CN111696077A - Wafer defect detection method based on wafer Det network - Google Patents
Wafer defect detection method based on wafer Det network Download PDFInfo
- Publication number
- CN111696077A CN111696077A CN202010391589.8A CN202010391589A CN111696077A CN 111696077 A CN111696077 A CN 111696077A CN 202010391589 A CN202010391589 A CN 202010391589A CN 111696077 A CN111696077 A CN 111696077A
- Authority
- CN
- China
- Prior art keywords
- wafer
- network
- defect
- detection method
- wafer 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
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
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
- G06T2207/30148—Semiconductor; IC; Wafer
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Quality & Reliability (AREA)
- Testing Or Measuring Of Semiconductors Or The Like (AREA)
- Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)
Abstract
The invention discloses a wafer Det network method applied to wafer defect detection. The wafer appearance defect data set is established, then the wafer defect characteristics are extracted through the EfficientNet neural network, the bidirectional FPN extracted aiming at the wafer defect characteristics is designed based on the characteristic pyramid (FPN), the multi-scale characteristics are further extracted, the region of interest is generated on the extracted characteristic map through the RPN network, finally, the defect category of the region of interest is identified through a classification layer, and the position of a defect marking frame is directly determined on the characteristic map through a regression layer. The improved whole network is named as 'WaferDet'. The invention realizes automatic wafer defect detection, and greatly improves the detection efficiency and the detection precision compared with the traditional method.
Description
Technical Field
The invention relates to the technical field of semiconductor defect detection and image processing, in particular to a wafer defect detection method based on a wafer Det network.
Background
As chips play an increasingly critical role in the development of the national high-tech industry, wafers are also of great importance as their raw materials. However, at present, because the difference between the manufacturing technology of the wafer in China and the advanced foreign manufacturers is large, and the feedback of related cooperative enterprises exists, in the production process of the wafer, more unqualified wafers with the defects of scratch, corrosion, edge defect and fracture exist. If these wafers with defects are not detected by the upstream manufacturer, the quality of the downstream customer products is difficult to be guaranteed, and the reputation and image of the wafer manufacturer are also damaged to a great extent. Therefore, it is a research goal of many research institutes to search for a wafer appearance defect detection method with high performance and high efficiency.
The related technical research of wafer appearance defect detection at home and abroad can be roughly divided into three categories. Firstly, the traditional artificial naked eye detection method is adopted. The detection effect of the method is completely dependent on the technical level of an inspector, and the fatigue phenomenon exists in the naked eyes, so that the false detection rate is increased; secondly, a detection method based on laser measurement. The main principle is that X-rays generated in an X-ray inspection machine are adopted to irradiate the surface of a wafer to penetrate through the wafer for imaging, so that the internal structure fracture condition of the wafer is clearly displayed. However, the method has expensive equipment cost and needs special personnel for maintenance; and finally, a machine vision-based detection method. The detection method has the advantages of high speed, high precision, high flexibility, non-contact property and the like. Research institutes such as northeast university, Shanghai Baogang company, Wuhan science and technology university have researched detection systems for identifying the surface defects of the cold-rolled steel sheets by adopting a machine vision method; chongqing university has studied surface defects of high temperature continuous casting slabs. In the field of package printing, a whole set of defect detection system is established by cognex in the United states, and electronic products and food and beverage can be detected at the same time; the JLIVISION company defect detection system based on machine vision is suitable for glass, plastic, steel and packaging products; in China. The document 'edible oil filling quality detection system based on deep learning' provides a method for detecting the quality of an edible oil tank based on Fast RCNN neural network, the detection rate is more than 99%, the detection speed is less than 100ms, and good effect is achieved; the document transparent plastic part defect detection based on machine vision carries out defect detection on the transparent plastic part based on the Faster RCNN neural network, wherein the accuracy rate is 85%, the recall rate is 90%, and the detection speed is 101 ms. However, in the field of wafer appearance defect detection, due to the confidential measures of manufacturers on wafer data, the wafer appearance defect detection lacks a representative data set, so that research and application of a related deep learning algorithm in the field are limited.
