WO2021232613A1 - 酒瓶表面缺陷检测方法、电子装置及存储介质 - Google Patents

酒瓶表面缺陷检测方法、电子装置及存储介质 Download PDF

Info

Publication number
WO2021232613A1
WO2021232613A1 PCT/CN2020/112555 CN2020112555W WO2021232613A1 WO 2021232613 A1 WO2021232613 A1 WO 2021232613A1 CN 2020112555 W CN2020112555 W CN 2020112555W WO 2021232613 A1 WO2021232613 A1 WO 2021232613A1
Authority
WO
WIPO (PCT)
Prior art keywords
neural network
wine bottle
deep learning
learning neural
image
Prior art date
Application number
PCT/CN2020/112555
Other languages
English (en)
French (fr)
Inventor
邓辅秦
黄永深
黄杰文
冯华
李伟科
Original Assignee
五邑大学
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 五邑大学 filed Critical 五邑大学
Publication of WO2021232613A1 publication Critical patent/WO2021232613A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/84—Systems specially adapted for particular applications
    • G01N21/88—Investigating the presence of flaws or contamination
    • G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/84—Systems specially adapted for particular applications
    • G01N21/88—Investigating the presence of flaws or contamination
    • G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
    • G01N2021/8887—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges based on image processing techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10024—Color image
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20092—Interactive image processing based on input by user
    • G06T2207/20104—Interactive definition of region of interest [ROI]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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 invention relates to the technical field of wine bottle surface defect detection, in particular to a wine bottle surface defect detection method, electronic device and storage medium.
  • Detection methods based on traditional machine vision mainly rely on template matching. At the same time, it is difficult to extract appropriate feature vectors for surface defect images of workpieces with complex structures, many types of defects, and large feature differences, resulting in poor final detection results.
  • the present invention aims to solve at least one of the technical problems existing in the prior art. For this reason, the present invention proposes a method for detecting surface defects of wine bottles, which can perform quality inspection on surface defects of bottled wine bottles, thereby greatly improving the production efficiency of enterprises, reducing production costs, and having the ability to learn and image for a large amount of inspection data. Feature extraction capabilities, without the need to manually design complex feature extraction algorithms, thereby reducing the difficulty of intelligent detection.
  • the present invention also provides an electronic device with the above-mentioned method for detecting surface defects of wine bottles.
  • the present invention also provides a computer-readable storage medium having the above-mentioned method for detecting surface defects of wine bottles.
  • the surface image of the wine bottle to be detected is input to the deep learning neural network to obtain surface defect information corresponding to the surface image of the wine bottle to be detected.
  • the method for detecting defects on the surface of a wine bottle has at least the following beneficial effects: input the image of the surface of the wine bottle in the training data set into the paired convolutional neural network, thereby training it, and obtaining a deep learning neural network with good detection performance Network, and then input the image of the surface of the wine bottle to be detected into the deep learning neural network to obtain surface defect information.
  • the surface defect information can represent the surface defect in the current image of the surface of the wine bottle, so the defects on the surface of the wine bottle can be detected quickly and accurately , Greatly improve the production efficiency of enterprises, reduce production costs, and convolutional neural networks can reduce the difficulty of intelligent detection.
  • the method further includes the following steps:
  • the contrast of the surface image of the wine bottle is enhanced.
  • the constructing a convolutional neural network and training the convolutional neural network based on the training data set to obtain a deep learning neural network includes the following steps:
  • the inputting the surface image of the wine bottle to be detected into the deep learning neural network to obtain the surface defect information corresponding to the surface image of the wine bottle to be detected includes the following steps:
  • the IoU value Calculate the IoU value according to the detection rectangular frame and the preset labeled rectangular frame. If the IoU value is greater than the preset IoU threshold, it is determined that the deep learning neural network has successfully detected; otherwise, the deep learning neural network is determined to be the deep learning neural network. The test failed.
  • the image of the surface of the wine bottle includes at least one piece of surface defect information.
  • the deep learning neural network includes a convolutional layer, an RPN layer, a first pooling layer, a second pooling layer, a third pooling layer, a first fully connected layer, and a second fully connected layer.
  • Layer, the third fully connected layer, the output information of the first fully connected layer is the input information of the second pooling layer, and the output information of the second fully connected layer is the input of the third pooling layer information.
  • the preset IoU thresholds of the first pooling layer, the second pooling layer, and the third pooling layer are different from each other, and the first fully connected layer, Both the second fully connected layer and the third fully connected layer perform the following steps:
  • the IoU value is calculated according to the detection rectangular frame and the preset labeled rectangular frame, and if the IoU value is greater than the preset IoU threshold, the detection rectangular frame is output.
  • the surface defect information includes any one of the following:
  • the category of the surface defect The category name of the surface defect; the detection rectangle corresponding to the position of the surface defect; the confidence ranking and value of the deep learning neural network; the judgment result.
  • An electronic device includes: a memory, a processor, and a computer program stored on the memory and capable of being run on the processor, and the processor executes the program as the present invention
  • an electronic device executes the method for detecting defects on the surface of a wine bottle as described in any one of the first aspect of the present invention, it has all the beneficial effects of the first aspect of the present invention.
  • a computer-readable storage medium stores computer-executable instructions for executing the wine bottle surface defect detection according to any one of the first aspect of the present invention method.
  • the computer-readable storage medium of the embodiment of the present invention stores computer executable instructions for executing the method for detecting defects on the surface of a wine bottle according to any one of the first aspect of the present invention, it has all of the first aspect of the present invention. Beneficial effect.
  • FIG. 1 is a flowchart of a method for detecting defects on the surface of a wine bottle according to an embodiment of the present invention
  • FIG. 2 is a flowchart after the step of obtaining a training data set based on the obtained wine bottle surface image of a method for detecting defects on the surface of a wine bottle provided by an embodiment of the present invention
