CN117011300B - Micro defect detection method combining instance segmentation and secondary classification - Google Patents
Micro defect detection method combining instance segmentation and secondary classification Download PDFInfo
- Publication number
- CN117011300B CN117011300B CN202311280842.2A CN202311280842A CN117011300B CN 117011300 B CN117011300 B CN 117011300B CN 202311280842 A CN202311280842 A CN 202311280842A CN 117011300 B CN117011300 B CN 117011300B
- Authority
- CN
- China
- Prior art keywords
- defect
- image
- model
- information
- combining
- 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 118
- 238000001514 detection method Methods 0.000 title claims abstract description 75
- 230000011218 segmentation Effects 0.000 title claims abstract description 19
- 238000012549 training Methods 0.000 claims abstract description 27
- 238000004806 packaging method and process Methods 0.000 claims abstract description 9
- 238000007781 pre-processing Methods 0.000 claims abstract description 9
- 238000004519 manufacturing process Methods 0.000 claims abstract description 8
- 230000005540 biological transmission Effects 0.000 claims abstract description 6
- 238000010276 construction Methods 0.000 claims abstract description 4
- 238000000034 method Methods 0.000 claims description 13
- 238000002372 labelling Methods 0.000 claims description 12
- 238000013145 classification model Methods 0.000 claims description 4
- 238000001914 filtration Methods 0.000 claims description 4
- 238000012937 correction Methods 0.000 claims description 3
- 238000013135 deep learning Methods 0.000 claims description 3
- 230000006872 improvement Effects 0.000 claims description 3
- 230000009467 reduction Effects 0.000 claims description 3
- 230000001360 synchronised effect Effects 0.000 claims description 3
- 238000013526 transfer learning Methods 0.000 claims description 3
- 230000003044 adaptive effect Effects 0.000 claims description 2
- 238000005516 engineering process Methods 0.000 claims description 2
- 238000013508 migration Methods 0.000 claims description 2
- 230000005012 migration Effects 0.000 claims description 2
- 238000012360 testing method Methods 0.000 description 6
- 238000004364 calculation method Methods 0.000 description 5
- 230000002950 deficient Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 2
- 230000004075 alteration Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 238000003708 edge detection Methods 0.000 description 1
- 230000005484 gravity Effects 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000001629 suppression Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
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
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/096—Transfer learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
- G06T7/62—Analysis of geometric attributes of area, perimeter, diameter or volume
-
- 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
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- 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
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- General Health & Medical Sciences (AREA)
- Databases & Information Systems (AREA)
- Multimedia (AREA)
- Medical Informatics (AREA)
- Biophysics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Geometry (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to the technical field of light transmission irradiation detection, in particular to a micro defect detection method combining example segmentation and secondary classification, which comprises the following steps: collecting a defect image; image annotation, namely generating a first micro defect data set; pre-training to obtain an optimal detection model; obtaining an optimal EfficientNet model; packaging the optimal detection model to complete the construction of a micro defect detection system combining the example segmentation and the secondary classification; the system is deployed on an actual production line, images of the tested piece are collected through a camera, image noise is reduced in an S1 preprocessing mode, and then the images are input into the system for defect detection, so that the output of defect types, positions and areas is realized. By manufacturing a high-quality defect data set and based on a yolov7-seg model, the example segmentation of defects is realized, the defect positions and mask frames can be synchronously acquired, and the defect area can be directly calculated based on the mask frames.
Description
Technical Field
The invention relates to the technical field of light transmission irradiation detection, in particular to a micro defect detection method combining example segmentation and secondary classification.
Background
In the field of light transmission irradiation detection, a canny edge detection algorithm is often used for defect identification, gaussian filtering is firstly adopted for detecting pictures to smooth images, noise in the images is removed, gradient amplitude and angle images are calculated, non-maximum suppression is then applied to gradients, finally double-threshold processing and connection analysis are used for detecting edges, minimum area rectangular fitting is carried out on defect positions after defects are identified, defect sizes are set in an upper computer, and defects are judged to exist in the detected pictures when the detected defect sizes are larger than the set sizes.
However, the fringes produced by gravity behave similarly to defects in the gray scale of the visual image, edges in the image can only be identified once for the canny operator, and image noise which may exist should not be identified as edges, and detection based on edges is poor for small fringes or non-defective targets having similar characteristics to defects, and false detection is likely to occur. The minimum area rectangular fitting has poor fitting effect on irregular defects and micro defects, and the rectangular area has deviation from the actual defect area, but for small detected stripes, the rectangular area is larger than the actual defect area, so that inaccurate detection is caused.
