CN113947601A - Plastic product surface defect detection method and system based on semi-supervised learning - Google Patents

Plastic product surface defect detection method and system based on semi-supervised learning Download PDF

Info

Publication number
CN113947601A
CN113947601A CN202111561559.8A CN202111561559A CN113947601A CN 113947601 A CN113947601 A CN 113947601A CN 202111561559 A CN202111561559 A CN 202111561559A CN 113947601 A CN113947601 A CN 113947601A
Authority
CN
China
Prior art keywords
loss
defect detection
network branch
network
supervised
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.)
Granted
Application number
CN202111561559.8A
Other languages
Chinese (zh)
Other versions
CN113947601B (en
Inventor
吴国平
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.)
Wuhan Xinguomao Packaging Co ltd
Original Assignee
Wuhan Xinguomao Packaging 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 Wuhan Xinguomao Packaging Co ltd filed Critical Wuhan Xinguomao Packaging Co ltd
Priority to CN202111561559.8A priority Critical patent/CN113947601B/en
Publication of CN113947601A publication Critical patent/CN113947601A/en
Application granted granted Critical
Publication of CN113947601B publication Critical patent/CN113947601B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

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
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • 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
    • G06T2207/20021—Dividing image into blocks, subimages or windows
    • 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

Landscapes

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

Abstract

The invention relates to the field of defect detection, in particular to a plastic product surface defect detection method and system based on semi-supervised learning, wherein the method comprises the following steps: constructing a surface defect detection network; semi-supervised training is carried out on the surface defect detection network: inputting labeled data, and calculating supervision loss according to the defect detection results and the labeled data output by the first network branch and the second network branch by combining a cross entropy function; inputting label-free data, and calculating unsupervised loss according to the defect detection results output by the first network branch and the second network branch; setting a weight value for the unsupervised loss, and performing semi-supervised training based on the weighted sum of the supervised loss and the unsupervised loss; calculating a weight value during current training according to historical supervision loss, wherein the larger the historical supervision loss value is, the smaller the weight value is; and detecting the surface defects of the plastic products by using the trained surface defect detection network. The invention reduces the training cost and ensures the training effect.

