WO2021235061A1 - 画像分類装置、画像分類方法、及び、画像分類プログラム - Google Patents

画像分類装置、画像分類方法、及び、画像分類プログラム Download PDF

Info

Publication number
WO2021235061A1
WO2021235061A1 PCT/JP2021/010287 JP2021010287W WO2021235061A1 WO 2021235061 A1 WO2021235061 A1 WO 2021235061A1 JP 2021010287 W JP2021010287 W JP 2021010287W WO 2021235061 A1 WO2021235061 A1 WO 2021235061A1
Authority
WO
WIPO (PCT)
Prior art keywords
label
image
classification
model
classified
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/JP2021/010287
Other languages
English (en)
French (fr)
Japanese (ja)
Inventor
均 服部
理也 栗原
一男 米倉
幸二 徳永
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.)
IHI Corp
Original Assignee
IHI Corp
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 IHI Corp filed Critical IHI Corp
Priority to EP21809682.4A priority Critical patent/EP4156092A4/en
Priority to JP2022524901A priority patent/JP7371776B2/ja
Publication of WO2021235061A1 publication Critical patent/WO2021235061A1/ja
Priority to US17/811,596 priority patent/US20220343632A1/en
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
    • 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
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/776—Validation; Performance evaluation
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/70—Labelling scene content, e.g. deriving syntactic or semantic representations

