CN116758368A - Semi-supervised equipment defect and safety monitoring image labeling method - Google Patents
Semi-supervised equipment defect and safety monitoring image labeling method Download PDFInfo
- Publication number
- CN116758368A CN116758368A CN202310570777.0A CN202310570777A CN116758368A CN 116758368 A CN116758368 A CN 116758368A CN 202310570777 A CN202310570777 A CN 202310570777A CN 116758368 A CN116758368 A CN 116758368A
- Authority
- CN
- China
- Prior art keywords
- image
- labeling
- safety monitoring
- equipment
- equipment defect
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000002372 labelling Methods 0.000 title claims abstract description 89
- 230000007547 defect Effects 0.000 title claims abstract description 65
- 238000012544 monitoring process Methods 0.000 title claims abstract description 61
- 238000001514 detection method Methods 0.000 claims abstract description 41
- 238000000034 method Methods 0.000 claims abstract description 21
- 238000012549 training Methods 0.000 claims abstract description 13
- 238000003860 storage Methods 0.000 claims abstract description 7
- 238000004806 packaging method and process Methods 0.000 claims description 18
- 230000000694 effects Effects 0.000 claims description 5
- 238000007689 inspection Methods 0.000 claims 2
- 238000013135 deep learning Methods 0.000 abstract description 8
- 238000011160 research Methods 0.000 description 5
- 238000013473 artificial intelligence Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 238000004519 manufacturing process Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 230000006978 adaptation Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000004140 cleaning Methods 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 238000000586 desensitisation Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000012827 research and development Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/94—Hardware or software architectures specially adapted for image or video understanding
- G06V10/945—User interactive design; Environments; Toolboxes
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Evolutionary Computation (AREA)
- Databases & Information Systems (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a semi-supervised equipment defect and safety monitoring image labeling method, which comprises the following steps: step 1: training to obtain a detection model of the equipment defect or safety monitoring target; step 2: obtaining a result of automatic labeling of the system; step 3: manually correcting inaccurate labeling results; step 4: the corrected accurate marking information storage file and the corresponding equipment defect or safety monitoring image are used as input, and the equipment defect or safety monitoring target detection model is retrained; step 5: repeating the steps 2 to 4, and iteratively updating the model; step 6: and embedding the target detection model which is finally trained and obtained and reaches the precision into an image labeling system to finish the labeling work of the residual equipment defects or the safety monitoring images. According to the method, the deep learning target detection model is effectively combined with iterative image labeling work, so that equipment defects or safety monitoring image labeling in a semi-supervision mode is realized.
Description
Technical Field
The invention belongs to the technical field of deep learning, and particularly relates to a semi-supervised equipment defect and safety monitoring image labeling method.
Background
The data labeling is to classify, sort, edit, correct, mark and annotate the waiting labeling data of text, image, voice and video, and add labels to the waiting labeling data to produce machine readable data codes meeting the requirements of machine learning and neural network training. The data labeling is a necessary link for upgrading the traditional manufacturing to the intelligent manufacturing and upgrading the information computing to the artificial intelligence, and is a key link for effectively operating an artificial intelligence algorithm, and the quality of the data labeling directly determines the degree of machine intelligence.
Data annotation first requires the determination of a data set, which is generally divided into a test data set and a training data set. The training data set is firstly provided with desensitization data by an artificial intelligent production domain, after pretreatment works such as cleaning, processing, format conversion, feature extraction and the like, the pretreatment data are marked and classified by an artificial intelligent platform picture marking system, and finally the marked data are divided into a training data set and a testing data set, so that sample data and testing data are provided for algorithm research and development of an algorithm realization layer. In processing artificial intelligence business requirements, a data set is formed after feature processing and data labeling of business data through a large amount of time for model training optimization and model verification evaluation to be deployed on line after passing a model.
Because the artificial intelligence deep learning algorithm needs a large number of data samples, the problem that training sample labeling is time-consuming and labor-consuming is common in related research. Although most of conventional image labeling tasks can be completed by using the existing labeling tools, the situation that the number of detection targets is large and dense exists in some images, and the situation that labeling is possibly lost due to subtle detection targets exists, if the targets in the images are labeled one by using the conventional labeling tools without an automatic labeling function, on one hand, the workload is large, and on the other hand, the accuracy, consistency and integrity of image labeling during participation of multiple people are difficult to ensure. In addition, the semi-automatic labeling function of the existing image labeling tool has the problems of limited applicable object types and poor automatic labeling effect.