Disclosure of Invention
The invention aims to provide a wafer defect detection method based on a wafer Det network, which realizes automatic detection of wafer defects and improves the wafer defect detection efficiency and detection precision compared with the traditional method. The specific technical scheme is as follows:
a wafer Det network method applied to wafer defect detection comprises the following steps:
(1) inputting marked wafer defect image data (the size of the image data is 640x640, and the wafer appearance acquisition equipment is shown in FIG. 1);
(2) inputting the defect image of the whole wafer into an EfficientNet feature extraction network (the structure of the EfficientNet feature extraction network is shown on the left side of the figure 3), and extracting features;
(3) performing feature downsampling on EfficientNet layers P3, P4, P5, P6 and P7, constructing a bidirectional feature pyramid by using a sampled feature map, and fusing features;
(4) generating suggestion windows on each output layer of the characteristic pyramid by using an RPN (resilient packet network), determining an interested area, and generating 1000 suggestion windows for each picture;
(5) generating a wafer characteristic map with a fixed size for each RoI through a RoI pooling layer;
(6) training the network by utilizing a Softmax Loss function (wafer defect classification Loss function) and a Smooth L1 Loss function (wafer defect border regression Loss function);
(7) and detecting the wafer picture which is not marked by the trained network model, and detecting whether the wafer picture has defects, defect types and positions.
The method steps (3) (4) (5) are based on an improvement or variation of the conventional FPN: and build the enhancement of the bottom-up path by using the idea of PANET; removing only one input node, and directly transmitting the information of the input node of the deleted node to the output node of the deleted node; introducing a skip link idea of a ResNet residual module, and transmitting the characteristics of a shallow network to a deep layer;
the momentum parameter of SGD of the method is 0.9, and the weight attenuation parameter is 4 e-5; when the bidirectional RPN is trained, for each input image, 256 training sets are extracted, wherein positive example overlap (IoU) is greater than 0.35, negative example overlap is less than 0.001, and the positive example proportion is not more than 20%; during the training of ResNet-101, the RPN is set to provide 1000 RoIs, 256 pieces of RoIs are randomly selected from the 1000 RoIs as a training set, wherein positive case overlap (IoU) is greater than 0.3, negative case overlap is less than 0.001, the positive case proportion is not more than 5%, and the learning rate lr =0.00005 is set.
Drawings
FIG. 1a is a diagram of the FPN architecture;
FIG. 1b is a diagram of a bi-directional FPN architecture;
FIG. 2 is a schematic diagram of a wafer defect detection network framework;
fig. 3 is a graph showing results of the WaferDet assay.
Detailed Description
The technical scheme of the invention is further explained by combining the drawings in the specification.
1. Software and hardware Environment for experiments
The computer hardware and software configuration adopts a Ubuntu 64-bit system, and the processor is i 78700 k, the memory 24G and the display card 1070 ti. The software environment is PyTorch 1.0 and imgauge data enhancement library.
2. Image acquisition of wafers
The method adopts 2600 ten thousand HDMI industrial cameras to collect images, wherein a 90-270V wide voltage adjustable light source is used for polishing wafers. And globally scanning through a macro fixed-focus lens to obtain a rectangular wafer original color image. The size of the photosensitive chip of the camera is 2/3' (8.8mm 6.6mm), the distance precision between each pixel can reach 3.6um, the detection precision of 50um is realized for lmm wafers, and the precision requirement of actual production defect detection can be completely met. The acquisition device is configured in detail as: 2600 million HDMI industrial camera, QJY-508 mirror, QJY-508 lens support, 90-270V wide voltage camera power supply, 90-270V wide voltage adjustable light source, camera measuring software CD and correcting ruler.
3. Wafer appearance defect detection method selection
And detecting appearance defects of the wafer by adopting a deep learning method.
There are three main levels of understanding of wafer defect images: and (4) classifying, detecting and segmenting. For the detection task of the appearance defects of the wafer, the appearance defects needing to be detected mainly comprise glass depressions, chip edge breakage, glass on a welding surface, crystal grain breakage, leftover materials and residual oxide films on the surface. If the defect type is regarded as a classification task, the defect types are more, the background interference among the types is larger, and the identification effect cannot be ensured by a general training classifier. The final aim of the method is to eliminate the defective wafer and not to carry out pixel-level segmentation on the wafer defects. Therefore, in order to detect the type and position of the wafer defect, it is the best solution to mark the defect by using the bounding box.
