CN111932639B - Detection method of unbalanced defect sample based on convolutional neural network - Google Patents
Detection method of unbalanced defect sample based on convolutional neural network Download PDFInfo
- Publication number
- CN111932639B CN111932639B CN202010939183.9A CN202010939183A CN111932639B CN 111932639 B CN111932639 B CN 111932639B CN 202010939183 A CN202010939183 A CN 202010939183A CN 111932639 B CN111932639 B CN 111932639B
- Authority
- CN
- China
- Prior art keywords
- image
- data
- defect
- images
- channel
- 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.)
- Active
Links
- 230000007547 defect Effects 0.000 title claims abstract description 95
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 26
- 238000001514 detection method Methods 0.000 title claims abstract description 26
- 238000013528 artificial neural network Methods 0.000 claims abstract description 26
- 230000002950 deficient Effects 0.000 claims abstract description 21
- 238000012549 training Methods 0.000 claims abstract description 16
- 238000004519 manufacturing process Methods 0.000 claims abstract description 13
- 238000000034 method Methods 0.000 claims description 13
- 238000005070 sampling Methods 0.000 claims description 11
- 239000013604 expression vector Substances 0.000 claims description 9
- 238000012937 correction Methods 0.000 claims description 7
- 101100295091 Arabidopsis thaliana NUDT14 gene Proteins 0.000 claims description 3
- 238000012935 Averaging Methods 0.000 claims description 3
- 238000012545 processing Methods 0.000 abstract description 5
- 238000013135 deep learning Methods 0.000 description 5
- 239000000463 material Substances 0.000 description 4
- 230000007246 mechanism Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 230000016776 visual perception Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 1
- 238000005520 cutting process Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000004438 eyesight Effects 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 239000000696 magnetic material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 239000013598 vector Substances 0.000 description 1
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
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06T5/00—Image enhancement or restoration
- G06T5/50—Image enhancement or restoration using two or more images, e.g. averaging or subtraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/90—Determination of colour characteristics
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a detection method of an unbalanced defect sample based on a convolutional neural network, which comprises the following steps: (1) collecting early data: collecting two types of data, wherein one type of data is original defect data of a product, and making an original image of a training set; the second is that data without defects, called non-defective data, produced in the production environment forms a non-defective image; (2) making a label image of original defect data; (3) deep neural network training: (4) defect reasoning: and for a new image in the actual production environment, randomly extracting a plurality of non-defective images, cascading each non-defective image and the new image to form 6 channels of input, sending the input into a network to obtain output, and voting to obtain the final defective pixel. The invention collects the defect data and the non-defect data, namely adds a large number of non-defect samples to form new data input, thereby improving the detection performance and being capable of processing some new types of defects.
Description
Technical Field
The invention relates to the technical field of defect detection and identification, in particular to a detection method of unbalanced defect samples based on a convolutional neural network.
Background
Convolutional Neural Networks (CNN) are a class of feed forward Neural Networks (fed forward Neural Networks) that contain convolution computations and have a deep structure, and are one of the representative algorithms for deep learning (deep learning). Convolutional Neural Networks have a feature learning (representation learning) capability, and can perform Shift-Invariant classification (Shift-Invariant classification) on input information according to a hierarchical structure thereof, and are also called Shift-Invariant Artificial Neural Networks (SIANN). Studies on convolutional neural networks began in the 80 to 90 th century, with time delay networks and LeNet-5 being the earliest convolutional neural networks. After the twenty-first century, with the introduction of deep learning theory and the improvement of numerical computing equipment, convolutional neural networks have been rapidly developed and applied to the fields of computer vision, natural language processing, and the like. The convolutional neural network is constructed by imitating a visual perception (visual perception) mechanism of a living being, can perform supervised learning and unsupervised learning, and has the advantages that the convolutional neural network can learn grid-like topologic features such as pixels and audio with small calculation amount, has stable effect and has no additional feature engineering (feature engineering) requirement on data due to the fact that convolutional kernel parameter sharing in an implicit layer and sparsity of connection between layers. Convolutional neural networks have long been one of the core algorithms in the field of image recognition and have stable performance when the learning data is sufficient. For a general large-scale image classification problem, the convolutional neural network can be used for constructing a hierarchical classifier (hierarchical classifier) and can also be used for extracting discriminant features of an image in fine-classification recognition (fine-grained-classification recognition) for other classifiers to learn. For the latter, feature extraction can be performed by artificially inputting different parts of an image into a convolutional neural network respectively, or by extracting the different parts of the image by the convolutional neural network through unsupervised learning.