Disclosure of Invention
Based on the above situation, the invention provides a micro defect detection method combining example segmentation and secondary classification, which solves the problems of easy false detection of a non-defect target containing approximate defect characteristics and larger difference between a defect area calculation result and actual detection caused by vision-based contour detection in the prior art.
The invention provides the following technical scheme: a method for detecting micro defects by combining instance segmentation and secondary classification comprises the following steps:
s1, acquiring a defect image, performing image acquisition on a detected piece, preprocessing the acquired defect image to obtain first original data, and acquiring an image without defects to be used as a first original background image;
s2, labeling the images, namely respectively labeling the first original data according to categories, storing the first original data as json format labels containing image information, expanding the original data and the corresponding labels, and combining the first original background images to generate a first micro defect data set;
s3, pre-training, namely dividing 10% of the first micro defect data set into a first pre-training data set, training a yolov7-seg deep learning network model to obtain a first initial detection model, and performing migration learning and training on the first initial detection model to obtain an optimal detection model;
s4, the first image acquisition program A1 cuts an original image in the tiny defect data set through the acquired defect position information to obtain a 224 x 224 size image containing a defect target, combines the acquired image, makes a second defect classification data set based on the acquired defect labeling information, trains a rapid classification model EfficientNet, and obtains an optimal EfficientNet model;
s5, packaging an optimal detection model, packaging an image acquisition program A1 and the optimal Efficient Net model into the optimal detection model, finishing improvement of the optimal detection model, outputting and decoding the Efficient Net model, obtaining classification information and confidence coefficient, carrying out logic operation B1 on the classification information and the decoded classification information of the optimal detection model, obtaining final classification information, combining the output information of the optimal detection model obtained by decoding, finishing output of all detection results, and finishing construction of a micro defect detection system combining example segmentation and secondary classification;
s6, deploying the system on an actual production line, collecting an image of a tested piece through a camera, reducing image noise in a preprocessing mode in S1, then inputting the image noise into the system for defect detection, and finally realizing the output of defect types, positions and areas.
In step S1, when an image of a measured object is acquired, information of a micro defect which is located in the measured object and cannot be directly observed and a non-defect target containing similar defect characteristics is acquired in a light source transmission mode, the information is used as first original data after mean filtering noise reduction and adaptive gamma correction and equalization image brightness preprocessing, the acquired first original data is screened, and the information is classified according to defect sizes and characteristics.
In step S2, the original defect image and the corresponding label are expanded by adopting data expansion and adopting rotation, scaling, mirroring and brightness adjustment modes, and a first micro defect data set is generated by combining the first original background image, wherein the data set size is 640 x 640, and the label is in txt format.
The first original data are 5000 pieces in total, original defect images and corresponding labels are expanded by adopting the modes of random clockwise rotation and anticlockwise rotation by 0-15 degrees, random scaling by 0.85-1.15 times, mirroring and random brightness adjustment by 0.8-1.1 times, and each mode is applied to each image and label with 50% probability and is expanded to 4 times.
In step S3, after the first initial detection model is obtained, the first initial detection model is used as a basic model to be loaded, a transfer learning technology is adopted, data except a training data set is used for training, the performance of the trained model is verified, and an optimal detection model is selected.
When the optimal detection model is packaged, model output is called, and information such as a labeling category, a labeling frame, a mask and the like is decoded and obtained, so that synchronous output of defect category, size, position and area information is realized.
As can be seen from the above description, the beneficial effects of the present solution are: 1. by manufacturing a high-quality defect data set and training a yolov7-seg model based on accurate annotation information, the example segmentation of defects can be realized, defect positions and mask frames can be synchronously acquired, the defect area can be directly calculated based on the mask frames, and the calculation result is closer to the actual area. 2. The defect type and position information of the yolov7-seg model are read through an image acquisition program A1, the production of a high-quality defect classification data set is completed, a high-precision EfficientNet model is trained for secondary classification of defects, the type information and the confidence coefficient in the output of the yolov7-seg model are obtained through decoding, and the accurate differentiation of the defects is realized by combining with a logic operation B1B1, so that the detection accuracy is improved, and the false detection rate of non-defect targets containing approximate defect characteristics is reduced; 3. the double-model can improve the detection rate and simultaneously reduce the false detection rate by carrying out logic operation on the extracted two model result information; 4. the defect area is calculated through the mask output by the real force dividing branch, the accuracy is higher than that of the minimum rectangle method, and the calculated area is closer to the actual area.