Description

Plastic product surface defect detection method and system based on semi-supervised learning
Technical Field
The invention relates to the field of defect detection, in particular to a plastic product surface defect detection method and system based on semi-supervised learning.
Background
The surface quality problem often appears on the surface of the plastic product in the production and processing process of the plastic product, the quality of the surface quality is closely related to the performance of the plastic product, and meanwhile, the surface quality of the plastic product has certain influence on the sale of the plastic product. In the prior art, a deep learning technology is used for detecting surface defects, and the deep learning needs a large amount of label data, but for the processing and manufacturing industry, the defects of products are irregular, and the large amount of label data can increase a large amount of cost.
Disclosure of Invention
In order to solve the technical problems, the invention aims to provide a plastic product surface defect detection method and system based on semi-supervised learning, and the adopted technical scheme is as follows:
in a first aspect, an embodiment of the present invention provides a method for detecting surface defects of a plastic product based on semi-supervised learning, including the following specific steps:
constructing a surface defect detection network, wherein the surface defect detection network comprises a first network branch and a second network branch which are different in structure;
semi-supervised training is carried out on the surface defect detection network: inputting labeled data, and calculating supervision loss according to the defect detection results and the labeled data output by the first network branch and the second network branch by combining a cross entropy function; inputting label-free data, and calculating unsupervised loss according to the defect detection results output by the first network branch and the second network branch; setting a weight value for the unsupervised loss, and performing semi-supervised training based on the weighted sum of the supervised loss and the unsupervised loss; calculating a weight value during current training according to historical supervision loss, wherein the larger the historical supervision loss value is, the smaller the weight value is;
and detecting the surface defects of the plastic products by using the trained surface defect detection network.
Further, obtaining historical supervised average loss according to historical supervised loss after a plurality of times of training before the current time, and performing negative correlation mapping on the historical supervised average loss to obtain a weight value during the current training.
Further, the value range of the weight is [0, 1 ].
Further, the input of the first network branch is a surface image of the plastic product, and the output is a defect segmentation graph; dividing the surface image of the plastic product into image blocks, inputting the second network branch into the image blocks, outputting the image blocks into a significant thermodynamic diagram, and obtaining a defect segmentation diagram based on the spliced integral significant thermodynamic diagram; wherein the parameters of the coding modules in the first network branch and the second network branch are shared.
Further, there is a supervision loss L1The acquisition specifically comprises the following steps: l is1=L11+L12，L11Calculating cross entropy loss for the defect segmentation map and the label data output from the first network branch; the second network branch also outputs block marks of the image blocks, the block marks represent whether defects exist in the image blocks, block mark loss is calculated according to the difference value between the block marks of the image blocks output by the network branch and the block mark labels, and L is obtained by combining the block mark loss and cross entropy loss calculated according to a defect segmentation graph and label data output by the second network branch12。
Further, unsupervised loss L2The defect detection method is used for representing whether the defect detection results output by the first network branch and the second network branch are the same or not.
Further, unsupervised loss L2The acquisition specifically comprises the following steps: calculating pixel difference absolute values of pixel positions by pixel for the defect segmentation maps output by the first network branch and the second network branch, wherein the sum of the pixel difference absolute values is unsupervised loss L2。
In a second aspect, another embodiment of the present invention provides a semi-supervised learning based plastic product surface defect detecting system, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the semi-supervised learning based plastic product surface defect detecting method.
The embodiment of the invention at least has the following beneficial effects: in consideration of the fact that label data of surface defects of plastic products need to consume a large amount of manpower in the industrial processing process, and cost is increased, the method fully utilizes the existing label data, simultaneously combines easily obtained label-free data to perform semi-supervised training of a surface defect detection network, and utilizes the surface defect detection network to obtain accurate surface defect detection results. In addition, the invention calculates the unsupervised loss weight value based on the historical training results of a plurality of times without manually setting the weight value, can ensure the accurate adjustment of the signal intensity of the supervised loss and the unsupervised loss in the training process, enhances the auxiliary effect of the non-label data in the training process in due time and ensures the training effect.
Detailed Description