At present, in the field of defect detection, a convolutional neural network is widely applied, and deep learning is playing an increasingly important role. For example, chinese patent application CN109993094A discloses a material defect intelligent detection method based on machine vision, which comprises the following steps: 1) the data acquisition module acquires image data of the microscope equipment in real time; 2) a user configures a training basic characteristic picture library of the representative material defect to be detected through a user interface module or manually; 3) extracting image features, and performing image optimization to obtain feature data; 4) classifying and identifying the image characteristics, and performing classification training on the characteristics of various defect types of the material; 5) performing multi-layer network iterative computation on the extracted features and classification by using a mature neural network convolution algorithm to obtain feature results and context data of all types of defects of the material; 6) and outputting a deep learning result, and using the result and the characteristic context data thereof in a training learning library to gradually improve the identification efficiency and accuracy of the magnetic material crack problem. The method adopts the defect characteristics based on machine vision, does not depend on a professional detection mechanism or manual detection completely, and continuously iterates through the whole detection process by a defect characteristic library, so that the detection efficiency is continuously improved; in addition, the accumulated defect data will form a defect feature database, which is a prototype of the large data in the industry in the long term.
However, the above method has the following disadvantages: (1) in some practical production environments, the types of defects are very complex and variable, some defects are easy to collect, the data volume is large, other defects are few, the proportion of the tiny defects in the whole data set is small, and the defects cannot be well trained by the model due to the small proportion, so that the defects cannot be detected; (2) in actual production, the proportion of good products is generally high, the proportion of defective products is low, and therefore, the defect data which can be collected is also small, and for the practical problem, the prior art takes a lot of time to collect the defect data (such as the method of CN 109993094A), but the method is too time-consuming, and takes too long time and too large cost for enterprises.
Therefore, a new detection method for the unbalanced defect sample based on the convolutional neural network, which can be rapidly put into use, is developed, and various defects (including common defects and uncommon defects) can be comprehensively detected, so that a large amount of labor cost is avoided, and the efficiency is improved.
Disclosure of Invention
The invention aims to provide a detection method of unbalanced defect samples based on a convolutional neural network.
In order to achieve the purpose of the invention, the technical scheme adopted by the invention is as follows: a detection method of unbalanced defect samples based on a convolutional neural network comprises the following steps:
(1) collecting early data: collecting two types of data, wherein one type of data is original defect data of a product, and making an original image of a training set; the second is that data without defects, called non-defective data, produced in the production environment forms a non-defective image;
(2) making a label image of the original defect data: marking the characteristics in the task target by using an image marking tool, and making a corresponding label image; for the non-defective data, this step is not processed;
(3) deep neural network training:
(a) and (3) coordinate correction: firstly, position correction is carried out on the label images, the absolute positions of central pixel points of all the label images are ensured to be consistent when the central pixel points are collected, and all the label images are cut to be the same size;
(b) sampling forms the neural network input: adopting 6-channel input, randomly taking two label images, fixing one label image as a non-defective image, randomly extracting the other label image with a defect or the non-defective image, and then cascading the two images to form 6-channel input;
finally obtaining a trained model;
(4) defect reasoning: and (4) reasoning by using the trained model in the step (3) and the collected defect-free image, randomly extracting a plurality of defect-free images for a new image in the actual production environment, cascading each defect-free image and the new image to form 6 channels for input, sending the input into a network to obtain output, and voting to obtain the final defect pixel.
