CN109978029B - Invalid image sample screening method based on convolutional neural network - Google Patents
Invalid image sample screening method based on convolutional neural network Download PDFInfo
- Publication number
- CN109978029B CN109978029B CN201910188287.8A CN201910188287A CN109978029B CN 109978029 B CN109978029 B CN 109978029B CN 201910188287 A CN201910188287 A CN 201910188287A CN 109978029 B CN109978029 B CN 109978029B
- Authority
- CN
- China
- Prior art keywords
- sample
- invalid
- model
- samples
- images
- 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
- 238000000034 method Methods 0.000 title claims abstract description 17
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 15
- 238000012216 screening Methods 0.000 title claims abstract description 13
- 238000001914 filtration Methods 0.000 claims abstract description 10
- 238000004140 cleaning Methods 0.000 claims description 8
- 238000009776 industrial production Methods 0.000 claims description 2
- 238000004519 manufacturing process Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 238000007670 refining Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/10—Pre-processing; Data cleansing
-
- 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
Abstract
The invention discloses a method for screening invalid image samples based on establishment of a convolutional neural network filtering sample model, wherein a large number of invalid sample (redundant sample) images such as fuzzy, blank shooting and damage are screened out from an original sample through the filtering sample model established by the convolutional neural network, and the rest samples are image samples with better quality and more representativeness and can be used as an effective sample set for image classification. The method can be finally realized as follows: and the invalid samples are screened out through an algorithm, so that the working hours consumed by screening out a large number of invalid samples are reduced, and the labor cost is reduced.
Description
Technical Field
The invention relates to the field of machine learning, in particular to a method for screening invalid image samples based on establishment of a convolutional neural network filtering sample model.
Background
When the convolutional neural network is used for image classification, a large number of image samples are required to be classified as a sample library for constructing a model. In an actual industrial process, compared with a normal industrial process, some invalid image data, such as a blurred image, an empty shot image and a damaged image, which are acquired during pipeline production, often exist in acquired product image data, and the images belong to invalid data for model construction.
Disclosure of Invention
The invention mainly solves the technical problem of providing a method for screening invalid image samples based on establishing a convolutional neural network filtering sample model, which can screen out fuzzy images, blank shot images and the like from collected samples with less labor hour and labor cost and realize the cleaning of original samples.
In order to solve the technical problems, the invention adopts a technical scheme that: and (3) blurring the part of the sample image which is not marked, screening out the image with extremely blurred sample, and performing primary blurring cleaning.
The processed samples are classified into invalid images, namely blurred images, blank shot images and the like, and effective sample images through manual classification.
The extremely blurred images processed by the filter are also divided into invalid images, two types of image division are formed, and a sample library is formed.
The CNN algorithm adopts the sample library to construct a classifier model containing two classification conditions, and then the model is used for cleaning samples, so that invalid samples in a large number of samples are screened out, and the samples are cleaned.
The invention has the beneficial effects that: the method adopts the convolutional neural network to clean the samples, realizes the screening of invalid samples in a large number of samples, reduces the working hours consumed by image screening, and reduces the labor cost.
Drawings
FIG. 1 is a schematic flow diagram of a method for invalid image sample screening based on building a convolutional neural network filter sample model;
FIG. 2 is a schematic flow diagram of a process for constructing a model of a filtered sample;
Detailed Description
The following detailed description of the preferred embodiments of the present invention, taken in conjunction with the accompanying drawings, will make the advantages and features of the invention easier to understand by those skilled in the art, and thus will clearly and clearly define the scope of the invention.
Referring to fig. 1 and 2, an embodiment of the present invention includes:
a method for screening invalid image samples based on a convolutional neural network filtering sample model is characterized in that a convolutional neural network is used for constructing the model, the model is applied to collected samples, less labor hour and labor cost are spent on screening invalid data such as fuzzy images and blank shot images from the collected samples, and the original samples are cleaned.
The first embodiment is as follows: blurring and refining the sample
(1) Preparation of an original sample: image data of a produced product is collected in industrial production, and 1000 sheets are taken as a basic quantity.
(2) A fuzzy filter: the method of cv2 Laplacian () of opencv can be used for realizing the filtering of pictures with high fuzzy degree and realizing one-time cleaning work on samples.
Fuzzy and refined sample obtaining mode: and (4) the prepared basic sample is processed by a fuzzy filter, a picture with higher fuzzy degree is filtered out to be used as a fuzzy sample, and the rest samples are used as refined samples.
Example two: sample collection
(1) Manual filtration: when the sample model is not filtered, a manual filtering mode is needed to classify the refined samples into an invalid sample type and a valid sample type.
(2) Type of invalid sample: and taking sample images such as the blurred image, the blank shot image and the damage as invalid sample types.
(3) Valid sample types: and other clear and obvious images serve as effective sample types.
(4) Sample set: the two types of invalid samples and valid samples form a sample set.