Definitions

  • This disclosure relates to an image classification device, an image classification method, and an image classification program.
  • Patent Document 1 discloses a system capable of constructing a database of image data by a user grouping captured images while referring to the captured images.
  • the present disclosure automates the classification work of images of similar components, which is difficult to improve the accuracy of classification without confirming the detailed features of the components, and reduces the work time and cost in the classification work. It is an object of the present invention to provide an image classification device, an image classification method, and an image classification program which can be used.
  • the image classification device includes a receiving unit that receives an image obtained by capturing an image of an object, and a controller that determines an image classification label based on the first model and the second model.
  • the first model is a model generated based on the first teacher data in which the classified image and the classification label of the classified image are paired.
  • the second model is a model generated based on the second teacher data, which is a set of the classified image, the classification label of the classified image, and the area set in the classified image.
  • the controller calculates the first label based on the image and the first model, and determines whether or not the first label is a predetermined label. Then, when the first label is not the predetermined label, the first label is set as the classification label of the image.
  • the second label is calculated based on the image and the second model, and the second label is set as the classification label of the image.
  • the second model may be a model generated based only on the second teacher data in which the classification label of the classified image is the predetermined label.
  • the classification label of the classified image whose correct answer rate is equal to or less than the predetermined threshold value may be the predetermined label.
  • the classified image and the label calculated based on the first model are used as reproduction labels, and the reproduction label calculated for each classification label of the classified image and the classification label of the classified image match.
  • the percentage of correct answers may be the correct answer rate.
  • the first model may be a model generated by machine learning based on the first teacher data.
  • the area may be an area in which the characteristic portion of the object in the image is reflected.
  • the second model may be a model that estimates the region from the image using a detection algorithm.
  • the above detection algorithm is an object detection method (Faster R-CNN (Regions with Convolutional Neural Networks), YOLO (You Only Look None), SSD (Single Shot MultiBox, at least one of the algorithms, Segment). It may be included.
  • the above object may be a component of a machine.
  • the machine may be, for example, an aircraft engine.
  • the classification label of the image obtained by capturing the object is determined based on the first model and the second model.
  • the first model is a model generated based on the first teacher data in which the classified image and the classification label of the classified image are paired.
  • the second model is a model generated based on the second teacher data, which is a set of the classified image, the classification label of the classified image, and the area set in the classified image.
  • the image classification method calculates a first label based on the image and the first model, and determines whether or not the first label is a predetermined label. Then, when the first label is not the predetermined label, the first label is set as the classification label of the image. On the other hand, when the first label is the predetermined label, the second label is calculated based on the image and the second model, and the second label is set as the classification label of the image.
  • the image classification program determines the classification label of the image obtained by capturing the object based on the first model and the second model.
  • the first model is a model generated based on the first teacher data in which the classified image and the classification label of the classified image are paired.
  • the second model is a model generated based on the second teacher data, which is a set of the classified image, the classification label of the classified image, and the area set in the classified image.
  • the image classification program causes a computer to calculate a first label based on the image and the first model, and determines whether or not the first label is a predetermined label. Then, when the first label is not the predetermined label, the first label is set as the classification label of the image. On the other hand, when the first label is the predetermined label, the second label is calculated based on the image and the second model, and the second label is set as the classification label of the image.
  • FIG. 1 is a block diagram showing a configuration of an image classification device.
  • the image classification device 20 includes a receiving unit 21, a database 23, a controller 25, and an output unit 27.
  • the controller 25 is connected so as to be able to communicate with the receiving unit 21, the database 23, and the output unit 27.
  • the output unit 27 may be provided by the image classification device 20 itself, or may be installed outside the image classification device 20 and connected to the image classification device 20 by a wireless or wired network. good.
  • the receiving unit 21 is connected wirelessly or by wire so as to be able to communicate with the image pickup device 10.
  • the receiving unit 21 receives an image of an object captured by the imaging device 10.
  • the receiving unit 21 may receive a time stamp indicating the date and time when the image was acquired together with the image.
  • the database 23 records the first teacher data and the second teacher data for the image for which the classification label has already been set by a method different from that of the image classification device 20 (hereinafter referred to as the classified image). good.
  • the first teacher data is data in which a classified image and a classification label of the classified image are paired.
  • the second teacher data is data in which the classified image, the classification label of the classified image, and the area set in the classified image are set as a set.
  • the image classification label is a label set for the image and represents a group to which the image belongs.
  • the classification label is the name of the object that appears in the image. If the object in the image is a component of the machine, the classification label may be the name of the component of the machine. More specifically, when the object shown in the image is a component constituting the aircraft engine, the classification label may be the name of the component constituting the aircraft engine.