  • FIG. 3 is a flowchart of the steps of constructing a convolutional neural network and training the convolutional neural network based on the training data set to obtain a deep learning neural network for a method for detecting defects on the surface of a wine bottle provided by an embodiment of the present invention
  • Fig. 4 is a step of inputting a picture of the surface of a wine bottle to be inspected into a deep learning neural network in a method for detecting surface defects of a wine bottle provided by an embodiment of the present invention to obtain surface defect information corresponding to the picture of the surface of the wine bottle to be inspected Flow chart
  • FIG. 5 is a diagram of the overall network structure of a deep learning neural network of a method for detecting defects on the surface of a wine bottle provided by an embodiment of the present invention
  • FIG. 6 is a schematic structural diagram of an electronic device provided by an embodiment of the present invention.
  • the electronic device 100 the processor 101, and the memory 102.
  • multiple means two or more.
  • the present invention provides a method, electronic device and storage medium for detecting surface defects of wine bottles.
  • Faster R-CNN and Cascade R-CNN networks are used as the basic convolutional neural network structure for surface defect detection of bottled wine bottles, combined with ResNeXt.
  • Residual network structure to increase the depth of the convolutional neural network, improve the ability of the convolutional neural network to extract deep-level features, and then use the FPN network structure to extract better feature information, thereby improving the final detection effect of the convolutional neural network;
  • the training data set of surface defects of wine bottles is enhanced to increase the number of training data.
  • the training data set is used to train the convolutional neural network to obtain a deep learning neural network capable of detecting surface defects of wine bottles.
  • the surface image of the wine bottle is input to the deep learning neural network, and the surface defects in the surface image of the wine bottle are artificially framed by the labeled rectangle to indicate the actual position of the defect.
  • the deep learning neural network finds the surface defects in the surface image of the wine bottle and uses it to detect Rectangular box to frame it, and then calculate the IoU value of the detection rectangular box and the labeled rectangular box. If the IoU value is greater than the preset IoU threshold, the detection is considered successful and judged as a positive example, otherwise the detection fails and judged as a negative example, and finally The deep learning neural network outputs surface defect information.
  • an electronic device 100 provided by an embodiment of the first aspect of the present invention includes a memory 102 and a processor 101.
  • a processor 101 and a memory 102 are taken as an example.
  • the processor and the memory may be connected through a bus or in other ways.
  • the connection through a bus is taken as an example.
  • the memory 102 can be used to store non-transitory software programs and non-transitory computer-executable programs.
  • the memory 102 may include a high-speed random access memory, and may also include a non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state storage devices.
  • the memory 102 may optionally include a memory 102 remotely provided with respect to the processor, and these remote memories may be connected to the electronic device 100 via a network. Examples of the aforementioned networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.
  • the device structure shown in FIG. 6 does not constitute a limitation on the electronic device 100, and may include more or fewer components than shown, or a combination of certain components, or a different component arrangement.
  • the processor 101 in the electronic device 100 may be used to call a method for detecting defects on the surface of a wine bottle stored in the memory 102, and execute the following steps:
  • FIG. 1 it is a method for detecting surface defects of wine bottles according to an embodiment of the second aspect of the present invention, including:
  • the convolutional neural network Input the image of the wine bottle surface in the detection data set into the convolutional neural network to train the convolutional neural network.
  • Each training generates a detection model, and then the convolutional neural network calculates the detection model with the smallest loss function, and the smallest loss function
  • the detection model is the deep learning neural network required in this embodiment, so as to obtain a deep learning neural network with good detection ability for the surface defects of the wine bottle, and then input the image of the wine bottle surface to be detected into the trained deep learning neural network
  • the deep learning neural network recognizes and extracts the surface defect features in the surface image of the wine bottle, that is, the surface defect information can be obtained.
  • the method for detecting the surface defect of the wine bottle according to the embodiment of the present invention can save the work of detecting the surface defect of the bottled wine bottle.
  • the convolutional neural network in the embodiment of the present invention has the learning ability for a large amount of detection data.
  • the convolutional neural network can continuously improve the convolutional neural model during the training process, thereby eliminating the need for artificially designing complex feature extraction
  • the algorithm can have powerful image feature extraction capabilities, greatly reducing the difficulty of intelligent detection.
  • step S110 obtains a training data set based on the obtained image of the surface of the wine bottle
  • the method further includes the following steps:
  • Data enhancement of the images in the training data set can expand the original training data set. Under the premise of ensuring the accuracy and reliability of the data, new valid data can be obtained. If the bottled wine bottle data in the training data set has many types of defects, and the amount of data is small In the training process, it is easy to cause the trained model to have a tendency towards certain defects, resulting in partial defects with small weights and fail to be fully identified. Data enhancement methods on the training data set can avoid this problem.
  • the Y-axis mirroring is to change the position of the defects on the surface of the wine bottle, which conforms to the characteristics of random occurrence of defects, and at the same time does not coincide with the original position.
  • the color enhancement and contrast enhancement are to adapt to the light problem in the process of obtaining the image of the wine bottle. Different lighting conditions also highlight the contour of the defect and the difference from the non-defective area.
  • step S120 constructs a convolutional neural network and trains the convolutional neural network based on the training data set, and obtaining a deep learning neural network includes the following steps:
  • the Faster R-CNN network and the Cascade R-CNN network are both commonly used R-CNN neural network structures.
  • the Faster R-CNN network uses the RPN layer to replace the traditional selective search algorithm, which greatly increases the final detection rate, but there is How to choose the IoU threshold;