Example three: constructing a model of a filtered sample
(1) The algorithm is as follows: and realizing sample cleaning based on a deep convolutional neural network algorithm.
(2) Judging the number and proportion of samples: if the positive and negative samples are not balanced, the following method can be adopted:
oversampling: the number of minority class samples in the samples is increased. A few samples are copied or random noise is added into the few samples, and interference data generate certain samples through certain rules.
Down-sampling: reducing the number of majority samples. Most samples are randomly removed until most and few samples are the same.
(3) And (3) constructing a model, and starting to calculate the model when the number of positive and negative samples in the sample set is proper and proportional and balanced, such as the flow of fig. 2.
Accuracy (accuracuracy):
recall (recall) is a measure of coverage, and there are several positive examples divided into positive examples:
when the model accuracy rate ACC is low or the recall rate recall is small, the calculated model does not meet the requirements.
(4) And putting the calculated new model into the flow of the figure 1 for processing.
Example four: sample library meeting requirements
(1) Judging whether the sample library meets the requirements: the number of sample banks, etc. meet the requirements.
(2) And (3) circulating treatment: generally, a sufficient number of sample libraries cannot be obtained once, so when the sample libraries do not meet the requirements, circulation is started, the original samples are filtered by the fuzzy filter to obtain fuzzy images, the sample images are classified by the filtered sample model, the classification of the new images is judged, and sample cleaning is realized.
Claims (5)
1. A method for carrying out invalid image sample screening based on establishment of a convolutional neural network filtering sample model comprises the following steps:
s1: collecting image data of products in industrial production as a basis to form an original sample set;
s2: constructing a fuzzy filter to filter pictures with high fuzzy degree;
s3: the residual image filtered by the fuzzy filter, namely the refined sample, is manually classified to form an invalid sample type and an effective sample type, the extremely-fuzzy image processed by the fuzzy filter is also classified into the invalid sample type, and the invalid sample type comprises the following steps: blurred images, blank shot images and damaged images;
s4: the two types of the invalid sample and the valid sample form a sample set;
s5: constructing a convolutional neural network filtering sample model;
s6: and classifying a large number of original sample images by using the fuzzy filter of S2 and the model loop of S5 to realize sample cleaning until a sample library meeting the requirements is built.
2. The method of claim 1, wherein: before the step of S3, a blur filter is added in the step of S2 to filter out a sample image with a high degree of blur in advance.
3. The method of claim 1, wherein: the refined sample is manually processed in step S3.
4. The method of claim 1, wherein: and step S5, calculating the model when the quantity of the positive and negative samples in the sample set is proper and proportional balance is achieved, and abandoning the model when the model recall rate and the model accuracy rate do not meet the requirements.
5. The method of claim 1, wherein: and when the sample library does not meet the requirement, starting a loop in the step S6, classifying a large number of original sample images, and cleaning the samples.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910188287.8A CN109978029B (en) | 2019-03-13 | 2019-03-13 | Invalid image sample screening method based on convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910188287.8A CN109978029B (en) | 2019-03-13 | 2019-03-13 | Invalid image sample screening method based on convolutional neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109978029A CN109978029A (en) | 2019-07-05 |
| CN109978029B true CN109978029B (en) | 2021-02-09 |
Family
ID=67078702
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910188287.8A Active CN109978029B (en) | 2019-03-13 | 2019-03-13 | Invalid image sample screening method based on convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109978029B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110990917A (en) * | 2019-11-19 | 2020-04-10 | 北京长空云海科技有限公司 | BIM model display method, device and system |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5590218A (en) * | 1993-10-18 | 1996-12-31 | Bayer Corporation | Unsupervised neural network classification with back propagation |
| CN106067020A (en) * | 2016-06-02 | 2016-11-02 | 广东工业大学 | The system and method for quick obtaining effective image under real-time scene |
| CN108960409A (en) * | 2018-06-13 | 2018-12-07 | 南昌黑鲨科技有限公司 | Labeled data generation method, equipment and computer readable storage medium |
| CN109379557A (en) * | 2018-09-30 | 2019-02-22 | 田东县文设芒果专业合作社 | Mango insect pest intelligent monitor system based on image recognition |
Family Cites Families (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7991223B2 (en) * | 2006-09-05 | 2011-08-02 | Bruker Daltonik Gmbh | Method for training of supervised prototype neural gas networks and their use in mass spectrometry |