Drawings
FIG. 1 is a flow chart of an embodiment of the present invention.
Fig. 2 shows PR curves trained with different parameters.
Fig. 3 is a flowchart of a first image acquisition procedure A1.
Fig. 4 is an effect net training result.
Fig. 5 is a schematic diagram of a packaging mode.
Fig. 6 is a schematic diagram of a logic operation B1.
Fig. 7 is a schematic diagram of an area calculation method M1.
Detailed Description
The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is apparent that the described embodiment is only one embodiment of the present invention, not all embodiments. All other embodiments, which can be made by one of ordinary skill in the art without undue burden from the present disclosure, are within the scope of the present disclosure.
As can be seen from the accompanying drawings, the method for detecting micro defects by combining example segmentation and secondary classification of the present embodiment comprises the following steps
And acquiring a defect image, acquiring an image of a measured piece, acquiring information of a tiny defect which is positioned in the measured piece and cannot be directly observed and a non-defect target containing approximate defect characteristics in a light source transmission mode, and preprocessing the brightness of the image by mean filtering, noise reduction and self-adaptive gamma correction and equalization to serve as first original data. And screening the obtained first original data, and classifying according to the defect size and the characteristics. A few images without defects are acquired for use as the first original background image.
Marking an original image, namely marking first original data according to categories, marking a tiny defect as A, marking a non-defect target mark containing approximate defect characteristics as B, and storing the tiny defect as json format labels containing image information, wherein the total number of the tiny defects is about 5000; the data expansion adopts the modes of random clockwise and anticlockwise rotation by 0-15 degrees, random scaling by 0.85-1.15 times, mirroring, random brightness adjustment by 0.8-1.1 times and the like to expand the original defect image and the corresponding label, and each mode is applied to each picture and label with 50% probability to 4 times of quantity. And combining the first original background image to generate a first micro defect data set, wherein the size of the data set is 640 x 640, and finally, the label is converted into txt format, and the data set image is about 20000 sheets.
Pre-training, 10% of the first micro defect data set is divided into a first pre-training data set, 10% of the first pre-training data set is taken as a test set, and the rest is taken as a formal training data set. And training the yolov7-seg deep learning network model by adopting a first pre-training data set to obtain a first initial detection model. Training parameters set epoch=200, batch size=16, woker=4, using gpu training. The training is then performed using data outside the formal training data set using a transfer learning technique by using the first initial detection model as a base model load, with the training parameters set epoch=1500, batch size=16, woker=4, and epoch=1500, batch size=8, woker=8 each trained once. The performance of the trained model is verified through the test set, as shown in fig. 2, it can be seen that the model trained by the batch size=8 training parameters has higher accuracy, and the model is selected as the optimal detection model.
And packaging the optimal detection model, calling the model to output, decoding to obtain information such as the labeling category, the labeling frame, the mask and the like, and synchronously outputting the information such as the defect category, the size, the position, the area and the like. As shown in fig. 3, the first image acquisition program A1 acquires 224×224 size images including the target by acquiring the defect position, the category information, and the measured image information of the optimum detection model label and cropping the corresponding original image in the dataset. And combining the acquired images, wherein A1 is used for manufacturing a second defect classification data set in the same way as before based on the acquired defect marking information, and about 4000 manufactured data set images are manufactured. Wherein class a labeled by the yolov7-seg model is classified as inc and class B is classified as err, and a rapid classification model, namely a high-speed classification model EfficientNet, is trained by using gpu according to parameter settings of epoch=50, batch size=8 and woker=4, and the trained model precision reaches 1.0, as shown in fig. 4.
And (5) packaging the image acquisition program A1 and the optimal EfficientNet model into an optimal detection model to finish the improvement of the optimal detection model, wherein the packaging mode is shown in figure 5. Inputting class information output by a yolov7-seg classification branch into an image acquisition program A1, and reading class confidence degrees C1 and C2 (corresponding to two classes A\B respectively); inputting position information of yolov7-seg detection branch output into an image acquisition program A1, reading upper left corner points (x 0, y 0) of a defect labeling frame, determining a clipping center, and performing (224 ) range clipping on a model input image; the cropped image is then input into the optimal EfficientNet model, which outputs the detected class information and together with the class information output by the yolov7-seg classification branch is input into the logic operation B1, thereby obtaining the final class, as shown in FIG. 6.
After the final category information is obtained, the output of all detection results is completed by combining with other output information of the optimal detection model obtained by decoding, and the construction of a micro defect detection system combining the example segmentation and the secondary classification is completed. The system inputs an acquired image of a detected target, outputs information such as defect positions, categories, areas and the like, and can perform defect labeling.