To further illustrate the technical means and effects of the present invention adopted to achieve the predetermined objects, the following detailed description is provided with reference to the preferred embodiments for a method and a system for detecting surface defects of plastic products based on semi-supervised learning, and the detailed implementation, structure, features and effects thereof according to the present invention. In the following description, different "one embodiment" or "another embodiment" refers to not necessarily the same embodiment. Furthermore, the particular features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.
Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
Aiming at the problems in the prior art, the invention provides a method for detecting the surface defects of plastic products, which comprises the steps of firstly constructing a surface defect network; secondly, collecting a data set of the surface defects of the plastic product, wherein the data set comprises tag data and non-tag data; finally, completing semi-supervised training of the surface defect detection network by using the data set; and obtaining the detection result of the surface defect of the plastic product by utilizing the trained surface defect detection network. The specific scheme of the plastic product surface defect detection method and system based on semi-supervised learning provided by the invention is specifically described below.
One embodiment of the invention provides a plastic product surface defect detection method based on semi-supervised learning, which comprises the following steps:
constructing a surface defect detection network, wherein the surface defect detection network comprises a first network branch and a second network branch which are different in structure;
semi-supervised training is carried out on the surface defect detection network: inputting labeled data, and calculating supervision loss according to the defect detection results and the labeled data output by the first network branch and the second network branch by combining a cross entropy function; inputting label-free data, and calculating unsupervised loss according to the defect detection results output by the first network branch and the second network branch; setting a weight value for the unsupervised loss, and performing semi-supervised training based on the weighted sum of the supervised loss and the unsupervised loss; calculating a weight value during current training according to historical supervision loss, wherein the larger the historical supervision loss value is, the smaller the weight value is;
and detecting the surface defects of the plastic products by using the trained surface defect detection network.
The following steps are detailed:
step S1, constructing a surface defect detection network, wherein the surface defect detection network includes a first network branch and a second network branch which are different in structure.
In order to perform semi-supervised training of a subsequent network, the surface defect detection network comprises a first network branch and a second network branch which are different in structure, wherein the first network branch and the second network branch can independently process input surface images of plastic products and output defect detection results.
Preferably, the input of the first network branch is a surface image of the plastic product, and the output is a defect segmentation graph; dividing the surface image of the plastic product into image blocks, inputting the second network branch into the image blocks, outputting the image blocks into a significant thermodynamic diagram, and obtaining a defect segmentation diagram based on the spliced integral significant thermodynamic diagram; wherein the parameters of the coding modules in the first network branch and the second network branch are shared. Specifically, the first network branch in the embodiment is a semantic segmentation network branch and comprises a first encoder and a first decoder, wherein the first encoder performs feature extraction on a surface image of a plastic product to obtain a feature map, and the first decoder performs up-sampling on the feature map to obtain a first defect segmentation map; the second branch comprises a second encoder, the second encoder performs feature extraction on each image block, then completes classification tasks by using Global Average Pooling (GAP), obtains a significant thermodynamic diagram (CAM) of each image block according to classification results, splices the significant thermodynamic diagrams (CAM) of all the image blocks to obtain an overall significant thermodynamic diagram, and performs binarization operation on the overall significant thermodynamic diagram to obtain a second defect segmentation diagram. It should be noted that the parameters of the encoding modules in the first network branch and the second network branch, i.e. the first encoder and the second encoder, are shared, and joint training of the two network branches ensures that the encoder can extract features related to the defect. Preferably, in the embodiment, the surface image of the plastic product is divided into 9 image blocks, and the surface image of the plastic product is divided into a plurality of image blocks, so that the data volume can be increased, the characteristics of the defect area can be accurately extracted by the second encoder, and the network performance is guaranteed; in the defect segmentation map, the defective pixel value is 1, and the non-defective pixel value is 0.
Step S2, performing semi-supervised training on the surface defect detection network: inputting labeled data, and calculating supervision loss according to the defect detection results and the labeled data output by the first network branch and the second network branch by combining a cross entropy function; inputting label-free data, and calculating unsupervised loss according to the defect detection results output by the first network branch and the second network branch; setting a weight value for the unsupervised loss, and performing semi-supervised training based on the weighted sum of the supervised loss and the unsupervised loss; and calculating the weight value during the current training according to the historical supervised loss, wherein the larger the historical supervised loss value is, the smaller the weight value is.
