CN114037679A - Product image defect detection method and device based on unsupervised feature combination - Google Patents

Product image defect detection method and device based on unsupervised feature combination Download PDF

Info

Publication number
CN114037679A
CN114037679A CN202111307737.4A CN202111307737A CN114037679A CN 114037679 A CN114037679 A CN 114037679A CN 202111307737 A CN202111307737 A CN 202111307737A CN 114037679 A CN114037679 A CN 114037679A
Authority
CN
China
Prior art keywords
image
network
memory
training
product
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.)
Pending
Application number
CN202111307737.4A
Other languages
Chinese (zh)
Inventor
胡亮
张聃
郑敏娥
展华益
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.)
Sichuan Qiruike Technology Co Ltd
Original Assignee
Sichuan Qiruike Technology 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 Sichuan Qiruike Technology Co Ltd filed Critical Sichuan Qiruike Technology Co Ltd
Priority to CN202111307737.4A priority Critical patent/CN114037679A/en
Publication of CN114037679A publication Critical patent/CN114037679A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • 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
    • 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
    • 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
    • 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
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • 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
    • G06N3/08—Learning methods
    • G06N3/088—Non-supervised learning, e.g. competitive 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
    • G06T2207/20081—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
    • G06T2207/20084—Artificial neural networks [ANN]
    • 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

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

Abstract

The invention discloses a product image defect detection method and a device based on unsupervised feature combination, wherein the method comprises the steps of collecting a small number of non-defective product images as training samples, and dividing the images into small blocks with equal size; dividing each image into two parts, wherein one part is an image small block, and the other part is a covered image obtained by covering the corresponding image small block; training a memory network and an image restoration network by using the image small blocks and the covering images, wherein the characteristics of the memory network and the image restoration network are combined and share a decoder; and carrying out defect detection on the image of the product to be detected by using the memory network and the image restoration network which are obtained through training. The invention uses the memory network to extract the whole structure information of the image small block, simultaneously uses the image restoration network to extract the local field information of the adjacent image small block, and finally fuses the whole information and the local information, thereby reconstructing the details of the complex image, having high detection precision and extremely low false detection rate.