In the above, in the step (1), it is necessary to pay special attention to the fact that the environmental variables of the two data are completely consistent when the two data are collected. In addition, technicians determine the types of defects which may occur according to the production conditions of the products, and the acquisition personnel use the photographic equipment to acquire early-stage data which are as diverse as possible according to the types of the defects and make original images of the training set.
Preferably, in the step (1), the proportion of the defect data to the total data is 20% or more, and the proportion of the non-defect data to the total data is 50% or more. Preferably, the proportion of the defect data to the total data is greater than or equal to 30%, and the proportion of the non-defect data to the total data is greater than or equal to 60%.
Preferably, in the step (1), a ratio of the defect data to the non-defect data is 1:3 to 5. More preferably, the ratio of the defect data to the non-defect data is 1: 4.
Preferably, in the step (1), a photographic device is used to collect original defect data of the product according to the defect type, and an original image of the training set is made.
Preferably, in the step (3), the neural network adopts renet as a backbone network, and the cavity convolution and the ASPP structure are utilized to expand the receptive field so as to adapt to various defects with different sizes, lengths and shapes.
Preferably, the coordinates in step (3) are corrected by:
(1) setting the pixel value of the background to 255;
(2) averaging the length and the width of all the images to obtain a standard length and width value, and adjusting the length and the width of all the images to the standard value;
(3) the foreground pixel of each image has four boundary pixels of top, bottom, left and right, the coordinate values of the four boundary pixels of all the images are taken and respectively averaged, and a rectangular frame is determined according to the four average value coordinates, wherein the rectangular frame is a target rectangular frame of the foreground;
and adjusting foreground pixels of all images to ensure that the central point is close to the central point of the rectangular frame, adjusting the foreground size to ensure that the foreground is completely in the rectangular frame, and ensuring that four pixels at the top, the bottom, the left and the right are on corresponding sides of the rectangular frame.
Preferably, the sampling in step (3) (b) forms a neural network input as: a complete random sampling is taken to extract a defect-free image from the database.
In another corresponding technical solution, the sampling in (b) in step (3) forms a neural network input as follows: the method comprises the steps of utilizing a classification neural network which is trained in advance to obtain expression vectors of all images in a database, when an image to be detected is input, firstly sending the image to be detected into the classification neural network to obtain the expression vectors of the image to be detected, then calculating the expression vectors of all images or partial images in the database, and selecting the images according to the similarity.
Of the two schemes described above, the first scheme is suitable for fast processing and the second scheme is suitable for accurate processing.
Preferably, the cascade of step (b) in step (3) forms a 6-channel input, specifically: and forming 6-channel input by adopting a mode that a first image RGB channel and a second image RGB channel are cascaded.
Preferably, the cascade of step (b) in step (3) forms a 6-channel input, specifically: the 6-channel input is formed in a mode that a first image R channel, a second image R channel, a first image G channel, a second image G channel, a first image B channel and a second image B channel are cascaded.
Due to the application of the technical scheme, compared with the prior art, the invention has the following advantages:
the invention improves the detection performance by collecting defect data and non-defect data, namely adding a large amount of non-defect samples to form new data input, and can process some new types of defects which cannot be detected in the existing detection method (such as the method of CN 109993094A);
the method adopts the defect data and the non-defect data at the same time, and carries out intensive comparison, so that a plurality of defect data samples are not needed, the problems of time and cost caused by the need of spending a large amount of time on collecting the defect data in the prior art are solved, the application can be rapidly realized, the model reasoning capability is strong, and the requirements of different product detection can be met only by small amount of adjustment;
3, the method adopts 6-channel input sampling to form neural network input, does not need to classify defects, only needs to make a model pay attention to what the defects are and what the defects are not, and has good self-adaptive capacity for the condition of unbalanced defect samples and even the condition that the defect samples never appear;
4, every time a defect sample is added, a large number of training samples can be added to the model, so that the data labeling work with large input of labor cost is avoided;
5, the method is simple and easy to implement, has low cost and is suitable for popularization and application.
Drawings
Fig. 1 is a schematic diagram of a first embodiment of the present invention.