The calculation mode M1 of the defect area adopts mask output of yolov7-seg, compared with 80 pixels obtained by directly calculating the minimum circumscribed rectangular area, the actual area is 12mm 2 The mask calculated pixel is 71 and the area is about 11mm 2 Closer to the actual area as shown in fig. 7.
When the system is used, the system is deployed on an actual production line, the image noise of a detected piece acquired by a camera is reduced by the system in a preprocessing mode in S1, then the image noise is input into a model for defect detection, and finally the output of defect types, positions and areas is realized.
In the test process, a system only using a yolov7-seg model is adopted for comparison test, the system operates for about 1 month, the detection quantity of products exceeds 10000, about 300 defective products are detected, the detection rate of the defective products is 100% through on-site manual rechecking, and the false detection quantity accounts for 0.5% of the total detection quantity. Comparing the outline detection based on vision of synchronous test, the defect detection rate is 90%, and the false detection quantity is about 20% of the total detection quantity; make onlyAnd the defect detection rate of a system using the yolov7-seg model is 97%, and the false detection quantity accounts for about 5% of the total detection quantity. At the time of sampling test, about 10.5mm for the actual area 2 The area calculated by the minimum bounding rectangle method is about 12mm 2 While the defect area calculated by the area calculation method M1 is 11mm 2 Closer to the actual area.
While particular embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations may be made therein without departing from the principles and spirit of the invention, the scope of which is defined in the appended claims and their equivalents.
Claims (6)
1. A method for detecting micro defects by combining example segmentation and secondary classification is characterized by comprising the following steps:
s1, acquiring a defect image, performing image acquisition on a detected piece, preprocessing the acquired defect image to obtain first original data, and acquiring an image without defects to be used as a first original background image;
s2, labeling the images, namely respectively labeling the first original data according to categories, storing the first original data as json format labels containing image information, expanding the original data and the corresponding labels, and combining the first original background images to generate a first micro defect data set;
s3, pre-training, namely dividing 10% of the first micro defect data set into a first pre-training data set, training a yolov7-seg deep learning network model to obtain a first initial detection model, and performing migration learning and training on the first initial detection model to obtain an optimal detection model;
s4, the first image acquisition program A1 cuts an original image in the tiny defect data set through the acquired defect position information to obtain a 224 x 224 size image containing a defect target, combines the acquired image, makes a second defect classification data set based on the acquired defect labeling information, trains a rapid classification model EfficientNet, and obtains an optimal EfficientNet model;
s5, packaging an optimal detection model, packaging an image acquisition program A1 and the optimal Efficient Net model into the optimal detection model, finishing improvement of the optimal detection model, outputting and decoding the Efficient Net model, obtaining classification information and confidence coefficient, carrying out logic operation B1 on the classification information and the decoded classification information of the optimal detection model, obtaining final classification information, combining the output information of the optimal detection model obtained by decoding, finishing output of all detection results, and finishing construction of a micro defect detection system combining example segmentation and secondary classification;
s6, deploying the system on an actual production line, collecting an image of a tested piece through a camera, reducing image noise in a preprocessing mode in S1, then inputting the image noise into the system for defect detection, and finally realizing the output of defect types, positions and areas.
2. The method for detecting a minute defect by combining instance segmentation with secondary classification according to claim 1,
in step S1, when an image of a measured object is acquired, information of a micro defect which is located in the measured object and cannot be directly observed and a non-defect target containing similar defect characteristics is acquired in a light source transmission mode, the information is used as first original data after mean filtering noise reduction and adaptive gamma correction and equalization image brightness preprocessing, the acquired first original data is screened, and the information is classified according to defect sizes and characteristics.
3. The method for detecting a minute defect by combining an example segmentation with a secondary classification according to claim 1 or 2, wherein,
in step S2, the original defect image and the corresponding label are expanded by adopting data expansion and adopting rotation, scaling, mirroring and brightness adjustment modes, and a first micro defect data set is generated by combining the first original background image, wherein the data set size is 640 x 640, and the label is in txt format.
4. The method for detecting a minute defect by combining instance segmentation with secondary classification according to claim 3,
the first original data are 5000 pieces in total, original defect images and corresponding labels are expanded by adopting the modes of random clockwise rotation and anticlockwise rotation by 0-15 degrees, random scaling by 0.85-1.15 times, mirroring and random brightness adjustment by 0.8-1.1 times, and each mode is applied to each image and label with 50% probability and is expanded to 4 times.