Description

Product image defect detection method and device based on unsupervised feature combination
Technical Field
The invention relates to the technical field of product defect detection, in particular to a product image defect detection method and device based on unsupervised feature combination.
Background
The detection of the image defects of the products is one of the most important links for ensuring the qualified product quality. Previously, product image defects were manually detected by trained workers in a manner that was very time consuming and inefficient. Later, classical machine vision methods replaced manual work in many instances. However, with the arrival of the 4.0 era of industry, the production line is developing towards the direction of generalization, the development period of the classic machine vision method of manually designing features is long, the features are not universal, and the rapid and flexible adaptation to new products cannot be ensured. After the deep learning develops, the method gradually replaces the classic machine vision method because the method can automatically extract features and has strong generalization capability.
When a supervised deep learning model is used for detecting the defects of the product image, a sample with labeled information is needed; however, in the actual production and manufacturing process, some product defects are rarely found, samples can hardly be collected, and some product defects are marked by spending a lot of time and labor. For example, chinese patent CN202110669165.8 and chinese patent CN202011493629.6 describe supervised defect detection methods, which are expensive in sample collection and labeling cost.
When the unsupervised deep learning model is used for detecting the product image defects, no label is needed, only a defect-free training sample is needed, but the detection precision is generally inferior to that of the supervised model. Chinese patent CN202110649145.4 proposes an unsupervised defect detection scheme based on an auto-encoder, however, the reconstruction of a complex image is blurred by the scheme, so that the detection accuracy is affected; the Chinese patent CN202110415805.2 proposes to use knowledge distillation to detect defects, but the construction of an industrial data set to be distilled is time-consuming and labor-consuming; the Chinese patent CN202110196131.1 uses the generation of the countermeasure network to detect defects, but the problem of mode collapse of the generation of the countermeasure network is difficult to solve; some methods pre-train an initial network from the ImageNet dataset and then find defects on a feature level by distillation compression, as represented by the document "PadiM: a Patch Distribution Modeling Framework for analysis Detection and Localization", however this method ignores the differences between industrial products and natural images.
Disclosure of Invention
The invention provides a product image defect detection method and device based on unsupervised feature combination, which aim to solve the problems that some product defect types are rarely generated in the actual production and manufacturing process, samples can hardly be collected, a lot of time and labor are needed for marking some product defects, and the detection precision is usually inferior to that of a supervised model when the unsupervised deep learning model is used for detecting the product image defects.
The technical scheme adopted by the invention is as follows: discloses a product image defect detection method based on unsupervised feature combination, which comprises the following steps
S1: collecting a small number of non-defective product images as training samples, and dividing the images into small blocks with equal size;
s2: dividing each image into two parts, wherein one part is an image small block, and the other part is a covered image obtained by covering the corresponding image small block;
s3: training a memory network and an image restoration network by using the image small blocks and the covering images, wherein the characteristics of the memory network and the image restoration network are combined and share a decoder;
s4: and carrying out defect detection on the image of the product to be detected by using the memory network and the image restoration network which are obtained through training.
As a preferred mode of the unsupervised feature combination-based product image defect detection method, the dividing of the image into small blocks with equal size means that each image is divided into different image areas with equal size by a grid, and a training sample is expanded.
As a preferable mode of the unsupervised feature combination-based product image defect detection method, the S2 includes:
s2.1: selecting an image of a training sample, and taking out a certain image small block on the image;
s2.2: covering the corresponding position of the image small block with a full black small block to obtain a covered image;
s2.3: the small image blocks and the corresponding covering images form a training sample pair;
s2.4: taking out another image small block from the image selected in the step S2.1, and carrying out the operations of the steps S2.2 and S2.3 to form another training sample pair;
s2.5: repeating S2.4 until each image small block in the image selected in S2.1 is selected, wherein the image selected in S2.1 forms a plurality of training sample pairs, and the number of the training sample pairs is equal to the number of the image small blocks of the image;
s2.6: steps S2.1-S2.5 are performed for each image of the training sample.
As a preferable mode of the unsupervised feature combination-based product image defect detection method, the S3 includes:
s3.1: constructing a memory network and an image restoration network, combining the characteristics output by the respective encoder parts of the memory network and the image restoration network, and simultaneously sharing a decoder;
s3.2: and (3) taking the training sample pairs formed in the steps S2.1-S2.6 as input, and training the memory network and the image restoration network to enable the network to reconstruct the image small blocks which are basically the same.
As a preferred mode of the unsupervised feature combination-based product image defect detection method, in S3.2, training the memory network and the image restoration network includes:
the characteristics of the memory network are replaced by the memories in the memory base, and the replacement formula is as follows:
Figure BDA0003340746300000031
q represents the features extracted by the encoder of the memory network, miWhich represents the memory of the user,
Figure BDA0003340746300000032
representing features after replacement with memory in a memory bank,
Figure BDA0003340746300000033
represents the weight, j represents the number of memory contained in the memory bank;
the characteristics after memory replacement are used in the memory network and the characteristics extracted from the image modifying network are connected and combined according to the channel;
the loss function used for network training is:
Figure BDA0003340746300000041
x represents a small block of the original image,
Figure BDA0003340746300000042
representing a reconstructed image patch.
As a preferable mode of the unsupervised feature combination-based product image defect detection method, the S4 includes:
s4.1: dividing an image to be detected into different image areas with equal size by using a grid;
s4.2: converting the image to be detected into a training sample pair consisting of small image blocks and corresponding covering images according to the steps S2.1-S2.6;
s4.3: inputting training sample pairs into the memory network and the image restoration network obtained through training to obtain small reconstructed image blocks;
s4.4: and splicing the small blocks of the reconstructed image into a reconstructed image, comparing the reconstructed image with the image to be detected, and judging whether the image to be detected has defects and the specific positions of the defects according to pixel-by-pixel difference.
As a preferred mode of the product image defect detection method based on unsupervised feature combination, S4.4 includes:
s4.41: calculating pixel-by-pixel absolute difference between the reconstructed image and the image to be detected, and selecting a threshold value for the difference image to carry out binarization;
s4.42: eliminating the influence of noise by using corrosion and expansion operation;
s4.43: judging and outputting whether the image to be detected has defects and the specific positions of the defects, and if the final difference value does not exceed the threshold value, outputting the image to be detected as the image of a normal product; and if the final difference value exceeds the threshold value, outputting that the image to be detected is the image of the defective product, and the corresponding area is the area where the defect exists.