  • Various names of parts constituting an aircraft engine include, for example, a fan rotor, an inlet cone, an inner shroud, a blade, and the like.
  • the names of the objects in the image, especially the names of the parts that make up the aircraft engine, are not limited to the examples given here.
  • the area set in the image is an area set on the image, and is an area in which the characteristic portion of the object reflected in the image is reflected.
  • the area set in the image may be an area composed of only the pixels corresponding to the characteristic portion of the object on the image.
  • the region set in the image may be a region including pixels corresponding to the characteristic portion of the object on the image.
  • the area set in the image may be a rectangular area including pixels corresponding to the characteristic portion of the object on the image.
  • the area set in the image may be an area having a polygon having a plurality of vertices as a boundary.
  • the characteristic part of an object is a part that distinguishes the object from other objects. That is, the characteristic portion of the object of interest is a dissimilar portion between the object of interest and an object other than the object of interest.
  • the second teacher data may be composed only of data in which the classification label of the classified image is a predetermined label.
  • the predetermined label will be described later.
  • the database 23 may record the image received by the receiving unit 21. Further, the database 23 may record the first model and the second model described later.
  • the output unit 27 outputs the information generated by the controller 25, which will be described later.
  • the output unit 27 outputs the classification label set for each image by the controller 25 to the user or the like.
  • the output unit 27 may be a display that presents information to the user by displaying figures and characters by combining a plurality of display pixels.
  • the output unit 27 may be a speaker that notifies the user of information by voice.
  • the method of outputting information by the output unit 27 is not limited to the examples given here.
  • the controller 25 (control unit) is a general-purpose microcomputer including a CPU (central processing unit), a memory, and an input / output unit.
  • a computer program (image classification program) for functioning as the image classification device 20 is installed in the controller 25.
  • the controller 25 By executing the computer program, the controller 25 functions as a plurality of information processing circuits (251, 255, 255, 257) included in the image classification device 20.
  • the computer program (image classification program) may be stored in a storage medium that can be read and written by a computer.
  • This disclosure shows an example of realizing a plurality of information processing circuits (251, 255, 255, 257) by software.
  • an information processing circuit (251, 253, 255, 257) by preparing dedicated hardware for executing each of the following information processing.
  • a plurality of information processing circuits (251, 255, 255, 257) may be configured by individual hardware.
  • the information processing circuit (251, 255, 255, 257) may also be used as a control unit used for monitoring or controlling the image pickup apparatus 10.
  • the controller 25 has a plurality of information processing circuits (251, 255, 255, 257) as a first label calculation unit 251, a second label calculation unit 253, a determination unit 255, and a classification label setting unit 257. Equipped with.
  • the first label calculation unit 251 performs "learning” based on the first teacher data, and then performs “estimation” of the label based on the image in which the classification label is not set.
  • the second label calculation unit 253 performs "learning” based on the second teacher data, and then “estimates” the label based on the image to which the classification label is not set.
  • the first label calculation unit 251 performs machine learning based on the first teacher data and generates the first model.
  • the second label calculation unit 253 performs machine learning based on the second teacher data and generates the second model.
  • the generated first model and the second model may be those stored in the database 23.
  • the first label calculation unit 251 and the second label calculation unit 253 generate the first model and the second model, respectively, using the neural network.
  • the first label calculation unit 251 generates a first model using a first neural network that inputs an image and outputs a label. At that time, the first label calculation unit 251 calculates an error between the label obtained when the classified image is input to the first neural network and the classification label corresponding to the input classified image.
  • the first label calculation unit 251 adjusts the parameters that define the first neural network so that the error is minimized, and learns the feature expressing the first teacher data.
  • the first model is represented by the first neural network.
  • the second label calculation unit 253 generates the second model by using the second neural network that inputs the image and outputs the label and the area information (information indicating the area set on the image). At that time, the second label calculation unit 253 calculates an error between the label obtained when the classified image is input to the second neural network and the classification label corresponding to the input classified image.
  • the second label calculation unit 253 calculates an error between the area information obtained when the classified image is input to the second neural network and the area information set in the input classified image.
  • the second label calculation unit 253 adjusts the parameters that define the second neural network so that the error related to the label and the error related to the area information are minimized, and learns the feature expressing the second teacher data.
  • the second model is represented by the second neural network.
  • the second label calculation unit 253 may calculate the area information from the image input to the second neural network by using the detection algorithm. That is, the second model generated by the second label calculation unit 253 may be a model that estimates the region from the image using a detection algorithm.
  • Examples of the detection algorithm include an object detection method or semantic segmentation.
  • Examples of the object detection method include Faster R-CNN (Regions with Convolutional Neural Networks), YOLO (You Only Look Noise), SSD (Single Shot MultiBox), and the like.
  • examples of the detection algorithm include those containing these algorithms internally. The detection algorithm is not limited to the examples given here.
  • the above-mentioned neural network includes an input layer into which an image is input, an output layer in which an output value is output, and at least one hidden layer provided between the input layer and the output layer, and includes an input layer and a hidden layer.