5. The method for detecting a minute defect by combining instance segmentation with secondary classification according to claim 3,
in step S3, after the first initial detection model is obtained, the first initial detection model is used as a basic model to be loaded, a transfer learning technology is adopted, data except a training data set is used for training, the performance of the trained model is verified, and an optimal detection model is selected.
6. The method for detecting a micro defect by combining instance segmentation with secondary classification according to claim 4,
when the optimal detection model is packaged, model output is called, and the marking category, the marking frame and the mask are obtained through decoding, so that the synchronous output of defect category, size, position and area information is realized.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311280842.2A CN117011300B (en) | 2023-10-07 | 2023-10-07 | Micro defect detection method combining instance segmentation and secondary classification |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311280842.2A CN117011300B (en) | 2023-10-07 | 2023-10-07 | Micro defect detection method combining instance segmentation and secondary classification |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117011300A CN117011300A (en) | 2023-11-07 |
| CN117011300B true CN117011300B (en) | 2023-12-12 |
Family
ID=88567552
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311280842.2A Active CN117011300B (en) | 2023-10-07 | 2023-10-07 | Micro defect detection method combining instance segmentation and secondary classification |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117011300B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117409261B (en) * | 2023-12-14 | 2024-02-20 | 成都数之联科技股份有限公司 | Element angle classification method and system based on classification model |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2022236876A1 (en) * | 2021-05-14 | 2022-11-17 | 广州广电运通金融电子股份有限公司 | Cellophane defect recognition method, system and apparatus, and storage medium |
| CN115471727A (en) * | 2022-08-24 | 2022-12-13 | 安徽工程大学 | Composite material defect detection method based on transfer learning |
-
2023
- 2023-10-07 CN CN202311280842.2A patent/CN117011300B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2022236876A1 (en) * | 2021-05-14 | 2022-11-17 | 广州广电运通金融电子股份有限公司 | Cellophane defect recognition method, system and apparatus, and storage medium |
| CN115471727A (en) * | 2022-08-24 | 2022-12-13 | 安徽工程大学 | Composite material defect detection method based on transfer learning |
Non-Patent Citations (1)
| Title |
|---|
| 基于Mask R-CNN的铸件X射线DR图像缺陷检测研究;蔡彪;沈宽;付金磊;张理泽;;仪器仪表学报(03);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117011300A (en) | 2023-11-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP6879431B2 (en) | Image processing equipment, image processing method and image processing program | |
| CN107123111B (en) | Deep residual error network construction method for mobile phone screen defect detection | |
| CN112348787B (en) | Training method of object defect detection model, object defect detection method and device | |
| US20180253836A1 (en) | Method for automated detection of defects in cast wheel products | |
| CN111932511B (en) | Electronic component quality detection method and system based on deep learning | |
| CN117011300B (en) | Micro defect detection method combining instance segmentation and secondary classification | |
| WO2021261168A1 (en) | Inspection device, unit selection device, inspection method, and inspection program | |
| CN112033965A (en) | 3D arc surface defect detection method based on differential image analysis | |
| CN110796669A (en) | Vertical frame positioning method and equipment | |
| CN110189375A (en) | A kind of images steganalysis method based on monocular vision measurement | |
| CN117974601B (en) | Method and system for detecting surface defects of silicon wafer based on template matching | |
| CN112465746A (en) | Method for detecting small defects in radiographic film | |
| CN114719749A (en) | Metal surface crack detection and real size measurement method and system based on machine vision | |
| CN109685030A (en) | A kind of mug rim of a cup defects detection classification method based on convolutional neural networks | |
| CN117351314A (en) | Glass bottle defect identification method and system | |
| CN113506281B (en) | Bridge crack detection method based on deep learning framework | |
| CN113343856B (en) | Image recognition method and system | |
| CN112802022B (en) | Method for intelligently detecting defective glass image, electronic device and storage medium | |
| CN113256563A (en) | Method and system for detecting surface defects of fine product tank based on space attention mechanism | |
| CN117853778A (en) | Improved HTC casting DR image defect identification method | |
| CN112488986A (en) | Cloth surface flaw identification method, device and system based on Yolo convolutional neural network | |
| JP2005283197A (en) | Detecting method and system for streak defect of screen | |
| CN114486916A (en) | Mobile phone glass cover plate defect detection method based on machine vision | |
| CN115272310B (en) | Method and device for detecting defects of workpiece | |
| CN117541579B (en) | Method and system for detecting defects of packaging support |
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 |