The invention also discloses a product image defect detection device based on the unsupervised feature combination, which comprises a camera and a detection module, wherein the camera is fixed on a product production line and is used for shooting a product image; the detection module comprises a computer readable storage medium and a processor and is connected with the camera through a data line; the computer readable storage medium stores a computer program that executes, by the processor, the steps of the unsupervised feature combination-based product image defect detection method described above.
The invention has the beneficial effects that: the invention uses the memory network to extract the whole structure information of the image small block, simultaneously uses the image restoration network to extract the local field information of the adjacent image small block, and finally fuses the whole information and the local information, thereby reconstructing the details of the complex image, having high detection precision and extremely low false detection rate; according to the invention, the number of samples is increased by carrying out grid separation on the product images, so that the model can be trained by only a small number of normal sample images; the method only carries out unsupervised learning, does not need to collect missing item samples or label, has low cost and easy implementation, and judges the positions of the defects more accurately.
Drawings
FIG. 1 is a schematic flow chart of a product image defect detection method based on unsupervised feature combination disclosed in the present invention.
FIG. 2 is a schematic diagram of a training sample pair consisting of an image patch and a masking image as disclosed in the present invention.
FIG. 3 is a schematic structural diagram of a memory network and an image restoration network according to the present invention.
FIG. 4 is a schematic flow chart of the product image defect detection apparatus based on unsupervised feature combination according to the present disclosure.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention clearer, the present invention will be described in further detail below with reference to the accompanying drawings, but embodiments of the present invention are not limited thereto.
Example 1:
referring to fig. 1, the present embodiment discloses a product image defect detection method based on unsupervised feature combination, which specifically includes the following steps:
s1: a small number of non-defective product images are collected as training samples and the images are divided into small blocks of equal size.
S2: each image is divided into two parts, one part is an image small block, and the other part is a covering image obtained by covering the corresponding image small block.
S3: training a memory network and an image restoration network by using the image patches and the covering images, wherein the characteristics of the memory network and the image restoration network are combined and share a decoder.
S4: and carrying out defect detection on the image of the product to be detected by using the memory network and the image restoration network which are obtained through training.
In this example, 120 training samples were collected without defects.
The step S1 of dividing the image into small blocks of equal size means that each image is divided into different image areas of equal size by a grid, and the training samples are expanded.
In this embodiment, the image size is 256 × 256 and is divided into 256 image tiles of equal size by a 16 × 16 network.
The S2 specifically includes:
s2.1: an image of the training sample is selected, and a certain image patch on the image is taken.
S2.2: and covering the corresponding position of the image small block by using a full black small block to obtain a covered image.
S2.3: the image patches and corresponding masking images form a training sample pair.
S2.4: and taking another image small block from the image selected in the step S2.1, and performing the operations in the steps S2.2 and S2.3 to form another training sample pair.
S2.5: s2.4 is repeated until each image patch in the image selected in S2.1 is selected, at which point the image selected in S2.1 forms a number of training sample pairs equal to the number of image patches for that image.
S2.6: steps S2.1-S2.5 are performed for each image of the training sample.
In this embodiment, each training sample is converted into 256 pairs of training samples consisting of image patches and masking images. To more clearly illustrate the relationship between the image patches and the training sample pairs formed by the masked images, see fig. 2, fig. 2 divides the image into 4 image patches, and the image is subjected to steps S2.1-S2.5 to obtain 4 training sample pairs. Similarly, if an image in this embodiment is divided into 256 image patches, 256 training sample pairs will be obtained.
The S3 specifically includes:
s3.1: and constructing a memory network and an image restoration network, wherein the characteristics output by the respective encoder parts of the memory network and the image restoration network are combined, and simultaneously the memory network and the image restoration network share a decoder.
S3.2: and (3) taking the training sample pairs formed in the steps S2.1-S2.6 as input, and training the memory network and the image restoration network to reconstruct image small blocks which are as same as possible by the network.
In the present embodiment, the network is constructed as shown in fig. 3.
In this embodiment, the memory bank contains 512 memories, each of which has the same size as the characteristic size of the output of the encoder portion of the memory network.
In this embodiment, the memory network and the image restoration network respectively receive the image patch and the masking image in the training sample pair as input, and extract features through respective encoders, wherein the features of the memory network are replaced by the memory in the memory library, and the replacement formula is as follows:
Figure BDA0003340746300000071
q represents the features extracted by the encoder of the memory network, miWhich represents the memory of the user,
Figure BDA0003340746300000072
representing features after replacement with memory in a memory bank,
Figure BDA0003340746300000073
representing the weight.
In this embodiment, the features after the memory replacement are used in the memory network, and are connected and combined with the features extracted from the image modifying network according to the channel.
In this embodiment, the loss function used for network training is:
Figure BDA0003340746300000081
x represents a small block of the original image,
Figure BDA0003340746300000082
representing a reconstructed image patch.
The S4 specifically includes:
s4.1: the image to be examined is divided into different image areas of equal size by a grid.
S4.2: and converting the image to be detected into a training sample pair consisting of the image small blocks and the corresponding covering images according to the steps S2.1-S2.6.
S4.3: and inputting the training sample pair into the memory network and the image restoration network obtained through training to obtain a reconstructed image small block.
S4.4: and splicing the small blocks of the reconstructed image into a reconstructed image, comparing the reconstructed image with the image to be detected, and judging whether the image to be detected has defects and the specific positions of the defects according to pixel-by-pixel difference.
In this embodiment, the image to be examined is divided into 256 image patches of equal size using a 16 x 16 network. And splicing 256 reconstructed image small blocks of the image to be detected into a reconstructed image in sequence.
In the embodiment, the pixel-by-pixel absolute difference between the reconstructed image and the image to be detected is calculated, a threshold value is selected for the difference image to carry out binarization, and then the influence of noise is eliminated by using corrosion and expansion operation; if the final difference image has no white area, the image to be detected is the image of a normal product; and if the final difference image has a white area, the image to be detected is the image of the defective product, and the white area is the area where the defect exists.
Example 2
The embodiment provides a product image defect detection device based on unsupervised feature combination, which comprises a camera and a detection module, wherein the camera is fixed on a product production line and used for shooting a product image as shown in fig. 4; the detection module comprises a computer readable storage medium and a processor and is connected with the camera through a data line; the computer readable storage medium stores a computer program, and the computer program executes the steps of the unsupervised feature combination-based product image defect detection method according to embodiment 1 by using the processor.
The above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those skilled in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.