  • Cascade R-CNN network by connecting multiple detection networks to improve the prediction results after model training, in which different IoU thresholds are set to determine the number of positive and negative samples to train the detection network Therefore, this embodiment combines the Faster R-CNN network and the Cascade R-CNN network to construct a convolutional neural network;
  • top-level feature map when predicting.
  • the top-level feature has rich semantic information, it also loses part of the precise location information of the target to be detected, while the low-level features have a lot of precise location information.
  • Reasonable use of low-level feature information is conducive to improving the accuracy of small object detection.
  • Adding the ResNeXt residual network structure to the convolutional neural network uses parallel stacked residual structure blocks to improve the accuracy of the detection effect without much change in the number of parameters, and then add the input and output of the network to obtain the features
  • the graph can achieve the purpose of improving the accuracy of the network model without significantly increasing the number of parameters.
  • the residual block structure of the ResNeXt network is the same, the corresponding hyperparameters of the training network will also be reduced.
  • the FPN network structure is combined with the convolutional neural network to extract better feature information, and the ResNeXt residual network structure is combined to increase the depth of the overall network, thereby improving the model's ability to extract deep features, thereby improving the final The detection effect.
  • step S130 inputting the surface image of the wine bottle to be inspected into the deep learning neural network to obtain the surface defect information corresponding to the image of the surface of the wine bottle to be inspected includes the following steps:
  • S133 If the IoU value is greater than the preset IoU threshold, it is determined that S134, the deep learning neural network detection is successful, otherwise it is determined that S135, the deep learning neural network detection failed.
  • the deep learning neural network uses a detection rectangle to frame the surface defects in the picture to be detected, and artificially uses the label rectangle to frame the surface defects in the picture to be detected in advance, the label rectangle represents the actual location of the surface defect, and then calculates the detection rectangle. Compare the IoU value with the preset IoU threshold with the IoU value of the labeled rectangular box. If the IoU value is greater than the IoU threshold, it means that the detection rectangular box and the labeled rectangular box are consistent, and the deep learning neural network has successfully found the image to be detected Otherwise, it means that the detection rectangle does not match the label rectangle, and the deep learning neural network detection fails.
  • the surface image of the wine bottle contains at least one surface defect information. If there are a large number of normal pictures in the training data set, there is no information on the surface defects of the bottled wine bottles that need to be detected in the normal pictures, which will not help the training of the model. At the same time, the defective workpieces also contain the features of the normal pictures. In the process of training the model, training a large number of normal images will cause a decrease in the ability of the model to detect surface defects of the workpiece. Therefore, in order to ensure the detection ability of the convolutional neural network, the images in the training data set contain at least one surface defect feature.
  • the deep learning neural network includes a convolutional layer, an RPN layer, a first pooling layer, a second pooling layer, a third pooling layer, a first fully connected layer, and a second fully connected layer.
  • Layer, the third fully connected layer, the output information of the first fully connected layer is the input information of the second pooling layer, and the output information of the second fully connected layer is the input information of the third pooling layer.
  • the recommended area represents the initial prediction position of the surface defect in the image to be detected, and the final feature map and the recommended area are together
  • Input the first pooling layer for processing, and the output of the first fully connected layer, the second fully connected layer, and the third fully connected layer are the types of surface defects and the detection rectangular frame indicating the location of the surface defects.
  • the preset IoU thresholds of the first pooling layer, the second pooling layer, and the third pooling layer are different.
  • the first fully connected layer, the second fully connected layer, and the third The fully connected layer performs the following steps:
  • the IoU thresholds of the information processed by the first pooling layer, the second pooling layer, and the third pooling layer are set to different values, from low to high.
  • the IoU threshold represents the accuracy requirements for the detection of the rectangular frame, the higher the value is , The higher the accuracy requirements for the detection of the rectangular frame, because the preset IoU threshold of the second pooling layer is higher than the preset IoU threshold of the first pooling layer, the first pooling layer and the first fully connected layer are adjusted Detect the position of the rectangular frame, and output a detection rectangular frame suitable for higher IoU threshold requirements to the second pooling layer, so that the detection rectangular frame output by the second fully connected layer is compared with the detection rectangular frame output by the first fully connected layer.
  • the working principle of the second fully connected layer and the third pooling layer are the same as above.
  • the cascaded pooling layer and fully connected layer can detect the rectangular frame every time The accuracy of both has been improved, getting closer and closer to the actual position of the surface defect in the picture, thereby improving the detection accuracy of the deep learning neural network.
  • the surface defect information includes any one of the following:
  • the category of the surface defect The category of the surface defect; the name of the category of the surface defect; the detection rectangle corresponding to the position of the surface defect; the confidence ranking and value of the deep learning neural network; the judgment result.
  • the types of bottled wine bottle surface defects involved in this embodiment include broken caps, deformed caps, broken sides of caps, cap twisting, cap breaks, skewed labels, wrinkled labels, bubbles on labels, and coding.
  • the computer-readable storage medium of the embodiment of the third aspect of the present invention stores computer-executable instructions, and the computer-executable instructions are used to execute the method for detecting defects on the surface of a wine bottle as described in the embodiment of the second aspect.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Artificial Intelligence (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Computational Linguistics (AREA)
  • Software Systems (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Evolutionary Biology (AREA)
  • Multimedia (AREA)
  • Signal Processing (AREA)
  • Chemical & Material Sciences (AREA)
  • Analytical Chemistry (AREA)
  • Biochemistry (AREA)
  • Immunology (AREA)
  • Pathology (AREA)
  • Quality & Reliability (AREA)
  • Image Analysis (AREA)
  • Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)