(a) Acquiring a training data set, wherein the training data is a plastic product surface image obtained by collecting an image of the surface of a plastic product by using an RGB (red, green and blue) camera; in the invention, semi-supervised training is carried out on the surface defect detection network, therefore, the training data set comprises labeled data and unlabelled data, the labeled data corresponding to the labeled data is a defect area in an artificially labeled plastic product surface image, the pixel value of the defect area is marked as 1, namely the label value of the defect pixel is 1, the pixel value of the normal area is marked as 0, namely the label value of the non-defect pixel is 0, and the cost for obtaining the label data is overhigh, so that only the label data of part of the training data needs to be obtained.
(b) Because the training data set contains tag data and non-tag data, in order to enable the non-tag data to play a role in supervision in the training process, different loss functions need to be set for different data:
(i) inputting labeled data, and calculating supervision loss according to the defect detection results and the labeled data output by the first network branch and the second network branch by combining a cross entropy function; preferably, there is a supervision loss L1The acquisition of (A) is as follows: l is1=L11+L12，L11Calculating cross entropy loss for the defect segmentation map and the label data output from the first network branch; the second network branch also outputs block marks of the image blocks, the block marks represent whether defects exist in the image blocks, block mark loss is calculated according to the difference value between the block marks of the image blocks output by the network branch and the block mark labels, and L is obtained by combining the block mark loss and cross entropy loss calculated according to a defect segmentation graph and label data output by the second network branch12(ii) a Specifically, the method comprises the following steps:
Figure DEST_PATH_IMAGE002
wherein the content of the first and second substances,
Figure DEST_PATH_IMAGE004
inputting size information of the surface image of the plastic product;
Figure DEST_PATH_IMAGE006
for input surface images of plastic articles
Figure DEST_PATH_IMAGE008
Middle pixel point
Figure DEST_PATH_IMAGE010
A corresponding tag value;
Figure DEST_PATH_IMAGE012
segmenting pixels in the map for the first defect
Figure 167584DEST_PATH_IMAGE010
Corresponding output result, i.e. pixel point
Figure 615883DEST_PATH_IMAGE010
A pixel value of (a);
Figure DEST_PATH_IMAGE014
in order to balance the influence of the difference of positive and negative sample data quantities on the network training result for a set hyper-parameter, the area of a defect region in the input plastic product surface image is a small region which is far smaller than the area of a normal region according to the prior, so the value of the embodiment is taken
Figure DEST_PATH_IMAGE016
And reducing the influence degree of the normal area sample on the training result.
Figure DEST_PATH_IMAGE018
Wherein the content of the first and second substances,
Figure DEST_PATH_IMAGE020
segmenting pixels in the map for the second defect
Figure 270986DEST_PATH_IMAGE010
Corresponding output result, i.e. pixel point
Figure 197354DEST_PATH_IMAGE010
A pixel value of (a);
Figure DEST_PATH_IMAGE022
for branching output of second network
Figure DEST_PATH_IMAGE024
The block flags of the image blocks are changed,
Figure DEST_PATH_IMAGE026
is as follows
Figure 621513DEST_PATH_IMAGE024
A block mark label corresponding to each image block; in embodiments the block markers are in the form of binary groupsOne dimension in the binary group represents that the image block has defects, the other dimension represents that the image block has no defects, preferably, the embodiment that the block mark is (1,0) represents that the image block has defects, the block mark is (0,1) represents that the image block has no defects,
Figure DEST_PATH_IMAGE028
representing the Euclidean distance; meaning and loss function of the remaining letters
Figure DEST_PATH_IMAGE030
In the same meaning, L12The second encoder may be further constrained from learning valid features associated with the defect. It should be noted that the block mark of the image block can be obtained by the second branch network according to whether there is a defective pixel in the second defect segmentation map.
(ii) Inputting non-label data, and calculating the unsupervised loss according to the defect detection results output by the first network branch and the second network branch, specifically:
unsupervised loss L2The defect detection method is used for representing whether the defect detection results output by the first network branch and the second network branch are the same or not. Due to the above-mentioned losses
Figure DEST_PATH_IMAGE032
The method is not suitable for label-free data, and the segmentation results output by two network branches should be consistent according to prior, so that no supervision loss L exists2The acquisition specifically comprises the following steps: calculating pixel difference absolute values of pixel positions by pixel for the defect segmentation maps output by the first network branch and the second network branch, wherein the sum of the pixel difference absolute values is unsupervised loss L2：
Figure DEST_PATH_IMAGE034
Wherein the content of the first and second substances,
Figure DEST_PATH_IMAGE036
and
Figure DEST_PATH_IMAGE038