Detailed Description
The invention is further described with reference to the following figures and examples:
the first embodiment is as follows:
referring to fig. 1, a method for detecting an unbalanced defect sample based on a convolutional neural network includes the following steps:
(1) collecting early data: two types of data need to be collected: one is original defect data of the product, technicians determine types of defects which may occur according to production conditions of the product, and acquisition personnel acquire early-stage data which is as diverse as possible according to the types of the defects by using photographic equipment and make original images of a training set;
the second is data without defects produced in the production environment, and the data can be collected in large quantity; special attention needs to be paid to the fact that when two kinds of data are collected, environment variables are completely consistent as much as possible;
specifically, the proportion of the defect data to the total data is 30%, and the proportion of the defect-free data to the total data is 70%;
(2) making a label image of the original defect data: using an image marking tool to manufacture a corresponding label image, and marking the characteristics in the task target; for the non-defective data, this step is not processed;
(3) deep neural network training:
(a) and (3) coordinate correction: firstly, correcting the position of a collected image, ensuring that the absolute positions of central pixel points of all images during collection are consistent, and simultaneously cutting all images to the same size;
(b) sampling to form a neural network input, wherein the neural network adopts 6-channel input, which is different from 3-channel or single-channel input of a traditional network, specifically, two images are randomly selected, one image is fixed as a non-defective image, the other image is randomly selected as a defective image or a non-defective image, and then the two images are cascaded to form the 6-channel input;
the formed 6-channel input can solve the problem of data imbalance of different defects because specific defect types are not distinguished; the deep neural network adopts the resnet as a backbone network, and enlarges the receptive field by utilizing the cavity convolution and the ASPP structure, thereby being suitable for various defects with different sizes, lengths and shapes; a large number of 6-channel inputs can be increased when a defective image is added, so that the labor cost is greatly reduced, and training samples are greatly increased, thereby helping to improve the performance of the model;
(4) defect reasoning: and (4) reasoning by using the trained model in the step (3) and the acquired defect-free image, randomly extracting a plurality of defect-free images for a new image in the actual production environment, cascading each defect-free image and the new image to form 6 channels for input, sending the input into a network to obtain output, and voting to obtain the final defect pixel.
The coordinate correction in the step (3) is as follows: first, a large amount of image data is acquired, ensuring that the background is pure white (pixel value = 255); averaging the length and the width of all the images to obtain a standard length and width value, and adjusting the length and the width of all the images to the standard value; then, the foreground pixel of each image has four boundary pixels of top, bottom, left and right, the coordinate values of the four boundary pixels of all the images are taken and respectively averaged, and a rectangular frame is determined according to the four average value coordinates, wherein the rectangular frame is a target rectangular frame of the foreground; and adjusting foreground pixels of all images to ensure that the central point is close to the central point of the rectangular frame, adjusting the foreground size to ensure that the foreground pixels can completely fall in the rectangular frame, and ensuring that four pixels at the top, the bottom, the left and the right fall on corresponding edges of the rectangular frame as far as possible.
Sampler technical details: the sampler has two options. First, a full random sampling is taken to extract a defect-free picture from the database. Secondly, obtaining expression vectors (characteristic vectors after passing through a global pooling layer) of all images in the database by utilizing a pre-trained classified neural network, when an image to be detected is input, firstly sending the image to be detected into the classified neural network to obtain the expression vectors of the image to be detected, then calculating the expression vectors of all images or partial images in the database, and selecting the images according to the similarity. The first solution is suitable for fast processing and the second solution is suitable for accurate processing.