  • the signal propagates in the order of layer and output layer.
  • Each layer of the input layer, the hidden layer, and the output layer is composed of one or more units.
  • the units between the layers are connected to each other, and each unit has an activation function (for example, a sigmoid function, a rectified linear function, a softmax function, etc.).
  • a weighted sum is calculated based on multiple inputs to the unit, and the value of the activation function with the sum value as a variable is the output of the unit.
  • the first label calculation unit 251 and the second label calculation unit 253 adjust the weights when calculating the weighted total in each unit among the parameters defining the neural network. Then, the first label calculation unit 251 and the second label calculation unit 253 minimize the error between the output of the neural network and the classification data.
  • the maximum likelihood estimation method or the like can be applied to minimize the error related to the output of the neural network for a plurality of teacher data.
  • the first label calculation unit 251 and the second label calculation unit 253 may use a gradient descent method, a stochastic gradient descent method, or the like.
  • the first label calculation unit 251 and the second label calculation unit 253 may use an error back propagation method for gradient calculation by the gradient descent method or the stochastic gradient descent method.
  • a method such as regularization that restricts the degree of freedom of weights at the time of learning may be used in order to alleviate overfitting.
  • a method such as a dropout that probabilistically selects units in the neural network and invalidates other units may be used.
  • methods such as data regularization, data standardization, and data expansion that eliminate bias in teacher data may be used.
  • the first label calculation unit 251 estimates the classification label for an image whose classification label is unknown by using the first model generated by the first teacher data. That is, the first label calculation unit 251 inputs an image to the first neural network representing the first model, and calculates the output of the first neural network. Then, the output of the first neural network is used as the first label (classification label estimated by the first model).
  • the second label calculation unit 253 estimates the classification label for an image whose classification label is unknown by using the second model generated by the second teacher data. That is, the second label calculation unit 253 inputs an image to the second neural network representing the second model, and calculates the output of the second neural network. Then, the output of the second neural network is used as the second label (classification label estimated by the second model).
  • Estimates based on the second model tend to have higher calculation costs than estimates based on the first model.
  • the estimation based on the second model tends to enable finer estimation than the estimation based on the first model. The reason for this is that, as compared with the first model, in the second model, the calculation regarding the characteristic portion of the object reflected in the image is performed.
  • Which of the first label and the second label is set as the classification label for the image for which the classification label is not set is determined by the processing in the determination unit 255 and the classification label setting unit 257 described below.
  • the first label calculation unit 251 may use the first model to estimate the classification label for the classified image and calculate the correct answer rate of the first model. That is, the first label calculation unit 251 uses the classified image and the label calculated based on the first model as the reproduction label, and sets the ratio at which the reproduction label and the classification label of the classified image match as the classification label of the classified image. It may be calculated for each. The correct answer rate is calculated for each classification label of the classified image.
  • the above-mentioned correct answer rate indicates the degree to which the first model reproduces the first teacher data.
  • the correct answer rate represents the probability that the classification label corresponding to the input classified image is output as a reproduction label.
  • the first label calculation unit 251 may set the classification label of the classified image whose correct answer rate is equal to or less than the predetermined threshold value as the predetermined label.
  • the predetermined label may be set based on the correct answer rate, or may be arbitrarily set by the user of the image classification device 20. The method of setting the predetermined label is not limited to the example given here.
  • the reason for setting the predetermined label is to construct the second teacher data by using the first teacher data to a small degree of reproduction by the first model.
  • the classification label is used for the image whose classification label is unknown using the second model. Is estimated.
  • the determination unit 255 determines whether or not the first label calculated by the first label calculation unit 251 is a predetermined label. Then, the result of the determination is output to the classification label setting unit 257. The determination unit 255 may control the second label calculation unit 253 so that the second label is calculated when the first label calculated by the first label calculation unit 251 is a predetermined label.
  • the classification label setting unit 257 sets one of the first label and the second label as the classification label for the image for which the classification label is not set, based on the determination result by the determination unit 255. Specifically, when the first label is not a predetermined label, the classification label setting unit 257 sets the first label as the classification label. On the other hand, when the first label is a predetermined label, the classification label setting unit 257 sets the second label as the classification label.
  • the processing of the flowchart shown in FIG. 2 is started when the user activates the image classification device. It is assumed that the first model and the second model have already been generated at the time when the image classification by the image classification device is started.
  • step S101 the receiving unit 21 receives the image of the object captured by the imaging device 10.
  • step S103 the first label calculation unit 251 calculates the first label based on the image and the first model. That is, the first label calculation unit 251 inputs an image to the first neural network representing the first model, and calculates the output of the first neural network. Then, the output of the first neural network is used as the first label.