respectively representing pixel points in the first defect segmentation graph and the second defect segmentation graph
Figure 584921DEST_PATH_IMAGE010
The pixel value of (2).
(iii) Setting a weight value for the unsupervised loss, and performing semi-supervised training based on the weighted sum of the supervised loss and the unsupervised loss; and calculating the weight value during the current training according to the historical supervised loss, wherein the larger the historical supervised loss value is, the smaller the weight value is.
Preferably, the embodiment obtains the historical supervised average loss according to the historical supervised loss after a plurality of times of training before the current time, performs negative correlation mapping on the historical supervised average loss to obtain the weight during the current training, specifically:
Figure DEST_PATH_IMAGE040
Figure DEST_PATH_IMAGE042
is the corresponding unsupervised weight loss in the training of the Tth time,
Figure DEST_PATH_IMAGE044
and
Figure DEST_PATH_IMAGE046
respectively representing L calculated after the t-th training11And L12N is a predetermined value, preferably, in the embodiment, the value of N is 15,
Figure DEST_PATH_IMAGE048
denotes the supervised loss L calculated after the t-th training1The average loss of (a) is,
Figure DEST_PATH_IMAGE050
representing the supervised average loss of the history obtained after N times of training before the current training; when T is less than or equal to N, the training is obtained after T-1 times of historical trainingHistorical supervised loss calculation historical supervised average loss.
Further, in the semi-supervised training process, the value of the loss function is continuously reduced along with the increase of the training time, and the performance of the surface defect detection network is better and better; in the early stage of training, in order to ensure that the output of the surface defect detection network is matched with a target task, the network needs to pay more attention to the supervision of labeled data; therefore, the signal strength of the labeled data and the signal strength of the unlabeled data are balanced by utilizing the real-time loss function of the labeled data in the training process, so that the surface defect detection network can meet the task requirement and can learn useful characteristics from the unlabeled data; therefore, the present application finally performs semi-supervised training based on a weighted sum of supervised and unsupervised losses, in particular:
Figure DEST_PATH_IMAGE052
wherein m is1And m2Respectively, the data amount of the labeled data and the unlabeled data in the training data set, each data in the training data set is correspondingly trained once, and further the data amount can also represent the training times,
Figure 71528DEST_PATH_IMAGE042
the value is a parameter which is changed along with the iteration of the training process and is used for balancing the signal intensity of the labeled data and the unlabeled data, and the value of the value is related to the magnitude of the historical supervised average loss. Weight value
Figure 67166DEST_PATH_IMAGE042
Has a value range of [0, 1]]The larger the historical supervised loss value is, the larger the historical supervised average loss is, the more the surface defect detection network is not fully trained, at this time, the signal intensity of the non-label data in the training process needs to be reduced, that is, the more the non-label data can not be used for network data at this time, and the smaller the weight value needs to be; when the historical supervised average loss is smaller, the surface defect detection network can learn the characteristics related to the surface defects, and the number of the non-labels is increasedAnd according to the signal intensity, the auxiliary effect of the non-tag data is enhanced.
Figure DEST_PATH_IMAGE054
For the supervised losses obtained after the tth training with labeled data,
Figure DEST_PATH_IMAGE056
the unsupervised loss obtained after the R-th training with the unlabeled data.
(c) And continuously updating network parameters by using a gradient descent method according to the training data set and the weighted sum of the supervised loss and the unsupervised loss to finish the semi-supervised training of the surface defect detection network.
And step S3, detecting the surface defects of the plastic products by using the trained surface defect detection network.
In the actual use process, the surface image of the plastic product acquired in real time is sent to a surface defect detection network, and the detection result of the surface defect of the plastic product, namely a defect segmentation graph, can be output by utilizing any one network branch.
Based on the same inventive concept as the method embodiment, an embodiment of the present invention provides a surface defect detecting system for plastic products based on semi-supervised learning, which comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein when the computer program is executed by the processor, the steps of the surface defect detecting method for plastic products based on semi-supervised learning are realized.
It should be noted that: the precedence order of the above embodiments of the present invention is only for description, and does not represent the merits of the embodiments. And specific embodiments thereof have been described above. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims may be performed in a different order than in the embodiments and still achieve desirable results. In some embodiments, multitasking and parallel processing may also be possible or may be advantageous.
The embodiments in the present specification are described in a progressive manner, and the same and similar parts among the embodiments are referred to each other, and each embodiment focuses on the differences from the other embodiments.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.