4. Wafer Det wafer appearance defect detection algorithm implementation
Firstly, wafer appearance defect characteristics are extracted by adopting EfficientNet as a basic network.
The EfficientNet is discovered by Google by adopting an architecture search technology, and the method obtains good expression on the accuracy and efficiency of feature extraction. On the ImageNet image dataset, EfficientNet-B7 surpassed the accuracy of the best previous GPipe model, with 88% fewer parameters and 6.1 times faster. Compared with the widely used ResNet-50 network, the Efficientnets-B4 improves the top-1 precision from 76.3% to 82.6% under similar FLOPS. In addition to the ImageNet dataset, Efficientnets performed well on other datasets, 5 out of 8 widely used datasets achieved the most advanced precision and reduced by 95% the parameter amount compared to the existing convolutional network.
The extracted features are then input into a bi-directional feature pyramid network, which is refined from the original feature pyramid network.
Conventional FPN as shown in fig. 1a, top-level features are fused with lower-level features in a linear manner with twice weighting, however, the PANet study finds that lower-level features contribute to the identification of large amounts of data. But it is increasingly difficult to access accurately located information from the underlying structure to the topmost features. Thus, the present invention takes advantage of the idea of PANET to create a bottom-up path enhancement, as shown in FIG. 1 b. In addition, in order to reduce the amount of calculation, only one input node is removed, a top-down propagation mode shown in fig. 1b is constructed, and meanwhile, the jump connection idea of a ResNet residual module is introduced, so that shallow features are transmitted to deep layers for learning. Finally, a bi-directional FPN structure as shown in FIG. 1b is constructed.
And finally, generating an interested area on the extracted feature map by adopting an RPN network, wherein a classification layer identifies the defect category of the interested area, and a regression layer directly determines the position of a defect marking frame on the feature map. The improved overall network is named as 'wafer det', and the schematic diagram of the network structure is shown in fig. 2.
The algorithm comprises the following specific steps:
(1) inputting the data of the marked wafer defect image;
(2) inputting the defect image of the whole wafer into an EfficientNet feature extraction network for feature extraction;
(3) performing feature downsampling on EfficientNet layers P3, P4, P5, P6 and P7, constructing a bidirectional feature pyramid by using a sampled feature map, and fusing features;
(4) generating suggestion windows on each output layer of the characteristic pyramid by using an RPN (resilient packet network), determining an interested area, and generating 1000 suggestion windows for each picture;
(5) generating a wafer characteristic map with a fixed size for each RoI through a RoI pooling layer;
(6) training the network by utilizing a Softmax Loss function (wafer defect classification Loss function) and a Smooth L1 Loss function (wafer defect border regression Loss function);
(7) and detecting the wafer picture which is not marked by the trained network model, and detecting whether the wafer picture has defects, defect types and positions.
5. Wafer Det wafer appearance defect detection algorithm training
And taking a 640x640 picture as input, training by an SGD method, setting the momentum to be 0.9, and setting the weight attenuation to be 4 e-5. In the training of bidirectional RPN, for each input image, 256 images are randomly extracted as a training set, wherein positive example overlap (IoU) is greater than 0.35, negative example overlap is less than 0.001, and positive example proportion is not more than 20%. The classifier uses a softmax loss function. When the EfficientNet is trained, the number of RoIs provided by the bidirectional RPN is controlled to be 1000, and 256 pieces of RoIs are randomly selected from the 1000 pieces of RoIs to serve as a training set, wherein positive example overlap (IoU) is greater than 0.3, negative example overlap is less than 0.001, the positive example proportion is not more than 5%, and the learning rate lr =0.005 is set. The classifier uses the softmax function, while the boundary regression uses the Smooth L1 Loss function.