  • step S105 the determination unit 255 determines whether or not the first label calculated by the first label calculation unit 251 is a predetermined label.
  • step S105 If it is determined in step S105 that the first label is not a predetermined label (NO in step S105), the process proceeds to step S107, and the classification label setting unit 257 sets the first label as the classification label for the image. do.
  • step S105 if it is determined in step S105 that the first label is a predetermined label (YES in step S105), the process proceeds to step S111, and the second label calculation unit 253 is based on the image and the second model. Calculate the second label. That is, the second label calculation unit 253 inputs an image to the second neural network representing the second model, and calculates the output of the second neural network. Then, the output of the second neural network is used as the second label.
  • step S113 the classification label setting unit 257 sets the second label as the classification label for the image.
  • step S107 or step S113 After performing the processing in step S107 or step S113, the processing of image classification shown in FIG. 2 is completed.
  • the classification label is estimated for the image whose classification label is unknown, and one of the first label and the second label is set as the classification label. Will be done.
  • the image classification device, the image classification method, and the image classification program according to the present disclosure determine the classification label of the image obtained by capturing the object based on the first model and the second model.
  • the first model is a model generated based on the first teacher data in which the classified image and the classification label of the classified image are paired.
  • the second model is a model generated based on the second teacher data, which is a set of the classified image, the classification label of the classified image, and the area set in the classified image.
  • the image classification method calculates a first label based on the image and the first model, and determines whether or not the first label is a predetermined label.
  • the first label is set as the classification label of the image.
  • the second label is calculated based on the image and the second model, and the second label is set as the classification label of the image.
  • the time required for the classification work of the captured images acquired for the maintenance and inspection of a machine having a large number of component parts can be shortened. Furthermore, it is not necessary to train workers who are proficient in the work of classifying captured images.
  • an aircraft engine is composed of parts as shown in FIGS. 3A, 3B, 3C, and 3D.
  • the fan rotor shown in FIG. 3A and the inlet cone shown in FIG. 3B are significantly different in appearance from the inner shrouds shown in FIGS. 3C and 3D. Therefore, by rough estimation based on the first model, it can be estimated that the component reflected in FIG. 3A is a fan rotor, and the component reflected in FIG. 3B is an inlet cone.
  • the parts reflected in FIGS. 3C and 3D are inner shrouds.
  • a "first inner shroud” and a “second inner shroud” are set as predetermined labels, and the parts reflected in FIGS. 3C and 3D are estimated separately from each other by detailed estimation based on the second model. ..
  • the area R1 is set in the image in which the first inner shroud is reflected in the second teacher data as shown in FIG. 3C. Further, in the image in which the second inner shroud is reflected, the area R2 is set as shown in FIG. 3D. Therefore, according to the detailed estimation based on the second model, the parts reflected in FIGS. 3C and 3D can be estimated separately from each other.
  • the names of the parts that make up the aircraft engine are not limited to the above examples. Further, the setting of the predetermined label is not limited to the above-mentioned example.
  • the second model may be a model generated based only on the second teacher data in which the classification label of the classified image is the predetermined label.
  • the second model becomes a model specialized in the classification of images that cannot be sufficiently classified by the classification based on the first model.
  • the learning time based on the second teacher data can be shortened.
  • the classification label of the classified image whose correct answer rate is equal to or less than the predetermined threshold value may be the predetermined label.
  • the classified image and the label calculated based on the first model are used as reproduction labels, and the reproduction label calculated for each classification label of the classified image and the classification label of the classified image match.
  • the percentage of correct answers may be the correct answer rate.
  • the first model may be a model generated by machine learning based on the first teacher data. This makes it possible to make a rough estimation of the image. In addition, it is possible to shorten the learning time for a sufficiently classified image by rough estimation based on the first model.
  • the area may be an area in which the characteristic portion of the object in the image is reflected. This makes it possible to improve the classification accuracy by fine estimation by the second model.
  • the second model may be a model that estimates the region from the image using a detection algorithm. This allows the second model to make finer estimates than the first model.
  • the detection algorithm includes Faster R-CNN (Regions with Convolutional Neural Networks), YOLO (You Only Look None), SSD (Single Shot MultiBox At least Semantic), and a semantic algorithm including SSD (Single Shot MultiBox Detector). May be good. This allows the second model to make finer estimates than the first model.
  • the above object may be a component of a machine. As a result, it is possible to shorten the time required for sorting the captured images acquired for maintenance and inspection of a machine having a large number of component parts. Furthermore, it is not necessary to train workers who are proficient in the work of classifying captured images.
  • the above machine may be an aircraft engine. As a result, it is possible to shorten the time required for sorting the captured images acquired for maintenance and inspection of an aircraft engine having a large number of components. Furthermore, it is not necessary to train workers who are proficient in the work of classifying captured images.
  • Processing circuits include programmed processors, electrical circuits, etc., as well as devices such as application specific integrated circuits (ASICs), or circuit components arranged to perform the described functions. Etc. are also included.
  • the classification work of images of similar components can be automated to reduce the work time and cost in the classification work, and thus, for example, United Nations-led sustainable development. It can contribute to Goal 12 “Ensuring sustainable production and consumption patterns” of the Goals (SDGs).