Claims (8)

1. A plastic product surface defect detection method based on semi-supervised learning is characterized by comprising the following steps:
constructing a surface defect detection network, wherein the surface defect detection network comprises a first network branch and a second network branch which are different in structure;
semi-supervised training is carried out on the surface defect detection network: inputting labeled data, and calculating supervision loss according to the defect detection results and the labeled data output by the first network branch and the second network branch by combining a cross entropy function; inputting label-free data, and calculating unsupervised loss according to the defect detection results output by the first network branch and the second network branch; setting a weight value for the unsupervised loss, and performing semi-supervised training based on the weighted sum of the supervised loss and the unsupervised loss; calculating a weight value during current training according to historical supervision loss, wherein the larger the historical supervision loss value is, the smaller the weight value is;
and detecting the surface defects of the plastic products by using the trained surface defect detection network.
2. The method of claim 1, wherein the historical supervised average loss is obtained according to the historical supervised losses after a plurality of times of training before the current time, and the negative correlation mapping is performed on the historical supervised average loss to obtain the weight during the current training.
3. The method of claim 2, wherein the weight value ranges from [0, 1 ].
4. The method of claim 3, wherein the input of the first network branch is a surface image of the plastic article and the output is a defect segmentation map; dividing the surface image of the plastic product into image blocks, inputting the second network branch into the image blocks, outputting the image blocks into a significant thermodynamic diagram, and obtaining a defect segmentation diagram based on the spliced integral significant thermodynamic diagram; wherein the parameters of the coding modules in the first network branch and the second network branch are shared.
5. The method of claim 4, wherein there is a tag loss L1The acquisition specifically comprises the following steps: l is1=L11+L12，L11Calculating cross entropy loss for the defect segmentation map and the label data output from the first network branch; the second network branch also outputs block marks of the image blocks, the block marks represent whether defects exist in the image blocks, block mark loss is calculated according to the difference value between the block marks of the image blocks output by the network branch and the block mark labels, and L is obtained by combining the block mark loss and cross entropy loss calculated according to a defect segmentation graph and label data output by the second network branch12。
6. The method of claim 5, in which there is no supervised loss L2The defect detection method is used for representing whether the defect detection results output by the first network branch and the second network branch are the same or not.
7. The method of claim 6, in which there is no supervised loss L2The acquisition specifically comprises the following steps: calculating pixel difference absolute values of pixel positions by pixel for the defect segmentation maps output by the first network branch and the second network branch, wherein the sum of the pixel difference absolute values is unsupervised loss L2。
8. A semi-supervised learning based surface defect detection system for plastic articles, comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the computer program, when executed by the processor, carries out the steps of the method according to any one of claims 1 to 7.
CN202111561559.8A 2021-12-20 2021-12-20 Plastic product surface defect detection method and system based on semi-supervised learning Active CN113947601B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111561559.8A CN113947601B (en) 2021-12-20 2021-12-20 Plastic product surface defect detection method and system based on semi-supervised learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111561559.8A CN113947601B (en) 2021-12-20 2021-12-20 Plastic product surface defect detection method and system based on semi-supervised learning