Abstract

一种酒瓶表面缺陷检测方法，并公开了具有酒瓶表面缺陷检测方法的电子装置和计算机可读存储介质，酒瓶表面缺陷检测方法包括：基于所获取到的酒瓶表面图片得到训练数据集（S110）；构建卷积神经网络并基于训练数据集训练卷积神经网络，得到深度学习神经网络（S120）；将待检测的酒瓶表面图片输入到深度学习神经网络，以获得待检测的酒瓶表面图片对应的表面缺陷信息（S130）；所述方法能够对瓶装酒瓶进行表面缺陷的质检，从而大大提高企业的生产效率，减少生产成本，并且具有针对大量检测数据的学习能力和图像特征提取能力，无需人为设计复杂的特征提取算法，从而降低智能检测难度。

Description

酒瓶表面缺陷检测方法、电子装置及存储介质 技术领域
本发明涉及酒瓶表面缺陷检测技术领域，特别涉及一种酒瓶表面缺陷检测方法、电子装置及存储介质。
背景技术
在瓶装酒瓶的工业生产过程中会受到原材料质量、酒瓶图纸设计方案、加工工艺(灌装)以及机床设备质量、生产环境等因素的影响，最终形成的瓶装酒中可能存在各类表面缺陷而影响到整体的产品质量，当今消费者对工业产品的要求在不断提高，消费者的消费欲望并不再局限于产品的质量好坏，还对产品的外观、视觉效果也有着额外的需求，故针对酿酒行业而言，瓶装酒瓶表面缺陷的质检工作将显得尤为重要，提高瓶装酒瓶表面缺陷的质检能力能在一定程度上影响瓶装酒的销售前景。
传统的人工目检工件表面缺陷或者人工抽样检测工件表面缺陷所造成的产品漏检率和错判率可能极高，针对产品的检测质量效果也因人而异并且缺乏效率，导致工件产品无法大批量生产，不但降低企业的生产效率，还增加企业的生产成本。
基于传统机器视觉的检测方法主要依赖于模板匹配，同时对结构复杂、缺陷种类多，特征差别大的工件表面缺陷图像难以提取合适的特征向量，导致最终的检测效果不佳。
发明内容
本发明旨在至少解决现有技术中存在的技术问题之一。为此，本发明提出一种酒瓶表面缺陷检测方法，能够对瓶装酒瓶进行表面缺陷的质检，从而大大提高企业的生产效率，减少生产成本，并且具有针对大量检测数据的学习能力和图像特征提取能力，无需人为设计复杂的特征提取算法，从而降低智能检测难度。
本发明还提出一种具有上述酒瓶表面缺陷检测方法的电子装置。
本发明还提出一种具有上述酒瓶表面缺陷检测方法的计算机可读存储介质。
根据本发明的第一方面实施例的酒瓶表面缺陷检测方法，包括：
基于所获取到的酒瓶表面图片得到训练数据集；
构建卷积神经网络并基于所述训练数据集训练所述卷积神经网络，得到深度学习神经网络；
将待检测的酒瓶表面图片输入到所述深度学习神经网络，以获得待检测的所述酒瓶表面图片对应的表面缺陷信息。
根据本发明实施例的酒瓶表面缺陷检测方法，至少具有如下有益效果：将训练数据集中的酒瓶表面图片输入对卷积神经网络，从而对其进行训练，得到具有良好检测性能的深度学习神经网络，然后将待检测的酒瓶表面图片输入深度学习神经网络，得到表面缺陷信息，表面缺陷信息能够表示当前酒瓶表面图片中的表面缺陷，因此能够快速、准确地检测到酒瓶表面的缺陷，大大提高企业的生产效率，减少生产成本，并且卷积神经网络能够降低智能检测难度。
根据本发明的一些实施例，所述基于所获取到的酒瓶表面图片得到训练数据集之后，还包括以下步骤：
对所述酒瓶表面图片进行Y轴镜像翻转；
对所述酒瓶表面图片进行增强颜色；
对所述酒瓶表面图片进行增强对比度。
根据本发明的一些实施例，所述构建卷积神经网络并基于所述训练数据集训练所述卷积神经网络，得到深度学习神经网络，包括以下步骤：
基于Faster R-CNN网络和Cascade R-CNN网络构建卷积神经网络；
基于ResNeXt残差网络结构和FPN网络结构优化所述卷积神经网络，得到第一卷积神经网络；
基于所述训练数据集训练所述第一卷积神经网络，得到所述深度学习神经网络。
根据本发明的一些实施例，所述将待检测的酒瓶表面图片输入到所述深度学习神经网络，以获得待检测的所述酒瓶表面图片对应的表面缺陷信息包括以下步骤：
基于所述深度学习神经网络中的检测矩形框将所述酒瓶表面图片中的表面缺陷框出；
根据所述检测矩形框与预设的标注矩形框计算IoU值，若所述IoU值大于预设的IoU阈值，则判断为所述深度学习神经网络检测成功，否则判断为所述深度学习神经网络检测失败。
根据本发明的一些实施例，所述酒瓶表面图片包含至少一个所述表面缺陷信息。
根据本发明的一些实施例，所述深度学习神经网络包括卷积层、RPN层、第一池化层、第二池化层、第三池化层、第一全连接层、第二全连接层、第三全连接层，所述第一全连接层的输出信息为所述第二池化层的输入信息，所述第二全连接层的输出信息为所述第三池化层的输入信息。
根据本发明的一些实施例，所述第一池化层、所述第二池化层、所述第三池化层预设的所述IoU阈值各不相同，所述第一全连接层、所述第二全连接层、所述第三全连接层均执行以下步骤：
基于所述深度学习神经网络中的所述检测矩形框将所述酒瓶表面图片中的表面缺陷框出；
根据所述检测矩形框与预设的所述标注矩形框计算IoU值，若所述IoU值大于预设的IoU阈值，则输出所述检测矩形框。
根据本发明的一些实施例，所述表面缺陷信息包括以下任意一种：
所述表面缺陷的类别；所述表面缺陷的类别名；表示所述表面缺陷的位置对应的检测矩形框；所述深度学习神经网络的置信度排名及数值；判定结果。
根据本发明的第二方面实施例的电子装置，包括：存储器、处理器及存储在所述存储器上并可在处理器上运行的计算机程序，所述处理器执行所述程序时实现如本发明第一方面中任一项所述的酒瓶表面缺陷检测方法。
由于本发明实施例的一种电子装置执行如本发明第一方面中任一项所述的酒瓶表面缺陷检测方法，因此具有本发明第一方面的所有有益效果。
根据本发明的第三方面实施例的计算机可读存储介质，存储有计算机可执行指令，所述计算机可执行指令用于执行如本发明第一方面中任一项所述的酒瓶表面缺陷检测方法。
由于本发明实施例的计算机可读存储介质上存储有用于执行如本发明第一方面中任一项所述的酒瓶表面缺陷检测方法的计算机可执行指令，因此具有本发明第一方面的所有有益效果。
本发明的附加方面和优点将在下面的描述中部分给出，部分将从下面的描述中变得明显，或通过本发明的实践了解到。