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Multimedia (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Databases & Information Systems (AREA)
  • Computing Systems (AREA)
  • Artificial Intelligence (AREA)
  • Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Image Analysis (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
PCT/JP2021/010287 2020-05-21 2021-03-15 画像分類装置、画像分類方法、及び、画像分類プログラム Ceased WO2021235061A1 (ja)

Priority Applications (3)

Application Number Priority Date Filing Date Title
EP21809682.4A EP4156092A4 (en) 2020-05-21 2021-03-15 Image classification device, image classification method, and image classification program
JP2022524901A JP7371776B2 (ja) 2020-05-21 2021-03-15 画像分類装置、画像分類方法、及び、画像分類プログラム
US17/811,596 US20220343632A1 (en) 2020-05-21 2022-07-11 Image classification device, image classification method, and image classification program

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
JP2020088815 2020-05-21
JP2020-088815 2020-05-21

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US17/811,596 Continuation US20220343632A1 (en) 2020-05-21 2022-07-11 Image classification device, image classification method, and image classification program

Publications (1)

Publication Number Publication Date
WO2021235061A1 true WO2021235061A1 (ja) 2021-11-25

Family

ID=78708450

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP2021/010287 Ceased WO2021235061A1 (ja) 2020-05-21 2021-03-15 画像分類装置、画像分類方法、及び、画像分類プログラム

Country Status (4)

Country Link
US (1) US20220343632A1 (https=)
EP (1) EP4156092A4 (https=)
JP (1) JP7371776B2 (https=)
WO (1) WO2021235061A1 (https=)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2024195055A1 (ja) * 2023-03-22 2024-09-26 日本電気株式会社 情報処理装置、情報処理方法、及び、記録媒体

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2023079703A1 (ja) * 2021-11-05 2023-05-11 楽天グループ株式会社 処理実行システム、処理実行方法、及びプログラム

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2005004564A (ja) 2003-06-13 2005-01-06 Joho Kankyo Design Kk 画像分類処理システム
JP2017084320A (ja) * 2015-03-06 2017-05-18 パナソニックＩｐマネジメント株式会社 学習方法およびプログラム
JP2018081629A (ja) * 2016-11-18 2018-05-24 住友電気工業株式会社 判定装置、判定方法および判定プログラム
JP2019212073A (ja) * 2018-06-06 2019-12-12 アズビル株式会社 画像判別装置および方法
JP2020088815A (ja) 2018-11-30 2020-06-04 Ｃｏｎｎｅｃｔｅｄ Ｄｅｓｉｇｎ株式会社 管理サーバ、情報処理方法およびプログラム