Technical details of the multi-channel fusion device: two solutions, firstly, a mode of cascade connection of a first image RGB channel and a second image RGB channel is adopted; secondly, a cascading mode of a first image R channel, a second image R channel, a first image G channel, a second image G channel, a first image B channel and a second image B channel is adopted.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to the above-described embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (8)
1. A detection method of unbalanced defect samples based on a convolutional neural network is characterized by comprising the following steps:
(1) collecting early data: collecting two types of data, wherein one type of data is original defect data of a product, and making an original image of a training set; the second is that data without defects, called non-defective data, produced in the production environment forms a non-defective image;
(2) making a label image of the original defect data: marking the characteristics in the task target by using an image marking tool, and making a corresponding label image; for the non-defective data, this step is not processed;
(3) deep neural network training:
(a) and (3) coordinate correction: firstly, position correction is carried out on the label images, the absolute positions of central pixel points of all the label images are ensured to be consistent when the central pixel points are collected, and all the label images are cut to be the same size;
(b) sampling forms the neural network input: adopting 6-channel input, randomly taking two label images, fixing one label image as a non-defective image, randomly extracting the other label image with a defect or the non-defective image, and then cascading the two images to form 6-channel input;
finally obtaining a trained model;
(4) defect reasoning: reasoning by using the trained model in the step (3) and the collected defect-free image, randomly extracting a plurality of defect-free images for a new image in the actual production environment, cascading each defect-free image and the new image to form 6 channels for input, sending the input into a network to obtain output, and voting to obtain the final defect pixel;
the cascade connection in the step (b) in the step (3) forms 6-channel input, specifically: forming 6-channel input in a mode of cascading a first image RGB channel and a second image RGB channel; or, the cascade connection in the step (b) in the step (3) forms a 6-channel input, specifically: the 6-channel input is formed in a mode that a first image R channel, a second image R channel, a first image G channel, a second image G channel, a first image B channel and a second image B channel are cascaded.
2. The detection method according to claim 1, characterized in that: in the step (1), the proportion of the defect data to the total data is 20% or more, and the proportion of the non-defect data to the total data is 50% or more.
3. The detection method according to claim 1, characterized in that: in the step (1), the ratio of the defect data to the non-defect data is 1: 3-5.
4. The detection method according to claim 1, characterized in that: in the step (1), the original defect data of the product is collected by using a photographic device according to the defect type, and an original image of a training set is made.
5. The detection method according to claim 1, characterized in that: in the step (3), the neural network adopts the resnet as a backbone network, and the receptive field is enlarged by utilizing the cavity convolution and the ASPP structure so as to adapt to various defects with different sizes, lengths and shapes.
6. The detection method according to claim 1, characterized in that: the coordinate correction in the step (3) is as follows:
(1) setting the pixel value of the background to 255;
(2) averaging the length and the width of all the images to obtain a standard length and width value, and adjusting the length and the width of all the images to the standard value;
(3) the foreground pixel of each image has four boundary pixels of top, bottom, left and right, the coordinate values of the four boundary pixels of all the images are taken and respectively averaged, and a rectangular frame is determined according to the four average value coordinates, wherein the rectangular frame is a target rectangular frame of the foreground;
and adjusting foreground pixels of all images to ensure that the central point is close to the central point of the rectangular frame, adjusting the foreground size to ensure that the foreground is completely in the rectangular frame, and ensuring that four pixels at the top, the bottom, the left and the right are on corresponding sides of the rectangular frame.
7. The detection method according to claim 1, characterized in that: the sampling in step (3) forms a neural network input as follows: a complete random sampling is taken to extract a defect-free image from the database.