附图说明
本发明的上述和/或附加的方面和优点从结合下面附图对实施例的描述中将变得明显和容易理解，其中：
图1为本发明实施例提供的一种酒瓶表面缺陷检测方法的流程图；
图2为本发明实施例提供的一种酒瓶表面缺陷检测方法的基于所获取到的酒瓶表面图片得到训练数据集这一步骤之后的流程图；
图3为本发明实施例提供的一种酒瓶表面缺陷检测方法的构建卷积神经网络并基于所述训练数据集训练卷积神经网络，得到深度学习神经网络这一步骤的流程图；
图4为本发明实施例提供的一种酒瓶表面缺陷检测方法的将待检测的酒瓶表面图片输入到深度学习神经网络，以获得待检测的酒瓶表面图片对应的表面缺陷信息这一步骤的流程图；
图5为本发明实施例提供的一种酒瓶表面缺陷检测方法的深度学习神经网络的整体网络结构图；
图6为本发明实施例提供的一种电子装置的结构示意图。
附图标记：
电子装置100、处理器101、存储器102。
具体实施方式
下面详细描述本发明的实施例，所述实施例的示例在附图中示出，其中自始至终相同或类似的标号表示相同或类似的元件或具有相同或类似功能的元件。下面通过参考附图描述的实施例是示例性的，仅用于解释本实用新型，而不能理解为对本实用新型的限制。
在本发明的描述中，多个的含义是两个以上。
本发明的描述中，除非另有明确的限定，设置、安装、连接等词语应做广义理解，所属技术领域技术人员可以结合技术方案的具体内容合理确定上述词语在本发明中的具体含义。
本发明提供了一种酒瓶表面缺陷检测方法、电子装置及存储介质，首先使用Faster R-CNN和Cascade R-CNN网络作为瓶装酒瓶表面缺陷检测的基础卷积神经网络的结构，同时结合ResNeXt残差网络结构来增加卷积神经网络的深度，提升卷积神经网络提取深层次特征能力，再搭配FPN网络结构来提取更好的特征信息，从而提高卷积神经网络最终的检测效果；然后对酒瓶表面缺陷的训练数 据集进行数据增强以扩增训练数据的数量，使用训练数据集对卷积神经网络进行训练，得到具有酒瓶表面缺陷检测能力的深度学习神经网络，然后，将待检测的酒瓶表面图片输入深度学习神经网络，并且人为使用标注矩形框将酒瓶表面图片中的表面缺陷框出，表示缺陷的实际位置，深度学习神经网络寻找酒瓶表面图片中的表面缺陷并用检测矩形框将其框出，然后计算检测矩形框与标注矩形框的IoU值，若IoU值大于预设的IoU阈值，则认为检测成功，判断为正例，否则检测失败并判断为负例，最后深度学习神经网络输出表面缺陷信息。
参照图6，为本发明第一方面实施例提供的一种电子装置100，包括存储器102、处理器101，图6中以一个处理器101和一个存储器102为例。
处理器和存储器可以通过总线或者其他方式连接，图6中以通过总线连接为例。
存储器102作为一种非暂态计算机可读存储介质，可用于存储非暂态软件程序以及非暂态性计算机可执行程序。此外，存储器102可以包括高速随机存取存储器，还可以包括非暂态存储器，例如至少一个磁盘存储器件、闪存器件、或其他非暂态固态存储器件。在一些实施方式中，存储器102可选包括相对于处理器远程设置的存储器102，这些远程存储器可以通过网络连接至该电子装置100。上述网络的实例包括但不限于互联网、企业内部网、局域网、移动通信网及其组合。
本领域技术人员可以理解，图6中示出的装置结构并不构成对电子装置100的限定，可以包括比图示更多或更少的部件，或者组合某些部件，或者不同的部件布置。
参照图1和图6所示，在本发明第一方面实施例的电子装置中，电子装置100中处理器101可以用于调用存储器102中存储的酒瓶表面缺陷检测方法，并执行以下步骤：
S110、基于所获取到的酒瓶表面图片得到训练数据集；
S120、构建卷积神经网络并基于训练数据集训练卷积神经网络，得到深度学习神经网络；
S130、将待检测的酒瓶表面图片输入到深度学习神经网络，以获得待检测的酒瓶表面图片对应的表面缺陷信息。
基于上述电子装置的硬件结构，提出本发明的一种酒瓶表面缺陷检测方法的各个实施例。
参照图1所示，为本发明第二方面实施例的酒瓶表面缺陷检测方法，包括：
S110、基于所获取到的酒瓶表面图片得到训练数据集；
S120、构建卷积神经网络并基于训练数据集训练卷积神经网络，得到深度学习神经网络；
S130、将待检测的酒瓶表面图片输入到深度学习神经网络，以获得待检测的酒瓶表面图片对应的表面缺陷信息。
将检测数据集中的酒瓶表面图片输入卷积神经网络，从而对卷积神经网络进行训练，每一次训练产生一个检测模型，然后卷积神经网络计算得到损失函数最小的检测模型，损失函数最小的检测模型即为本实施例所需的深度学习神经网络，从而得到针对酒瓶表面缺陷具有良好检测能力的深度学习神经网络，然后，将待检测的酒瓶表面图片输入训练好的深度学习神经网络中，深度学习神经网络识别并提取该酒瓶表面图片中的表面缺陷特征，即能得到表面缺陷信息，因此本发明实施例一种酒瓶表面缺陷检测方法能够节省对瓶装酒瓶表面缺陷检测工作的人力物力，并且，本发明实施例中的卷积神经网络具有针对大量检测数据的学习能力，卷积神经网络在训练过程中，能够不断改善卷积神经模型，从而无需人为设计复杂的特征提取算法就能够拥有强大的图像特征提取能力，大大降低智能检测的难度。
参照图2，在本实施例中，步骤S110基于所获取到的酒瓶表面图片得到训练数据集之后，还包括以下步骤：
S111、对酒瓶表面图片进行Y轴镜像翻转；
S112、对酒瓶表面图片进行增强颜色；
S113、对酒瓶表面图片进行增强对比度。
对训练数据集中的图片进行数据增强能够扩展原始的训练数据集，在保证数据精确可靠的前提下，获取新的有效数据，若训练数据集中的瓶装酒瓶数据的缺陷种类多，数据量较少，在训练的过程中容易导致训练好的模型出现对某种缺陷的倾向性，导致部分缺陷的权重较小，未能全面的识别出来，对训练数据集进行数据增强方法能够避免这一问题，其中Y轴镜像是为了改变酒瓶表面缺陷的位置，符合缺陷随意产生的特点，同时又不与原图位置相重合，颜色增强以及对比 度增强是为了适应酒瓶图像获取过程中的光线问题，适应不同的光照条件，同时也突出缺陷的轮廓以及与非缺陷区域的差异。
参照图3，在本实施例中，步骤S120构建卷积神经网络并基于训练数据集训练卷积神经网络，得到深度学习神经网络包括以下步骤：
