调整并作进一步训练，得到训练后的卷积神经网络。
在其中一个实施例中，如图6所示，建立单元240包括配准单元241、截取处理单元242；
配准单元241用于获得PCB板卡图像和PCB模板图，并以PCB模板图为参考，对PCB板卡图像进行位置配准；
截取处理单元242用于截取位置配准后PCB板卡图像上的电子元件图像，以PCB模板图中的各类电子元件对各类电子元件图像中相应的电子元件进行匹配，获得各类电子元件图像中相应电子元件的精确位置，根据相应电子元件的精确位置对相应电子元件进行调整，使相应电子元件位于各类电子元件图像的中心，获得各类电子元件的各种极性方向的图像样本集。
在其中一个实施例中，预训练单元250用于用外部数据集对卷积神经网络进行预训练，使卷积神经网络学习各层次的通用图像特征，获得预训练后的卷积神经网络的初始参数值。
在其中一个实施例中，调整单元260用于根据图像样本集，将初始参数值中的预训练后的卷积神经网络的最后一层节点数调整为各种电子元件的各种极性方向的类别数，并对调整后的卷积神经网络作进一步训练。
在其中一个实施例中，如图7所示，电子元件极性方向的识别系统还包括 报警单元270，用于若选取的概率最大的极性方向类别与预设的电子元件极性方向类别不同，则给出错误报警信息。
本发明的电子元件极性方向的识别系统与本发明的电子元件极性方向的识别方法一一对应，在上述电子元件极性方向的识别方法的实施例阐述的技术特征及其有益效果均适用于电子元件极性方向的识别系统的实施例中。
本发明还提供一种电子元件极性方向的标注系统，如图8所示，包括标注单元280以及上述的电子元件极性方向的识别系统，标注单元280用于根据上述电子元件极性方向的识别方法确定的目标电子元件的极性方向，在板式文件中标注电子元件的极性方向信息，板式文件用于保存电子元件的各种属性信息。
上述电子元件极性方向的标注系统可以应用于AOI板式制作，自动、精准地对PCB板卡上的元件进行极性方向标注，从而提升板式制作的自动化水平，改善板式制作的效率与准确性。
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. 一种电子元件极性方向的识别方法，其特征在于，包括以下步骤：
    获取包含目标电子元件的图像；
    使用训练后的卷积神经网络对所述图像作前向计算，得到所述目标电子元件属于各类电子元件的各种极性方向类别的概率分布；
    选取概率最大的极性方向类别作为所述目标电子元件的极性方向类别。
  2. 根据权利要求1所述的电子元件极性方向的识别方法，其特征在于，在所述获取包含目标电子元件的图像的步骤之后，包括以下步骤：
    获取所述目标电子元件的模板图，以所述目标电子元件的模板图对所述图像中的目标电子元件进行匹配，获得所述图像中的目标电子元件的精确位置，根据所述精确位置对所述图像中的目标电子元件进行调整，使所述图像中的目标电子元件位于所述图像的中心，调整后的图像供所述训练后的卷积神经网络作前向计算。
  3. 根据权利要求1所述的电子元件极性方向的识别方法，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：
    建立各类电子元件的各种极性方向的图像样本集；
    通过预设的外部数据集预训练卷积神经网络，其中所述外部数据集包含多个类别的多张已预先标注的自然图像；
    根据所述图像样本集，对预训练后的卷积神经网络进行调整并作进一步训练，得到所述训练后的卷积神经网络。
  4. 根据权利要求3所述的电子元件极性方向的识别方法，其特征在于，建立各类电子元件的各种极性方向的样本集的步骤包括以下步骤：
    获取PCB板卡图像和PCB模板图，并以所述PCB模板图为参考，对所述PCB板卡图像进行位置配准；
    截取位置配准后PCB板卡图像上的各类电子元件图像，以所述PCB模板图中的各类电子元件对所述各类电子元件图像中相应的电子元件进行匹配，获得所述各类电子元件图像中相应电子元件的精确位置，根据所述相应电子元件的精确位置对所述相应电子元件进行调整，使所述相应电子元件位于各类电子 元件图像的中心，获得所述各类电子元件的各种极性方向的图像样本集。
  5. 根据权利要求3所述的电子元件极性方向的识别方法，其特征在于，所述通过预设的外部数据集预训练卷积神经网络的步骤包括以下步骤：
    用所述外部数据集对卷积神经网络进行预训练，使卷积神经网络学习各层次的通用图像特征，获得预训练后的卷积神经网络的初始参数值。
  6. 根据权利要求5中所述的电子元件极性方向的识别方法，其特征在于，所述根据所述图像样本集，对预训练后的卷积神经网络进行调整并作进一步训练的步骤包括以下步骤：
    根据所述图像样本集，将所述初始参数值中的预训练后的卷积神经网络的最后一层节点数调整为各种电子元件的各种极性方向的类别数，并对调整后的卷积神经网络作进一步训练。
  7. 一种电子元件极性方向的标注方法，其特征在于，包括以下步骤：
    根据权利要求1至6中任意一项所述的电子元件极性方向的识别方法确定的目标电子元件的极性方向，在板式文件中标注所述目标电子元件的极性方向信息，所述板式文件用于保存电子元件的各种属性信息。
  8. 一种电子元件极性方向的识别系统，其特征在于，包括：
    获取单元，用于获取包含目标电子元件的图像；
    计算单元，用于使用训练后的卷积神经网络对所述图像作前向计算，得到所述目标电子元件属于各类电子元件的各种极性方向类别的概率分布；
    选取单元，用于选取概率最大的极性方向类别作为所述目标电子元件的极性方向类别。
  9. 根据权利要求8所述的电子元件极性方向的识别系统，其特征在于，还包括建立单元、预训练单元以及调整单元；
    所述建立单元用于建立各类电子元件的各种极性方向的图像样本集；
    所述预训练单元用于通过预设的外部数据集预训练卷积神经网络，其中所述外部数据集包含多个类别的多张已预先标注的自然图像；
    所述调整单元用于根据所述图像样本集，对预训练后的卷积神经网络进行调整并作进一步训练，得到所述训练后的卷积神经网络。
  10. 一种电子元件极性方向的标注系统，其特征在于，包括标注单元和如权利要求8或9所述的电子元件极性方向的识别系统，其中，所述标注单元用于在板式文件中标注所述目标电子元件的极性方向信息，所述板式文件用于保存电子元件的各种属性信息。
PCT/CN2016/098227 2015-11-23 2016-09-06 电子元件极性的识别方法和系统、标注方法和系统 Ceased WO2017088552A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201510822818.6A CN105513046B (zh) 2015-11-23 2015-11-23 电子元件极性的识别方法和系统、标注方法和系统
CN201510822818.6 2015-11-23