| KR102592076B1 (en) * | 2015-12-14 | 2023-10-19 | 삼성전자주식회사 | Appartus and method for Object detection based on Deep leaning, apparatus for Learning thereof |
| CN107909566A (en) * | 2017-10-28 | 2018-04-13 | 杭州电子科技大学 | A kind of image-recognizing method of the cutaneum carcinoma melanoma based on deep learning |
| CN108171175B (en) * | 2017-12-29 | 2020-06-23 | 苏州科达科技股份有限公司 | Deep learning sample enhancement system and operation method thereof |
| CN108154134B (en) * | 2018-01-11 | 2019-07-23 | 天格科技（杭州）有限公司 | Pornographic image detection method is broadcast live in internet based on depth convolutional neural networks |
| CN108764372B (en) * | 2018-06-08 | 2019-07-16 | Oppo广东移动通信有限公司 | Construction method and device, mobile terminal, the readable storage medium storing program for executing of data set |
| CN108986075A (en) * | 2018-06-13 | 2018-12-11 | 浙江大华技术股份有限公司 | A kind of judgment method and device of preferred image |
| CN109165671A (en) * | 2018-07-13 | 2019-01-08 | 上海交通大学 | Confrontation sample testing method based on sample to decision boundary distance |
| CN109117887A (en) * | 2018-08-17 | 2019-01-01 | 哈尔滨工业大学 | A kind of support vector machines accelerating method and device for low-dimensional data screening sample |
| CN109241903B (en) * | 2018-08-30 | 2023-08-29 | 平安科技（深圳）有限公司 | Sample data cleaning method, device, computer equipment and storage medium |
-
2019
- 2019-03-13 CN CN201910188287.8A patent/CN109978029B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5590218A (en) * | 1993-10-18 | 1996-12-31 | Bayer Corporation | Unsupervised neural network classification with back propagation |
| CN106067020A (en) * | 2016-06-02 | 2016-11-02 | 广东工业大学 | The system and method for quick obtaining effective image under real-time scene |
| CN108960409A (en) * | 2018-06-13 | 2018-12-07 | 南昌黑鲨科技有限公司 | Labeled data generation method, equipment and computer readable storage medium |
| CN109379557A (en) * | 2018-09-30 | 2019-02-22 | 田东县文设芒果专业合作社 | Mango insect pest intelligent monitor system based on image recognition |
Non-Patent Citations (3)
| Title |
|---|
| Consensus of Ambiguity: Theory and Application of Active Learning for Biomedical Image Analysis;Scott Doyle 等;《pattern recognition in bioinformatics-5th IAPR international conference》;20100930;314-324 * |
| Efficient Multiple Instance Convolutional Neural Networks for Gigapixel Resolution Image Classification;Le Hou 等;《arXiv：1504.07947v3》;20150519;1-9 * |
| 基于卷积神经网络的小样本图像识别方法;段萌 等;《计算机工程与设计》;20180131;第39卷(第1期);224-229 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109978029A (en) | 2019-07-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110782399B (en) | Image deblurring method based on multitasking CNN | |
| CN107507173B (en) | No-reference definition evaluation method and system for full-slice image | |
| CN105608679B (en) | A kind of image de-noising method of fusion structure tensor and the full variation of non-local | |
| CN105894460A (en) | Image filtering method and device | |
| CN109740721B (en) | Wheat ear counting method and device | |
| CN106910169A (en) | A kind of image salt-pepper noise minimizing technology for preventing edge blurry | |
| CN111598918B (en) | Video image stabilizing motion estimation method based on reference frame optimization and foreground and background separation | |
| CN110097522B (en) | Single outdoor image defogging method based on multi-scale convolution neural network | |
| CN110490822B (en) | Method and device for removing motion blur of image | |
| CN109978029B (en) | Invalid image sample screening method based on convolutional neural network | |
| CN106910170B (en) | A kind of minimizing technology of image salt-pepper noise | |
| CN109118434A (en) | A kind of image pre-processing method | |
| CN109801231B (en) | Image processing method of electrophoresis electronic paper detection equipment | |
| CN110942436A (en) | Image deblurring method based on image quality evaluation | |
| WO2017177559A1 (en) | Image management method and apparatus | |
| CN113554649A (en) | Defect detection method and device, computer equipment and storage medium | |
| CN108107611A (en) | A kind of adaptive defect inspection method, device and electronic equipment | |
| CN114663427A (en) | Boiler part size detection method based on image processing | |
| CN113034404A (en) | Traffic image deblurring method and device based on multi-scale counterstudy | |
| CN111104976B (en) | Blue algae coverage calculating method based on time sequence images | |
| CN107248143A (en) | A kind of depth image restorative procedure split based on image | |
| CN115008255B (en) | Tool wear identification method and device for machine tool | |
| CN115984613A (en) | Fundus image classification method, device, equipment and storage medium | |
| CN114486916A (en) | Mobile phone glass cover plate defect detection method based on machine vision | |
| CN114897214A (en) | Metal additive manufacturing time prediction system and method based on graphical processing |
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 | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20240129 Address after: Room 401-55, No. 821 Lianting Road, Min'an Street, Xiang'an District, Xiamen City, Fujian Province, 361101 Patentee after: Xiamen Xinhang Century Information Technology Co.,Ltd. Country or region after: China Address before: 100876 Beijing city Haidian District Xitucheng Road No. 10 Patentee before: Beijing University of Posts and Telecommunications Country or region before: China |
|
| TR01 | Transfer of patent right |