S121、基于Faster R-CNN网络和Cascade R-CNN网络构建卷积神经网络；
S122、基于ResNeXt残差网络结构和FPN网络结构优化卷积神经网络，得到第一卷积神经网络；
S123、基于训练数据集训练第一卷积神经网络，得到深度学习神经网络。
Faster R-CNN网络和Cascade R-CNN网络都是常用的R-CNN神经网络结构，Faster R-CNN网络使用RPN层来替换传统的选择性搜索算法，使得最终的检测速率大幅度提升，但存在如何选择IoU阈值的问题；Cascade R-CNN网络，通过连接多个检测网络来改进模型训练后的预测结果，其中通过设置不同IoU阈值来确定正例样本和负例样本的训练数量来训练检测网络，因此本实施例结合Faster R-CNN网络和Cascade R-CNN网络构建卷积神经网络；
传统目标检测算法在预测的时候只利用到最顶层的特征图谱，虽然顶层的特征语义信息丰富，但同时也丧失部分待检测目标的精确位置信息，而低层特征当中却拥有大量精确位置信息，若能合理使用低层特征信息，有利于提升对小物体检测的精度，FPN网络结构的顶层特征输出后不做预测，而是经过上采样之后再与低层特征做融合，最后再进行预测，同时保持每一层之间都是独立预测。通过这种网络结构能够保留低层和高层特征中的有效信息，从而，结合了每一层的特征信息，使所有尺度下的特征都具有丰富的语义信息；
在卷积神经网络中加入ResNeXt残差网络结构使用平行堆叠的残差结构块，在参数数量没有太大改变的情况下去提升检测效果的准确率，然后将网络的输入以及输出相加得到的特征图谱作为残差网络的输出，能够实现在不明显增加参数数量的情况下提升网络模型精确度的目的，同时还因为ResNeXt网络的残差块结构相同，训练网络相应的超参数也会减少，更有利于模型的移植；
本发明实施例在卷积神经网络中搭配FPN网络结构来提取更好的特征信息，再结合ResNeXt残差网络结构来增加整体网络的深度，以此改进模型提取深层次特征的能力，进而提高最终的检测效果。
参照图4，在本实施例中，步骤S130将待检测的酒瓶表面图片输入到深度 学习神经网络，以获得待检测的酒瓶表面图片对应的表面缺陷信息包括以下步骤：
S131、基于深度学习神经网络中的检测矩形框将酒瓶表面图片中的表面缺陷框出；
S132、根据检测矩形框与预设的标注矩形框计算IoU值，
S133、若IoU值大于预设的IoU阈值，则判断为S134、深度学习神经网络检测成功，否则判断为S135、深度学习神经网络检测失败。
首先，深度学习神经网络用检测矩形框框出待检测图片中的表面缺陷，并且人为事先用标注矩形框框出待检测图片中的表面缺陷，标注矩形框表示表面缺陷的实际位置，然后计算检测矩形框与标注矩形框的IoU值，将IoU值与预设的IoU阈值进行比较，若IoU值大于IoU阈值，则说明检测矩形框与标注矩形框吻合，深度学习神经网络已成功找出待检测图片中的表面缺陷，否则说明检测矩形框与标注矩形框不吻合，深度学习神经网络检测失败。
参照图3，在本实施例中，酒瓶表面图片包含至少一个表面缺陷信息。如果训练数据集中存在了大量的正常图片，正常图片中并没有需要检测的瓶装酒瓶表面缺陷信息，对模型的训练并没有任何帮助，同时带缺陷的工件也包含正常图片的特征，因此若在训练模型的过程中训练大量的正常图片会造成对模型检测工件表面缺陷能力有所下降，因此为了保证卷积神经网络的检测能力，训练数据集中的图片至少包含一个表面缺陷特征。
参照图5，在本实施例中，深度学习神经网络包括卷积层、RPN层、第一池化层、第二池化层、第三池化层、第一全连接层、第二全连接层、第三全连接层，第一全连接层的输出信息为第二池化层的输入信息，第二全连接层的输出信息为第三池化层的输入信息。将待检测图片的输入深度学习神经网络后，经过卷积层形成特征图谱，并且经过RPN层生成推荐区域，推荐区域表示待检测图片中的表面缺陷的初始预测位置，最后特征图谱和推荐区域一起输入第一池化层进行处理，第一全连接层、第二全连接层、第三全连接层的输出为表面缺陷的种类和表示表面缺陷位置的检测矩形框。
参照图5，在本实施例中，第一池化层、第二池化层、第三池化层预设的IoU阈值各不相同，第一全连接层、第二全连接层、第三全连接层均执行以下步骤：
基于深度学习神经网络中的检测矩形框将酒瓶表面图片中的表面缺陷框出；
根据检测矩形框与预设的标注矩形框计算IoU值，
若IoU值大于的IoU阈值，则输出检测矩形框。第一池化层、第二池化层、第三池化层处理信息的IoU阈值被设置为不同值，数值从低到高，IoU阈值表示对检测矩形框的准确度要求，其数值越高，对检测矩形框的准确度要求越高，由于第二池化层预设的IoU阈值比第一池化层预设的IoU阈值高，所以第一池化层和第一全连接层通过调整检测矩形框的位置，给第二池化层输出一个适合更高IoU阈值要求的检测矩形框，从而第二全连接层输出的检测矩形框与第一全连接层输出的检测矩形框相比，其准确度更高，更接近图片中表面缺陷的实际位置，第二全连接层与第三池化层的工作原理同上，通过级联的池化层与全连接层可使每次检测矩形框的精度都有所提高，越来越接近图片中表面缺陷的实际位置，从而提升深度学习神经网络的检测精度。
在本实施例中，表面缺陷信息包括以下任意一种：
表面缺陷的类别；表面缺陷的类别名；表示表面缺陷的位置对应的检测矩形框；深度学习神经网络的置信度排名及数值；判定结果。本实施例涉及的瓶装酒瓶表面缺陷类别包含瓶盖破损、瓶盖变形、瓶盖坏边、瓶盖打旋、瓶盖断点、标贴歪斜、标贴起皱、标贴气泡、喷码异常、喷码正常共十类；深度学习神经网络的置信度排名及其数值表示当前模型的准确度，可以作为提高深度学习神经网络的检测准确度的参考数据；判定结果表示卷积神经网络是否已准确找出当前图片中的表面缺陷。
本发明第三方面实施例的计算机可读存储介质，存储有计算机可执行指令，计算机可执行指令用于执行如上述第二方面实施例所述的酒瓶表面缺陷检测方法。
上面结合附图对本发明实施例作了详细说明，但是本发明不限于上述实施例，在所述技术领域普通技术人员所具备的知识范围内，还可以在不脱离本发明宗旨的前提下作出各种变化。