Publications (1)

Publication Number Publication Date
WO2017088552A1 true WO2017088552A1 (zh) 2017-06-01

Family

ID=55721003

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2016/098227 Ceased WO2017088552A1 (zh) 2015-11-23 2016-09-06 电子元件极性的识别方法和系统、标注方法和系统

Country Status (2)

Country Link
CN (1) CN105513046B (zh)
WO (1) WO2017088552A1 (zh)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111640088A (zh) * 2020-04-22 2020-09-08 深圳拓邦股份有限公司 基于深度学习的电子元件极性检测方法、系统及电子装置
CN113468833A (zh) * 2021-06-11 2021-10-01 山东英信计算机技术有限公司 一种原理图中元器件属性标记方法、装置、设备及介质
CN113804704A (zh) * 2020-06-11 2021-12-17 广东美的白色家电技术创新中心有限公司 电路板检测方法、视觉检测设备及具有存储功能的装置
EP4358020A1 (en) 2022-10-17 2024-04-24 Fitech sp. z o.o. Method of detecting errors in the placement of elements in the pcb

Families Citing this family (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105513046B (zh) * 2015-11-23 2019-03-01 广州视源电子科技股份有限公司 电子元件极性的识别方法和系统、标注方法和系统
CN105426917A (zh) * 2015-11-23 2016-03-23 广州视源电子科技股份有限公司 一种元件分类方法及装置
CN107563123A (zh) * 2017-09-27 2018-01-09 百度在线网络技术（北京）有限公司 用于标注医学图像的方法和装置
CN107886131A (zh) * 2017-11-24 2018-04-06 佛山科学技术学院 一种基于卷积神经网络检测电路板元器件极性方法和装置
CN108984992B (zh) * 2018-09-25 2022-03-04 郑州云海信息技术有限公司 一种电路板设计方法和装置
CN109859164B (zh) * 2018-12-21 2023-08-01 苏州绿控传动科技股份有限公司 一种通过快速型卷积神经网络对pcba外观检验的方法
CN115047308B (zh) * 2021-02-25 2025-07-04 北京小米移动软件有限公司 一种电路板导电性能的检测方法及装置
CN114399502A (zh) * 2022-03-24 2022-04-26 视睿（杭州）信息科技有限公司 一种适用于led芯片的外观缺陷检测方法、系统及存储介质
CN115221832B (zh) * 2022-08-03 2025-07-11 上海望友信息科技有限公司 一种封装极性标识设计方法、系统、电子设备和存储介质
CN117315029B (zh) * 2023-10-18 2025-11-04 北京泛融科技有限公司 一种电极方向识别方法、系统、设备及介质