Disclosure of Invention
The invention aims to solve the defects in the prior art, the traditional manual labeling workload is huge and is not accurate enough, and provides a semi-supervised image labeling tool and method, aiming at the problems that training sample labeling work is time-consuming and labor-consuming in deep learning research, the defect of electric equipment is taken as a research object, a method for effectively combining a deep learning target detection model with iterative image labeling work is adopted, a semi-supervised image labeling system based on deep learning is developed, and an iterative method of 'detection model training-target automatic detection-manual labeling correction-detection model updating' is adopted, so that the image labeling in a semi-supervised mode is realized, and the automatic detection performance of the system is continuously optimized, thereby improving the image labeling efficiency and reducing the labor cost investment of the image labeling.
The technical scheme adopted by the first aspect of the invention is as follows:
the method is applied to a semi-supervised equipment defect and safety monitoring image labeling system, wherein the system is provided with a target detection model and an image database, and the method comprises the following steps of:
s1, taking part of manually marked equipment defects or safety monitoring images and corresponding marked information storage files as input, and training to obtain an equipment defect or safety monitoring target detection model;
s2, loading part of equipment defects or safety monitoring images to be marked, and detecting the equipment defects or safety monitoring targets in the images by using an equipment defect or safety monitoring target detection model embedded in the system to obtain a system automatic marking result;
s3, manually correcting the inaccurate labeling result to obtain an accurate labeling result;
s4, taking the corrected accurate marking information storage file and the corresponding equipment defect or safety monitoring image as input, retraining an equipment defect or safety monitoring target detection model to obtain a detection model with higher accuracy, and automatically marking the equipment defect or safety monitoring target in the image;
s5, repeating the steps 2 to 4, and iteratively updating the model until the detection effect of the target detection model reaches ideal precision, namely the average accuracy rate is more than 90%;
s6, embedding the target detection model which is finally trained and reaches ideal precision into an image marking system, and finishing marking work of residual equipment defects or safety monitoring images, so that more than 90% of equipment defects or safety monitoring targets in the images are accurately marked.
Further, in step 3, the inaccurate labeling result includes missed labeling, wrong labeling, and distorted labeling frame.
Further, in step 3, the accurate labeling result means that all equipment defects or safety monitoring in the image are labeled and the labeling frame is attached to the equipment defects or safety monitoring edge, and no labeling frame which does not contain the equipment defects or safety monitoring is provided.
Further, in step S3, the step of manually correcting the inaccurate labeling result to obtain an accurate labeling result further includes the sub-steps of:
s31, acquiring file names and paths from the images of the accurate labeling results;
s32, acquiring an image through the file name and the path of the image;
s33, matching and comparing the acquired image with the image database image through a difference hash algorithm;
s34, if the image database is not matched with the acquired image, the acquired image is added to the image database, and if the image database is matched with the acquired image, the acquired image is not added to the image database.
Further, the images reaching the ideal progress are packaged into an image database.
Further, the image database is provided with a private packaging interface and a public packaging interface, wherein the private packaging interface is used for internal connection, and the public packaging interface is used for external connection.
A second aspect of the present invention proposes a semi-supervised equipment defect, safety monitoring image annotation system for performing the method proposed by the first aspect, the system being provided with a target detection model and an image database.
The invention has the technical effects that:
the technology of the invention provides a semi-supervised image marking function and an artificial image marking function, which obviously reduces the complexity of equipment defects or safety monitoring image marking work and greatly reduces the workload of artificial marking. The research result provides a high-efficiency labeling method and tool for training sample labeling work in deep learning research, which is beneficial to improving image labeling efficiency and reducing labor cost investment.
Drawings
Fig. 1 is a schematic flow chart of a method for labeling a semi-supervised equipment defect and safety monitoring image.
Detailed Description
The present invention will be specifically described with reference to examples below in order to make the objects and advantages of the present invention more apparent. It should be understood that the following text is intended to describe only one or more specific embodiments of the invention and does not limit the scope of the invention strictly as claimed.
Referring to fig. 1, a method for labeling a semi-supervised equipment defect and a safety monitoring image is applied to a semi-supervised equipment defect and safety monitoring image labeling system, wherein the system is provided with a target detection model and an image database, and the method comprises the following steps:
s1, taking part of manually marked equipment defects or safety monitoring images and corresponding marked information storage files as input, and training to obtain an equipment defect or safety monitoring target detection model;
s2, loading part of equipment defects or safety monitoring images to be marked, and detecting the equipment defects or safety monitoring targets in the images by using an equipment defect or safety monitoring target detection model embedded in the system to obtain a system automatic marking result;
s3, manually correcting the inaccurate labeling result to obtain an accurate labeling result;
s4, taking the corrected accurate marking information storage file and the corresponding equipment defect or safety monitoring image as input, retraining an equipment defect or safety monitoring target detection model to obtain a detection model with higher accuracy, and automatically marking the equipment defect or safety monitoring target in the image;
s5, repeating the steps 2 to 4, and iteratively updating the model until the detection effect of the target detection model reaches ideal precision, namely the average accuracy rate is more than 90%;
s6, embedding the target detection model which is finally trained and reaches ideal precision into an image marking system, and finishing marking work of residual equipment defects or safety monitoring images, so that more than 90% of equipment defects or safety monitoring targets in the images are accurately marked.