Publications (2)

Publication Number Publication Date
CN113947601A true CN113947601A (en) 2022-01-18
CN113947601B CN113947601B (en) 2022-03-22

Family

ID=79339413

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111561559.8A Active CN113947601B (en) 2021-12-20 2021-12-20 Plastic product surface defect detection method and system based on semi-supervised learning

Country Status (1)

Country Link
CN (1) CN113947601B (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114998309A (en) * 2022-07-12 2022-09-02 启东万惠机械制造有限公司 Unsupervised mechanical part surface defect detection method and unsupervised mechanical part surface defect detection system
CN116862903A (en) * 2023-07-31 2023-10-10 梅卡曼德（北京）机器人科技有限公司 Defect detection model training method and device, defect detection method and electronic equipment
CN116883390A (en) * 2023-09-04 2023-10-13 合肥中科类脑智能技术有限公司 Fuzzy-resistant semi-supervised defect detection method, device and storage medium

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110880176A (en) * 2019-11-19 2020-03-13 浙江大学 Semi-supervised industrial image defect segmentation method based on countermeasure generation network
CN112164040A (en) * 2020-09-17 2021-01-01 杭州电子科技大学 Steel surface defect identification method based on semi-supervised deep learning algorithm
CN113436169A (en) * 2021-06-25 2021-09-24 东北大学 Industrial equipment surface crack detection method and system based on semi-supervised semantic segmentation

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110880176A (en) * 2019-11-19 2020-03-13 浙江大学 Semi-supervised industrial image defect segmentation method based on countermeasure generation network
CN112164040A (en) * 2020-09-17 2021-01-01 杭州电子科技大学 Steel surface defect identification method based on semi-supervised deep learning algorithm
CN113436169A (en) * 2021-06-25 2021-09-24 东北大学 Industrial equipment surface crack detection method and system based on semi-supervised semantic segmentation

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114998309A (en) * 2022-07-12 2022-09-02 启东万惠机械制造有限公司 Unsupervised mechanical part surface defect detection method and unsupervised mechanical part surface defect detection system
CN116862903A (en) * 2023-07-31 2023-10-10 梅卡曼德（北京）机器人科技有限公司 Defect detection model training method and device, defect detection method and electronic equipment
CN116883390A (en) * 2023-09-04 2023-10-13 合肥中科类脑智能技术有限公司 Fuzzy-resistant semi-supervised defect detection method, device and storage medium
CN116883390B (en) * 2023-09-04 2023-11-21 合肥中科类脑智能技术有限公司 Fuzzy-resistant semi-supervised defect detection method, device and storage medium

Also Published As

Publication number Publication date
CN113947601B (en) 2022-03-22

Similar Documents

Publication Publication Date Title
CN113947601B (en) Plastic product surface defect detection method and system based on semi-supervised learning
Liu et al. A method of segmenting apples at night based on color and position information
CN113643268B (en) Industrial product defect quality inspection method and device based on deep learning and storage medium
CN107169985A (en) A kind of moving target detecting method based on symmetrical inter-frame difference and context update
WO2022236876A1 (en) Cellophane defect recognition method, system and apparatus, and storage medium
CN112862811A (en) Material microscopic image defect identification method, equipment and device based on deep learning
CN115830471B (en) Multi-scale feature fusion and alignment domain self-adaptive cloud detection method
CN114140665A (en) Dense small target detection method based on improved YOLOv5
CN110827312A (en) Learning method based on cooperative visual attention neural network
CN115147418B (en) Compression training method and device for defect detection model
CN111027538A (en) Container detection method based on instance segmentation model
CN111798470A (en) Crop image entity segmentation method and system applied to intelligent agriculture
CN113313684A (en) Video-based industrial defect detection system under dim light condition
CN113409355A (en) Moving target identification system and method based on FPGA
CN116363505A (en) Target picking method based on picking robot vision system
Chen et al. Segmentation of field grape bunches via an improved pyramid scene parsing network
CN112967271B (en) Casting surface defect identification method based on improved DeepLabv3+ network model
CN107239761B (en) Fruit tree branch pulling effect evaluation method based on skeleton angular point detection
CN113077438A (en) Cell nucleus region extraction method and imaging method for multi-cell nucleus color image
CN117437691A (en) Real-time multi-person abnormal behavior identification method and system based on lightweight network
CN117058669A (en) Deep learning-based litchi fruit identification method
CN104751487A (en) Method for detecting movement target based on colored RGB three-pane color-change frame difference
CN115661451A (en) Deep learning single-frame infrared small target high-resolution segmentation method
CN115100533A (en) Training and using method of litchi maturity recognition model
CN114429461A (en) Cross-scene strip steel surface defect detection method based on domain adaptation

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
GR01 Patent grant
GR01 Patent grant