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20060029265A1 (en) * 2004-08-04 2006-02-09 Samsung Electronics Co., Ltd. Face detection method based on skin color and pattern match
CN103544705A (zh) * 2013-10-25 2014-01-29 华南理工大学 一种基于深度卷积神经网络的图像质量测试方法
CN103927534A (zh) * 2014-04-26 2014-07-16 无锡信捷电气股份有限公司 一种基于卷积神经网络的喷码字符在线视觉检测方法
CN104459421A (zh) * 2014-12-30 2015-03-25 广州视源电子科技股份有限公司 一种二极管极性检测方法与系统
CN104537650A (zh) * 2014-12-16 2015-04-22 广州视源电子科技股份有限公司 一种铝电解电容器的极性方向判定方法及装置
CN105513046A (zh) * 2015-11-23 2016-04-20 广州视源电子科技股份有限公司 电子元件极性的识别方法和系统、标注方法和系统

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20060029265A1 (en) * 2004-08-04 2006-02-09 Samsung Electronics Co., Ltd. Face detection method based on skin color and pattern match
CN103544705A (zh) * 2013-10-25 2014-01-29 华南理工大学 一种基于深度卷积神经网络的图像质量测试方法
CN103927534A (zh) * 2014-04-26 2014-07-16 无锡信捷电气股份有限公司 一种基于卷积神经网络的喷码字符在线视觉检测方法
CN104537650A (zh) * 2014-12-16 2015-04-22 广州视源电子科技股份有限公司 一种铝电解电容器的极性方向判定方法及装置
CN104459421A (zh) * 2014-12-30 2015-03-25 广州视源电子科技股份有限公司 一种二极管极性检测方法与系统
CN105513046A (zh) * 2015-11-23 2016-04-20 广州视源电子科技股份有限公司 电子元件极性的识别方法和系统、标注方法和系统

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111640088A (zh) * 2020-04-22 2020-09-08 深圳拓邦股份有限公司 基于深度学习的电子元件极性检测方法、系统及电子装置
CN111640088B (zh) * 2020-04-22 2023-12-01 深圳拓邦股份有限公司 基于深度学习的电子元件极性检测方法、系统及电子装置
CN113804704A (zh) * 2020-06-11 2021-12-17 广东美的白色家电技术创新中心有限公司 电路板检测方法、视觉检测设备及具有存储功能的装置
CN113468833A (zh) * 2021-06-11 2021-10-01 山东英信计算机技术有限公司 一种原理图中元器件属性标记方法、装置、设备及介质
CN113468833B (zh) * 2021-06-11 2024-02-09 山东英信计算机技术有限公司 一种原理图中元器件属性标记方法、装置、设备及介质
EP4358020A1 (en) 2022-10-17 2024-04-24 Fitech sp. z o.o. Method of detecting errors in the placement of elements in the pcb

Also Published As

Publication number Publication date
CN105513046A (zh) 2016-04-20
CN105513046B (zh) 2019-03-01

Similar Documents

Publication Publication Date Title
WO2017088552A1 (zh) 电子元件极性的识别方法和系统、标注方法和系统
WO2017088553A1 (zh) 电子元件极性方向的快速识别、标注的方法和系统
US12283039B2 (en) Method and system for defect inspection based on deep learning
CN108985214B (zh) 图像数据的标注方法和装置
CN114627089B (zh) 缺陷识别方法、装置、计算机设备及计算机可读存储介质
WO2017215241A1 (zh) 电路板元件漏件检测方法和系统
WO2020244183A1 (zh) 数据标注
CN108961236A (zh) 电路板缺陷检测模型的训练方法及装置、检测方法及装置
WO2017088537A1 (zh) 一种元件分类方法及装置
CN110335274B (zh) 一种三维模具缺陷检测方法及装置
CN119048497B (zh) Pcb缺陷检测方法、系统、设备和存储介质
CN110992313A (zh) 一种基于机器视觉的摄像头图像中心检测及中心调整方法
CN119379600A (zh) 基于2d和3d图像视觉结合的pcb板缺陷检测方法和设备
CN110031471B (zh) 大口径光学元件表面缺陷增长分析方法、系统、装置
CN113920450B (zh) 基于本征图像分解的绝缘子rtv涂层识别方法及装置
CN106447045A (zh) 基于机器学习的adas系统的测评方法
CN109829887A (zh) 一种基于深度神经网络的图像质量评估方法
CN114841980A (zh) 基于巡线航拍图像的绝缘子缺陷检测方法及系统
CN118172338B (zh) 应用前馈神经网络的智能分析系统
CN118631973B (zh) 一种基于多目标定位分析的监控画面实时优化方法
CN116467479B (zh) 一种基于人工智能的电力场景图像数据筛选方法
CN110784330B (zh) 一种应用识别模型的生成方法及装置
CN115223153B (zh) 一种燃烧类烟雾分级识别方法、系统、电子设备及介质
CN114494038A (zh) 基于改进yolox-s的靶面透视失真矫正方法
CN106373161A (zh) 一种基于sift特征点的定位方法

Legal Events

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

Ref document number: 16867779

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 16867779

Country of ref document: EP

Kind code of ref document: A1