Claims (10)

  1. 一种酒瓶表面缺陷检测方法，其特征在于，包括：
    基于所获取到的酒瓶表面图片得到训练数据集；
    构建卷积神经网络并基于所述训练数据集训练所述卷积神经网络，得到深度学习神经网络；
    将待检测的酒瓶表面图片输入到所述深度学习神经网络，以获得待检测的所述酒瓶表面图片对应的表面缺陷信息。
  2. 根据权利要求1所述的一种酒瓶表面缺陷检测方法，其特征在于，所述基于所获取到的酒瓶表面图片得到训练数据集之后，还包括以下步骤：
    对所述酒瓶表面图片进行Y轴镜像翻转；
    对所述酒瓶表面图片进行增强颜色；
    对所述酒瓶表面图片进行增强对比度。
  3. 根据权利要求1所述的一种酒瓶表面缺陷检测方法，其特征在于，所述构建卷积神经网络并基于所述训练数据集训练所述卷积神经网络，得到深度学习神经网络，包括以下步骤：
    基于Faster R-CNN网络和Cascade R-CNN网络构建卷积神经网络；
    基于ResNeXt残差网络结构和FPN网络结构优化所述卷积神经网络，得到第一卷积神经网络；
    基于所述训练数据集训练所述第一卷积神经网络，得到所述深度学习神经网络。
  4. 根据权利要求1所述的一种酒瓶表面缺陷检测方法，其特征在于，所述将待检测的酒瓶表面图片输入到所述深度学习神经网络，以获得待检测的所述酒瓶表面图片对应的表面缺陷信息包括以下步骤：
    基于所述深度学习神经网络中的检测矩形框将所述酒瓶表面图片中的表面缺陷框出；
    根据所述检测矩形框与预设的标注矩形框计算IoU值，若所述IoU值大于预设的IoU阈值，则判断为所述深度学习神经网络检测成功，否则判断为所述深度学习神经网络检测失败。
  5. 根据权利要求1所述的一种酒瓶表面缺陷检测方法，其特征在于，所述酒瓶表面图片包含至少一个所述表面缺陷信息。
  6. 根据权利要求1所述的一种酒瓶表面缺陷检测方法，其特征在于，所述深度学习神经网络包括卷积层、RPN层、第一池化层、第二池化层、第三池化层、第一全连接层、第二全连接层、第三全连接层，所述第一全连接层的输出信息为所述第二池化层的输入信息，所述第二全连接层的输出信息为所述第三池化层的输入信息。
  7. 根据权利要求6所述的一种酒瓶表面缺陷检测方法，其特征在于，所述第一池化层、所述第二池化层、所述第三池化层预设的所述IoU阈值各不相同，所述第一全连接层、所述第二全连接层、所述第三全连接层均执行以下步骤：
    基于所述深度学习神经网络中的所述检测矩形框将所述酒瓶表面图片中的表面缺陷框出；
    根据所述检测矩形框与预设的所述标注矩形框计算IoU值，若所述IoU值大于预设的IoU阈值，则输出所述检测矩形框。
  8. 根据权利要求4所述的一种酒瓶表面缺陷检测方法，其特征在于，所述表面缺陷信息包括以下任意一种：
    所述表面缺陷的类别；所述表面缺陷的类别名；表示所述表面缺陷的位置对应的检测矩形框；所述深度学习神经网络的置信度排名及数值；判定结果。
  9. 一种电子装置，包括：存储器、处理器及存储在所述存储器上并可在处理器上运行的计算机程序，其特征在于：所述处理器执行所述程序时实现如权利要求1至8中任一项所述的酒瓶表面缺陷检测方法。
  10. 计算机可读存储介质，存储有计算机可执行指令，其特征在于：所述计算机可执行指令用于执行如权利要求1至8中任一项所述的酒瓶表面缺陷检测方法。
PCT/CN2020/112555 2020-05-22 2020-08-31 酒瓶表面缺陷检测方法、电子装置及存储介质 WO2021232613A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010440253.6 2020-05-22
CN202010440253.6A CN111709456A (zh) 2020-05-22 2020-05-22 酒瓶表面缺陷检测方法、电子装置及存储介质

Publications (1)

Publication Number Publication Date
WO2021232613A1 true WO2021232613A1 (zh) 2021-11-25

Family

ID=72538133

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/112555 WO2021232613A1 (zh) 2020-05-22 2020-08-31 酒瓶表面缺陷检测方法、电子装置及存储介质

Country Status (2)

Country Link
CN (1) CN111709456A (zh)
WO (1) WO2021232613A1 (zh)

Cited By (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113870263A (zh) * 2021-12-02 2021-12-31 湖南大学 一种路面缺陷损伤实时监测方法及系统
CN114113109A (zh) * 2021-11-26 2022-03-01 科士恩智能科技(上海)有限公司 一种自动化和智能化的汽车仪表台缺陷检测方法
CN114463284A (zh) * 2022-01-14 2022-05-10 阿丘机器人科技(苏州)有限公司 Pcb缺陷检测方法、装置、设备及存储介质
CN114529529A (zh) * 2022-02-21 2022-05-24 清华大学 基于图像仿真增强的花边布匹表面缺陷检测方法及装置
CN114627089A (zh) * 2022-03-21 2022-06-14 成都数之联科技股份有限公司 缺陷识别方法、装置、计算机设备及计算机可读存储介质
CN114647762A (zh) * 2022-03-23 2022-06-21 中国水利水电科学研究院 一种基于图谱对比的堤坝探测方法
CN114792300A (zh) * 2022-01-27 2022-07-26 河南大学 一种基于多尺度注意力x光断针检测方法
CN114820497A (zh) * 2022-04-20 2022-07-29 深圳市国电科技通信有限公司 污渍检测方法、装置、系统及存储介质、电子设备
CN115063725A (zh) * 2022-06-23 2022-09-16 中国民航大学 基于多尺度自适应ssd算法的飞机蒙皮缺陷识别系统
CN115330702A (zh) * 2022-08-01 2022-11-11 无锡雪浪数制科技有限公司 一种基于深度视觉的饮料瓶灌装缺陷识别方法
CN116452791A (zh) * 2023-03-27 2023-07-18 广州市斯睿特智能科技有限公司 多相机点位的缺陷区域定位方法、系统、装置及存储介质

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112581393B (zh) * 2020-12-15 2022-04-26 湖北工业大学 基于语义分割的旋开盖图像处理方法
CN113324993B (zh) * 2021-05-19 2022-09-27 吉林大学 一种全方位的药瓶外观缺陷检测方法