6. Test results and analysis
The average time of each graph of the trained WaferDet in 1070Ti GPU calculation mode is 87ms, and the parameter number is 50M. The obtained defect detection results are shown in table 1:
TABLE 1 statistical table of WaferDet test results
| Number of | Correction of | Error(s) in |
| With defects (positive type) | 2282 | 46 |
| Flawless (negative type) | 756 | 35 |
Accuracy FP =2282/(2282+46) =98.0%
Recall FR =2282/(2282+ 35) =98.5%
F1 score=2/(l/P+l/R)=98.25%
Wherein the detection result mAP = 0.835. The partial detection results are shown in FIG. 3.
According to the calculation analysis, the wafer Det network obtains the detection accuracy with the average accuracy of 98% under the condition that only 50M model parameters exist, and meanwhile, the model reasoning speed reaches 87 ms/piece.
7. Algorithmic comparative analysis
In order to prove that the algorithm has better performance in speed and precision than the current mainstream detection algorithm, the algorithm is compared with two mainstream detection algorithms, namely fast RCNN and FPN. Training two detection algorithms of fast RCNN and FPN by using the same data set of the training waferDet, and testing the fast RCNN and FPN algorithms by using the test set of the testing waferDet, wherein the test results are as follows:
the average time of each graph in the trained fast RCNN detection model 1070Ti GPU calculation mode is 210ms, and the parameter number is 135M. . The obtained defect detection results are shown in table 2:
TABLE 2 statistics of fast RCNN test results
| Number of | Correction of | Error(s) in |
| With defects (positive type) | 2090 | 180 |
| Flawless (negative type) | 785 | 145 |
Calculating to obtain:
precision FP =2090/(2090+180) =92.1%
Recall FR =2090/(2090+ 145) =93.5%
F1score=2/(l/P+l/R)=92.8%
Wherein the detection result mAP = 0.78.
The trained FPN detection model 1070Ti GPU takes 248ms on average for each graph in a calculation mode, and the parameter number is 187M. The obtained defect detection results are shown in table 3:
TABLE 3 statistical table of FPN test results
| Number of | Correction of | Error(s) in |
| With defects (positive type) | 2162 | 110 |
| Flawless (negative type) | 822 | 105 |
Calculating to obtain:
the precision rate FP =2162/(2162+110) =95.2%
Recall FR =2162/(2162+ 105) =95.4%
F1score=2/(l/P+l/R)=95.3%
Wherein the detection result mAP = 0.82.
As can be seen from comparative analysis, the three algorithms are superior to the Faster RCNN and FPN detection algorithms in the calculation precision and the reasoning speed. In particular, the WaferDet algorithm has a parameter of only 50M under the same conditions, but achieves the highest maps of 0.835. Compared with the current target detection method using two mainstream algorithms of fast RCNN and FPN, the method has great advantages.
Claims (5)
1. A wafer defect detection method based on a wafer Det network is characterized by comprising the following steps:
(1) inputting the data of the marked wafer defect image;
(2) inputting the defect image of the whole wafer into an EfficientNet feature extraction network for feature extraction;
(3) performing feature downsampling on EfficientNet layers P3, P4, P5, P6 and P7, constructing a bidirectional feature pyramid by using a sampled feature map, and fusing features;
(4) generating suggestion windows on each output layer of the characteristic pyramid by using an RPN (resilient packet network), determining an interested area, and generating 1000 suggestion windows for each picture;
(5) generating a wafer characteristic map with a fixed size for each RoI through a RoI pooling layer;
(6) training the network by utilizing a wafer defect classification Loss function (Softmax Loss) and a wafer defect border regression Loss function (Smooth L1 Loss);
(7) and detecting the wafer picture which is not marked by the trained network model, and detecting whether the wafer picture has defects, defect types and positions.
2. The wafer defect detection method based on the wafer det network as claimed in claim 1, wherein: the method is based on an improvement or variation of the traditional FPN framework: and build the enhancement of the bottom-up path by using the idea of PANET; removing only one input node, and directly transmitting the information of the input node of the deleted node to the output node of the deleted node; and introducing a hop link idea of a ResNet residual module to transfer the characteristics of the shallow network to the deep layer.