8. The detection method according to claim 1, characterized in that: the sampling in step (3) forms a neural network input as follows: the method comprises the steps of utilizing a classification neural network which is trained in advance to obtain expression vectors of all images in a database, when an image to be detected is input, firstly sending the image to be detected into the classification neural network to obtain the expression vectors of the image to be detected, then calculating the expression vectors of all images or partial images in the database, and selecting the images according to the similarity.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010939183.9A CN111932639B (en) | 2020-09-09 | 2020-09-09 | Detection method of unbalanced defect sample based on convolutional neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010939183.9A CN111932639B (en) | 2020-09-09 | 2020-09-09 | Detection method of unbalanced defect sample based on convolutional neural network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN111932639A CN111932639A (en) | 2020-11-13 |
CN111932639B true CN111932639B (en) | 2020-12-18 |
Family
ID=73308672
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010939183.9A Active CN111932639B (en) | 2020-09-09 | 2020-09-09 | Detection method of unbalanced defect sample based on convolutional neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111932639B (en) |
Families Citing this family (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112991236B (en) * | 2021-05-20 | 2021-08-13 | 南京甄视智能科技有限公司 | Image enhancement method and device based on template |
CN113724252A (en) * | 2021-10-11 | 2021-11-30 | 北京中科智眼科技有限公司 | Industrial image anomaly detection method based on deep dual network feature matching |
CN114140398A (en) * | 2021-11-15 | 2022-03-04 | 上海大学 | Few-sample defect detection method using defect-free image |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109993094A (en) * | 2019-03-26 | 2019-07-09 | 苏州富莱智能科技有限公司 | Fault in material intelligent checking system and method based on machine vision |
US10713769B2 (en) * | 2018-06-05 | 2020-07-14 | Kla-Tencor Corp. | Active learning for defect classifier training |
-
2020
- 2020-09-09 CN CN202010939183.9A patent/CN111932639B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US10713769B2 (en) * | 2018-06-05 | 2020-07-14 | Kla-Tencor Corp. | Active learning for defect classifier training |
CN109993094A (en) * | 2019-03-26 | 2019-07-09 | 苏州富莱智能科技有限公司 | Fault in material intelligent checking system and method based on machine vision |
Non-Patent Citations (2)
Title |
---|
基于聚类分析的不均衡数据标注技术研究;赵俊杰等;《计算机仿真》;20200229;第37卷(第02期);全文 * |
面向不均衡数据集的过抽样算法;崔鑫等;《计算机应用》;20200610(第06期);全文 * |
Also Published As
Publication number | Publication date |
---|---|
CN111932639A (en) | 2020-11-13 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108562589B (en) | Method for detecting surface defects of magnetic circuit material | |
CN110533084B (en) | Multi-scale target detection method based on self-attention mechanism | |
CN111932639B (en) | Detection method of unbalanced defect sample based on convolutional neural network | |
CN109242864B (en) | Image segmentation result quality evaluation method based on multi-branch network | |
CN108985169B (en) | Shop cross-door operation detection method based on deep learning target detection and dynamic background modeling | |
WO2022236876A1 (en) | Cellophane defect recognition method, system and apparatus, and storage medium | |
CN111815564B (en) | Method and device for detecting silk ingots and silk ingot sorting system | |
CN111967313B (en) | Unmanned aerial vehicle image annotation method assisted by deep learning target detection algorithm | |
CN112070727B (en) | Metal surface defect detection method based on machine learning | |
CN112132784A (en) | Method for classifying and segmenting industrial magnetic tile defect image based on small sample deep convolution neural network | |
CN111310756A (en) | Damaged corn particle detection and classification method based on deep learning | |
CN111914902B (en) | Traditional Chinese medicine identification and surface defect detection method based on deep neural network | |
CN112766334A (en) | Cross-domain image classification method based on pseudo label domain adaptation | |
CN111178177A (en) | Cucumber disease identification method based on convolutional neural network | |
CN113077444A (en) | CNN-based ultrasonic nondestructive detection image defect classification method | |
CN114170511A (en) | Pavement crack disease identification method based on Cascade RCNN | |
CN114359199A (en) | Fish counting method, device, equipment and medium based on deep learning | |
CN111161213B (en) | Industrial product defect image classification method based on knowledge graph | |
CN115731400A (en) | X-ray image foreign matter detection method based on self-supervision learning | |
CN116310548A (en) | Method for detecting invasive plant seeds in imported seed products | |
CN111598833A (en) | Method and device for detecting defects of target sample and electronic equipment | |
CN109615610B (en) | Medical band-aid flaw detection method based on YOLO v2-tiny | |
CN113077438B (en) | Cell nucleus region extraction method and imaging method for multi-cell nucleus color image | |
CN112528058B (en) | Fine-grained image classification method based on image attribute active learning | |
Zhang et al. | Design of tire damage image recognition system based on deep learning |
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 |