By effectively combining a deep learning target detection model with iterative image labeling work, the device defect or safety monitoring image labeling in a semi-supervision mode is realized by adopting iterative operation of detection model training, target automatic detection, manual labeling correction and detection model updating.
In step 3, the inaccurate labeling result includes missed labeling, wrong labeling and wrong labeling frame skew error.
In step 3, the accurate labeling result means that all equipment defects or safety monitoring in the image are labeled, and the labeling frame is attached to the equipment defects or safety monitoring edge, and no labeling frame which does not contain the equipment defects or safety monitoring is provided.
In step S3, the step of manually correcting the inaccurate labeling result to obtain an accurate labeling result further includes the sub-steps of:
s31, acquiring file names and paths from the images of the accurate labeling results;
s32, acquiring an image through the file name and the path of the image;
s33, matching and comparing the acquired image with the image database image through a difference hash algorithm;
s34, if the image database is not matched with the acquired image, the acquired image is added to the image database, and if the image database is matched with the acquired image, the acquired image is not added to the image database.
For example, by matching the acquired image with the image of the image data, the labeling efficiency can be improved, the problem of repeated image repeated labeling is avoided, for the process of the difference hash algorithm, the image has 72 pixels by reducing the size, namely to 9*8, the image is subjected to gray processing, the difference value is calculated to obtain the final hash value, and by comparing the left pixel and the right pixel of each row, if the left pixel is brighter than the right pixel (the left pixel value is larger than the right pixel value), the pixel is recorded as 1, otherwise, the pixel is 0. Since each row has 9 pixels, the left and right two are sequentially compared to obtain 8 values, so that the total of 8 rows of pixels can obtain 64 values, the hash value is a 0-1 sequence with the length of 64, and finally the Hamming distance is calculated through the picture pairing to obtain the similarity of the images.
And packaging the images reaching the ideal progress into an image database. For example, image data of an image database may be enriched and updated by encapsulating images that reach an ideal schedule into the image database.
The image database is provided with a private packaging interface and a public packaging interface, wherein the private packaging interface is used for internal connection, and the public packaging interface is used for external connection.
By means of the private packaging interface and the public packaging interface, the private packaging interface can facilitate access to the private packaging interface and avoid other users from accessing and acquiring image data, and the private packaging interface can be connected with an external image database, so that the volume of the image data of the database is greatly enriched, rapid iteration of a full model can be achieved, and the efficiency of image labeling is improved.
A second aspect of the present invention proposes a semi-supervised equipment defect, safety monitoring image annotation system for performing the method proposed by the first aspect, the system being provided with a target detection model and an image database.
The foregoing is merely a preferred embodiment of the present invention and it should be noted that modifications and adaptations to those skilled in the art may be made without departing from the principles of the present invention, which are intended to be comprehended within the scope of the present invention. Structures, devices and methods of operation not specifically described and illustrated herein, unless otherwise indicated and limited, are implemented according to conventional means in the art.
Claims (7)
1. The method is characterized by being applied to a semi-supervised equipment defect and safety monitoring image labeling system, wherein the system is provided with a target detection model and an image database, and the method comprises the following steps of:
s1, taking part of manually marked equipment defects or safety monitoring images and corresponding marked information storage files as input, and training to obtain an equipment defect or safety monitoring target detection model;
s2, loading part of equipment defects or safety monitoring images to be marked, and detecting the equipment defects or safety monitoring targets in the images by using an equipment defect or safety monitoring target detection model embedded in the system to obtain a system automatic marking result;
s3, manually correcting the inaccurate labeling result to obtain an accurate labeling result;
s4, taking the corrected accurate marking information storage file and the corresponding equipment defect or safety monitoring image as input, retraining an equipment defect or safety monitoring target detection model to obtain a detection model with higher accuracy, and automatically marking the equipment defect or safety monitoring target in the image;
s5, repeating the steps 2 to 4, and iteratively updating the model until the detection effect of the target detection model reaches ideal precision, namely the average accuracy rate is more than 90%;
s6, embedding the target detection model which is finally trained and reaches ideal precision into an image marking system, and finishing marking work of residual equipment defects or safety monitoring images, so that more than 90% of equipment defects or safety monitoring targets in the images are accurately marked.