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
GB201704373D0 (en) * 2017-03-20 2017-05-03 Rolls-Royce Ltd Surface defect detection

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2005004564A (ja) 2003-06-13 2005-01-06 Joho Kankyo Design Kk 画像分類処理システム
JP2017084320A (ja) * 2015-03-06 2017-05-18 パナソニックＩｐマネジメント株式会社 学習方法およびプログラム
JP2018081629A (ja) * 2016-11-18 2018-05-24 住友電気工業株式会社 判定装置、判定方法および判定プログラム
JP2019212073A (ja) * 2018-06-06 2019-12-12 アズビル株式会社 画像判別装置および方法
JP2020088815A (ja) 2018-11-30 2020-06-04 Ｃｏｎｎｅｃｔｅｄ Ｄｅｓｉｇｎ株式会社 管理サーバ、情報処理方法およびプログラム

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP4156092A4

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2024195055A1 (ja) * 2023-03-22 2024-09-26 日本電気株式会社 情報処理装置、情報処理方法、及び、記録媒体

Also Published As

Publication number Publication date
JPWO2021235061A1 (https=) 2021-11-25
US20220343632A1 (en) 2022-10-27
EP4156092A1 (en) 2023-03-29
EP4156092A4 (en) 2024-04-10
JP7371776B2 (ja) 2023-10-31

Similar Documents

Publication Publication Date Title
US20220245405A1 (en) Deterioration suppression program, deterioration suppression method, and non-transitory computer-readable storage medium
JP6798614B2 (ja) 画像認識装置、画像認識方法および画像認識プログラム
JP2015087903A (ja) 情報処理装置及び情報処理方法
CN110942015B (zh) 人群密度估计方法
JP7056259B2 (ja) 検査システム、識別システム、及び識別器評価装置
CN114399655A (zh) 目标检测方法、系统及存储介质
US20220343632A1 (en) Image classification device, image classification method, and image classification program
JPWO2019215904A1 (ja) 予測モデル作成装置、予測モデル作成方法、および予測モデル作成プログラム
CN112766427A (zh) 图像识别模型的训练方法、装置及设备
WO2021049119A1 (ja) 学習装置、学習方法及び学習プログラムが格納された非一時的なコンピュータ可読媒体
CN110276252A (zh) 一种基于生成式对抗网络的抗表情干扰的人脸识别方法
JP7070308B2 (ja) 推定器生成装置、検査装置、推定器生成方法、及び推定器生成プログラム
JP2019105871A (ja) 異常候補抽出プログラム、異常候補抽出方法および異常候補抽出装置
US20210406688A1 (en) Method and device with classification verification
US12327397B2 (en) Electronic device and method with machine learning training
JP7753782B2 (ja) 判定プログラム、判定方法および情報処理装置
WO2023053216A1 (ja) 機械学習プログラム、機械学習方法および機械学習装置
US20210319269A1 (en) Apparatus for determining a classifier for identifying objects in an image, an apparatus for identifying objects in an image and corresponding methods
WO2021111831A1 (ja) 情報処理方法、情報処理システム及び情報処理装置
JP4449483B2 (ja) 画像解析装置、および画像解析方法、並びにコンピュータ・プログラム
CN112633399B (zh) 一种稀疏协同联合表示模式识别方法
JP2024020924A (ja) 特定プログラム、特定方法および情報処理装置
US20260073663A1 (en) Data analysis apparatus, method, and non-transitory computer-readable storage medium
JP7517740B2 (ja) 製造プロセスのためのシステム、方法、および媒体
US20260011146A1 (en) System and method for generating a federated learning based model for improved video analytics

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: 21809682

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2022524901

Country of ref document: JP

Kind code of ref document: A

NENP Non-entry into the national phase

Ref country code: DE

ENP Entry into the national phase

Ref document number: 2021809682

Country of ref document: EP

Effective date: 20221221

WWW Wipo information: withdrawn in national office

Ref document number: 2021809682

Country of ref document: EP