Claims (8)

1. A product image defect detection method based on unsupervised feature combination is characterized by comprising
S1: collecting a small number of non-defective product images as training samples, and dividing the images into small blocks with equal size;
s2: dividing each image into two parts, wherein one part is an image small block, and the other part is a covered image obtained by covering the corresponding image small block;
s3: training a memory network and an image restoration network by using the image small blocks and the covering images, wherein the characteristics of the memory network and the image restoration network are combined and share a decoder;
s4: and carrying out defect detection on the image of the product to be detected by using the memory network and the image restoration network which are obtained through training.
2. The unsupervised feature combination-based product image defect detection method of claim 1, wherein the dividing of the image into small blocks of equal size means that each image is divided into different image areas of equal size by a grid to expand the training samples.
3. The unsupervised combination of features based product image defect detection method of claim 1, wherein said S2 comprises:
s2.1: selecting an image of a training sample, and taking out a certain image small block on the image;
s2.2: covering the corresponding position of the image small block with a full black small block to obtain a covered image;
s2.3: the small image blocks and the corresponding covering images form a training sample pair;
s2.4: taking out another image small block from the image selected in the step S2.1, and carrying out the operations of the steps S2.2 and S2.3 to form another training sample pair;
s2.5: repeating S2.4 until each image small block in the image selected in S2.1 is selected, wherein the image selected in S2.1 forms a plurality of training sample pairs, and the number of the training sample pairs is equal to the number of the image small blocks of the image;
s2.6: steps S2.1-S2.5 are performed for each image of the training sample.
4. The unsupervised combination of features based product image defect detection method of claim 3, wherein said S3 comprises:
s3.1: constructing a memory network and an image restoration network, combining the characteristics output by the respective encoder parts of the memory network and the image restoration network, and simultaneously sharing a decoder;
s3.2: and (3) taking the training sample pairs formed in the steps S2.1-S2.6 as input, and training the memory network and the image restoration network to enable the network to reconstruct the image small blocks which are basically the same.
5. The unsupervised feature combination-based product image defect detection method of claim 4, wherein in S3.2, training the memory network and the image restoration network comprises:
the characteristics of the memory network are replaced by the memories in the memory base, and the replacement formula is as follows:
Figure FDA0003340746290000021
q represents the features extracted by the encoder of the memory network, miWhich represents the memory of the user,
Figure FDA0003340746290000025
representing features after replacement with memory in a memory bank,
Figure FDA0003340746290000022
represents the weight, j represents the number of memory contained in the memory bank;
the characteristics after memory replacement are used in the memory network and the characteristics extracted from the image modifying network are connected and combined according to the channel;
the loss function used for network training is:
Figure FDA0003340746290000023
x represents a small block of the original image,
Figure FDA0003340746290000024
representing a reconstructed image patch.
6. The unsupervised combination of features based product image defect detection method of claim 5, wherein said S4 comprises:
s4.1: dividing an image to be detected into different image areas with equal size by using a grid;
s4.2: converting the image to be detected into a training sample pair consisting of small image blocks and corresponding covering images according to the steps S2.1-S2.6;
s4.3: inputting training sample pairs into the memory network and the image restoration network obtained through training to obtain small reconstructed image blocks;
s4.4: and splicing the small blocks of the reconstructed image into a reconstructed image, comparing the reconstructed image with the image to be detected, and judging whether the image to be detected has defects and the specific positions of the defects according to pixel-by-pixel difference.
7. The unsupervised combination of features based product image defect detection method of claim 6, wherein said S4.4 comprises:
s4.41: calculating pixel-by-pixel absolute difference between the reconstructed image and the image to be detected, and selecting a threshold value for the difference image to carry out binarization;
s4.42: eliminating the influence of noise by using corrosion and expansion operation;
s4.43: judging and outputting whether the image to be detected has defects and the specific positions of the defects, and if the final difference value does not exceed the threshold value, outputting the image to be detected as the image of a normal product; and if the final difference value exceeds the threshold value, outputting that the image to be detected is the image of the defective product, and the corresponding area is the area where the defect exists.
8. A product image defect detection device based on unsupervised feature combination is characterized by comprising a camera and a detection module, wherein the camera is fixed on a product production line and is used for shooting a product image; the detection module comprises a computer readable storage medium and a processor and is connected with the camera through a data line; the computer readable storage medium stores a computer program, which executes the steps of the unsupervised combination of features based product image defect detection method of any one of claims 1 to 7 by the processor.
CN202111307737.4A 2021-11-05 2021-11-05 Product image defect detection method and device based on unsupervised feature combination Pending CN114037679A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111307737.4A CN114037679A (en) 2021-11-05 2021-11-05 Product image defect detection method and device based on unsupervised feature combination