2. The method for labeling a semi-supervised equipment defect and safety monitoring image according to claim 1, wherein in the step 3, inaccurate labeling results include missed labeling, wrong labeling and mislabel frame deviation errors.
3. The method for labeling a semi-supervised equipment defect and safety monitoring image according to claim 1, wherein in step 3, the accurate labeling result means that all equipment defects or safety monitoring in the image are labeled and a labeling frame is attached to an equipment defect or safety monitoring edge, and no labeling frame which does not contain the equipment defect or safety monitoring is provided.
4. The method for labeling a semi-supervised equipment defect and safety inspection image according to claim 1, wherein in step S3, the inaccurate labeling result is manually corrected to obtain the accurate labeling result, further comprising the sub-steps of:
s31, acquiring file names and paths from the images of the accurate labeling results;
s32, acquiring an image through the file name and the path of the image;
s33, matching and comparing the acquired image with the image database image through a difference hash algorithm;
s34, if the image database is not matched with the acquired image, the acquired image is added to the image database, and if the image database is matched with the acquired image, the acquired image is not added to the image database.
5. The method for labeling a semi-supervised equipment defect and safety inspection image according to claim 1, wherein in step S5, the image reaching the ideal progress is packaged into an image database.
6. The method for labeling a semi-supervised equipment defect and security monitoring image according to claim 1, wherein the image database is provided with a private packaging interface and a public packaging interface, the private packaging interface is used for internal connection, and the public packaging interface is used for external connection.
7. A semi-supervised equipment defect, safety monitoring image annotation system, characterized in that the system is adapted to perform any of the methods according to claims 1-6, the system being provided with an object detection model and an image database.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310570777.0A CN116758368A (en) | 2023-05-18 | 2023-05-18 | Semi-supervised equipment defect and safety monitoring image labeling method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310570777.0A CN116758368A (en) | 2023-05-18 | 2023-05-18 | Semi-supervised equipment defect and safety monitoring image labeling method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116758368A true CN116758368A (en) | 2023-09-15 |
Family
ID=87948669
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310570777.0A Pending CN116758368A (en) | 2023-05-18 | 2023-05-18 | Semi-supervised equipment defect and safety monitoring image labeling method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116758368A (en) |
-
2023
- 2023-05-18 CN CN202310570777.0A patent/CN116758368A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108717545A (en) | A kind of bank slip recognition method and system based on mobile phone photograph | |
| EP3848888A1 (en) | Method and apparatus for generating vehicle damage image based on gan network | |
| CN110992349A (en) | Underground pipeline abnormity automatic positioning and identification method based on deep learning | |
| CN110648310A (en) | Weak supervision casting defect identification method based on attention mechanism | |
| CN110276587A (en) | The method, apparatus of project examination calculates equipment and computer readable storage medium | |
| CN111223081A (en) | Part hole opening recognition and detection method and system based on deep learning | |
| CN117785851A (en) | Product service identification marking method based on selected industry data cleaning | |
| CN116758368A (en) | Semi-supervised equipment defect and safety monitoring image labeling method | |
| CN116563735A (en) | Transmission tower inspection image focusing judgment method based on depth artificial intelligence | |
| CN114693554B (en) | Big data image processing method and system | |
| CN113920294B (en) | Method, system and storage medium for identifying massive invoices | |
| CN116089944A (en) | Cross-platform application program abnormality detection method and system based on transfer learning | |
| CN115908977A (en) | Image data labeling method and device, electronic equipment and storage medium | |
| CN112801013B (en) | Face recognition method, system and device based on key point recognition verification | |
| CN115457305A (en) | Semi-supervised target detection method and system | |
| CN110956174A (en) | Device number identification method | |
| CN112164040A (en) | Steel surface defect identification method based on semi-supervised deep learning algorithm | |
| Gao et al. | Intelligent appearance quality detection of air conditioner external unit and dataset construction | |
| CN116993721B (en) | Steel plate surface defect detection method based on weak supervision neural network | |
| CN111221995B (en) | Sequence matching method based on big data and set theory | |
| Zhang et al. | OCR-based aircraft maintenance report data structuring | |
| CN113283452B (en) | Large-scale equipment mounting and dismounting step detection method | |
| CN112749888B (en) | Multi-element random spot check method, system and device based on RANSAC algorithm | |
| CN110738209B (en) | License plate detection method based on deep learning | |
| CN110781932B (en) | Ultrahigh-definition film source color gamut detection method for multi-class image conversion and comparison |
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 |