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109598721A (zh) * 2018-12-10 2019-04-09 广州市易鸿智能装备有限公司 电池极片的缺陷检测方法、装置、检测设备和存储介质
CN109767427A (zh) * 2018-12-25 2019-05-17 北京交通大学 列车轨道扣件缺陷的检测方法
CN110232687A (zh) * 2019-06-21 2019-09-13 华北电力大学（保定） 一种电力巡检图像中带销螺栓缺陷的检测方法
CN110865077A (zh) * 2019-11-15 2020-03-06 上海电器科学研究所（集团）有限公司 用于rfid天线生产的外观缺陷视觉检测系统
WO2020059446A1 (ja) * 2018-09-20 2020-03-26 富士フイルム株式会社 学習装置及び学習方法
CN111709451A (zh) * 2020-05-21 2020-09-25 五邑大学 酒瓶表面缺陷检测方法、电子装置及存储介质

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2020059446A1 (ja) * 2018-09-20 2020-03-26 富士フイルム株式会社 学習装置及び学習方法
CN109598721A (zh) * 2018-12-10 2019-04-09 广州市易鸿智能装备有限公司 电池极片的缺陷检测方法、装置、检测设备和存储介质
CN109767427A (zh) * 2018-12-25 2019-05-17 北京交通大学 列车轨道扣件缺陷的检测方法
CN110232687A (zh) * 2019-06-21 2019-09-13 华北电力大学（保定） 一种电力巡检图像中带销螺栓缺陷的检测方法
CN110865077A (zh) * 2019-11-15 2020-03-06 上海电器科学研究所（集团）有限公司 用于rfid天线生产的外观缺陷视觉检测系统
CN111709451A (zh) * 2020-05-21 2020-09-25 五邑大学 酒瓶表面缺陷检测方法、电子装置及存储介质

Cited By (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114113109A (zh) * 2021-11-26 2022-03-01 科士恩智能科技(上海)有限公司 一种自动化和智能化的汽车仪表台缺陷检测方法
CN113870263B (zh) * 2021-12-02 2022-02-25 湖南大学 一种路面缺陷损伤实时监测方法及系统
CN113870263A (zh) * 2021-12-02 2021-12-31 湖南大学 一种路面缺陷损伤实时监测方法及系统
CN114463284A (zh) * 2022-01-14 2022-05-10 阿丘机器人科技(苏州)有限公司 Pcb缺陷检测方法、装置、设备及存储介质
CN114792300A (zh) * 2022-01-27 2022-07-26 河南大学 一种基于多尺度注意力x光断针检测方法
CN114792300B (zh) * 2022-01-27 2024-02-20 河南大学 一种基于多尺度注意力x光断针检测方法
CN114529529A (zh) * 2022-02-21 2022-05-24 清华大学 基于图像仿真增强的花边布匹表面缺陷检测方法及装置
CN114529529B (zh) * 2022-02-21 2024-04-09 清华大学 基于图像仿真增强的花边布匹表面缺陷检测方法及装置
CN114627089A (zh) * 2022-03-21 2022-06-14 成都数之联科技股份有限公司 缺陷识别方法、装置、计算机设备及计算机可读存储介质
CN114647762A (zh) * 2022-03-23 2022-06-21 中国水利水电科学研究院 一种基于图谱对比的堤坝探测方法
CN114820497A (zh) * 2022-04-20 2022-07-29 深圳市国电科技通信有限公司 污渍检测方法、装置、系统及存储介质、电子设备
CN115063725A (zh) * 2022-06-23 2022-09-16 中国民航大学 基于多尺度自适应ssd算法的飞机蒙皮缺陷识别系统
CN115063725B (zh) * 2022-06-23 2024-04-26 中国民航大学 基于多尺度自适应ssd算法的飞机蒙皮缺陷识别系统
CN115330702A (zh) * 2022-08-01 2022-11-11 无锡雪浪数制科技有限公司 一种基于深度视觉的饮料瓶灌装缺陷识别方法
CN116452791A (zh) * 2023-03-27 2023-07-18 广州市斯睿特智能科技有限公司 多相机点位的缺陷区域定位方法、系统、装置及存储介质
CN116452791B (zh) * 2023-03-27 2024-03-22 广州市斯睿特智能科技有限公司 多相机点位的缺陷区域定位方法、系统、装置及存储介质

Also Published As

Publication number Publication date
CN111709456A (zh) 2020-09-25

Similar Documents

Publication Publication Date Title
WO2021232613A1 (zh) 酒瓶表面缺陷检测方法、电子装置及存储介质
CN109146873B (zh) 一种基于学习的显示屏缺陷智能检测方法及装置
CN111814850A (zh) 缺陷检测模型训练方法、缺陷检测方法及相关装置
WO2021232611A1 (zh) 酒瓶表面缺陷模型的评价方法、电子装置及存储介质
CN109671058B (zh) 一种大分辨率图像的缺陷检测方法及系统
KR20190063839A (ko) 제조 공정에서 딥러닝을 활용한 머신 비전 기반 품질검사 방법 및 시스템
US11836967B2 (en) Method and device for small sample defect classification and computing equipment
CN105654067A (zh) 一种车辆检测方法及装置
CN111199543A (zh) 一种基于卷积神经网络的冰柜表面缺陷检测
CN111709451B (zh) 酒瓶表面缺陷检测方法、电子装置及存储介质
CN111209907A (zh) 一种复杂光污染环境下产品特征图像人工智能识别方法
CN113378969B (zh) 一种目标检测结果的融合方法、装置、设备及介质
CN113592832A (zh) 一种工业品缺陷检测方法和装置
CN115546211B (zh) 一种焊点缺陷分类方法、终端及计算机存储介质
US20200090319A1 (en) Machine learning method implemented in aoi device
CN115471476A (zh) 一种部件缺陷检测方法、装置、设备及介质
CN111709452B (zh) 酒瓶表面缺陷模型的评价方法、电子装置及存储介质
Li et al. Surface defect detection algorithm of aluminum based on improved faster RCNN
CN204429717U (zh) 一种应用在瓶装饮料生产设备中的瓶盖检测筛选装置
CN116309471A (zh) 一种质量缺陷预警方法、装置、电子设备以及介质
CN107578168B (zh) 一种用于缺陷库移植的方法、装置及电子设备
CN116030013A (zh) 多目标多类别破损缺陷检测方法、系统及计算机存储介质
Tatu et al. Fault Detection In Bottle Caps And Label Alignment Using Convolutional Neural Network
CN110728310B (zh) 一种基于超参数优化的目标检测模型融合方法及融合系统
CN112348102A (zh) 一种基于查询的自底向上视频定位方法和系统

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: 20936357

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: 20936357

Country of ref document: EP

Kind code of ref document: A1