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111307737.4A CN114037679A (en) 2021-11-05 2021-11-05 Product image defect detection method and device based on unsupervised feature combination

Publications (1)

Publication Number Publication Date
CN114037679A true CN114037679A (en) 2022-02-11

Family

ID=80136513

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111307737.4A Pending CN114037679A (en) 2021-11-05 2021-11-05 Product image defect detection method and device based on unsupervised feature combination

Country Status (1)

Country Link
CN (1) CN114037679A (en)

Similar Documents

Publication Publication Date Title
JP7138753B2 (en) Image defect detection method, device, electronic device, storage medium and product
US11715190B2 (en) Inspection system, image discrimination system, discrimination system, discriminator generation system, and learning data generation device
CN114724043B (en) Self-encoder anomaly detection method based on contrast learning
CN110648310A (en) Weak supervision casting defect identification method based on attention mechanism
CN112070135A (en) Power equipment image detection method and device, power equipment and storage medium
CN105931246A (en) Fabric flaw detection method based on wavelet transformation and genetic algorithm
CN113688665A (en) Remote sensing image target detection method and system based on semi-supervised iterative learning
CN112489092A (en) Fine-grained industrial motion mode classification method, storage medium, equipment and device
CN111667476A (en) Cloth flaw detection method and device, electronic equipment and readable storage medium
Kumar et al. Detection of concrete cracks using dual-channel deep convolutional network
CN113962951B (en) Training method and device for detecting segmentation model, and target detection method and device
CN116778137A (en) Character wheel type water meter reading identification method and device based on deep learning
CN115147418A (en) Compression training method and device for defect detection model
CN111461121A (en) Electric meter number identification method based on YO L OV3 network
CN114219762A (en) Defect detection method based on image restoration
CN112215301B (en) Image straight line detection method based on convolutional neural network
CN113506281A (en) Bridge crack detection method based on deep learning framework
CN113052103A (en) Electrical equipment defect detection method and device based on neural network
CN110363198B (en) Neural network weight matrix splitting and combining method
Fakhri et al. Road crack detection using gaussian/prewitt filter
CN111767324A (en) Intelligent associated self-adaptive data analysis method and device
CN114037679A (en) Product image defect detection method and device based on unsupervised feature combination
CN115908988A (en) Defect detection model generation method, device, equipment and storage medium
CN111882545B (en) Fabric defect detection method based on bidirectional information transmission and feature fusion
CN114758133A (en) Image flaw segmentation method based on super-pixel active learning and semi-supervised learning strategies

Legal Events

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