3. The wafer defect detection method based on the wafer det network as claimed in claim 1, wherein: the feature extraction adopts EfficientNet as a basic network to extract the appearance defect features of the wafer.
4. The wafer defect detection method based on the wafer det network as claimed in claim 1, wherein: the detection method adopts a wafer defect classification Loss function (Softmax Loss) and a wafer defect border regression Loss function (smoothen L1 Loss) to carry out combined training on the wafer defect classification probability and the border regression.
5. The wafer defect detection method based on the wafer det network as claimed in claim 1, wherein: in the training: the momentum parameter of the SGD is 0.9, and the weight attenuation parameter is 4 e-5; when the bidirectional RPN is trained, extracting 256 input images as a training set, wherein positive example overlap is >0.35, negative example overlap is <0.001, and the positive example proportion is not more than 20%; during training of ResNet-101, the RPN is set to provide 1000 RoIs, 256 pieces of RoIs are randomly selected from the 1000 RoIs as a training set, wherein positive case overlap is >0.3, negative case overlap is <0.001, the positive case proportion is not more than 5%, and the learning rate lr =0.00005 is set.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010391589.8A CN111696077A (en) | 2020-05-11 | 2020-05-11 | Wafer defect detection method based on wafer Det network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010391589.8A CN111696077A (en) | 2020-05-11 | 2020-05-11 | Wafer defect detection method based on wafer Det network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111696077A true CN111696077A (en) | 2020-09-22 |
Family
ID=72477108
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010391589.8A Pending CN111696077A (en) | 2020-05-11 | 2020-05-11 | Wafer defect detection method based on wafer Det network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111696077A (en) |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112180903A (en) * | 2020-10-19 | 2021-01-05 | 江苏中讯通物联网技术有限公司 | Vehicle state real-time detection system based on edge calculation |
| CN112487862A (en) * | 2020-10-28 | 2021-03-12 | 南京云牛智能科技有限公司 | Garage pedestrian detection method based on improved EfficientDet model |
| CN112528810A (en) * | 2020-12-04 | 2021-03-19 | 北京中科慧眼科技有限公司 | Semantic segmentation method, system and equipment suitable for mobile terminal deployment |
| CN112733686A (en) * | 2020-12-31 | 2021-04-30 | 武汉兴图新科电子股份有限公司 | Target object identification method and device used in image of cloud federation |
| CN112819748A (en) * | 2020-12-16 | 2021-05-18 | 机科发展科技股份有限公司 | Training method and device for strip steel surface defect recognition model |
| CN112884769A (en) * | 2021-04-12 | 2021-06-01 | 深圳中科飞测科技股份有限公司 | Image processing method, image processing apparatus, optical system, and computer-readable storage medium |
| CN112884744A (en) * | 2021-02-22 | 2021-06-01 | 深圳中科飞测科技股份有限公司 | Detection method and device, detection equipment and storage medium |
| CN112991267A (en) * | 2021-02-08 | 2021-06-18 | 上海工程技术大学 | Strip steel surface defect detection method based on improved efficientNet-RCNN |
| CN113222967A (en) * | 2021-05-28 | 2021-08-06 | 长江存储科技有限责任公司 | Wafer detection method and system |
| CN113313668A (en) * | 2021-04-19 | 2021-08-27 | 石家庄铁道大学 | Subway tunnel surface disease feature extraction method |
| WO2022105143A1 (en) * | 2020-11-18 | 2022-05-27 | 淮阴工学院 | Lightweight fire-det flame detection method and system |
| CN114998324A (en) * | 2022-07-19 | 2022-09-02 | 深圳新视智科技术有限公司 | Training method and device for semiconductor wafer defect detection model |
| CN115035082A (en) * | 2022-06-24 | 2022-09-09 | 西安电子科技大学芜湖研究院 | YOLOv4 improved algorithm-based aircraft transparency defect detection method |
| KR102516199B1 (en) * | 2021-12-22 | 2023-03-30 | 호서대학교 산학협력단 | Apparatus and method for vision inspection using artificial neural network including filter for feature detection |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108765389A (en) * | 2018-05-18 | 2018-11-06 | 浙江大学 | A kind of microcosmic wafer surface defects image detecting method |
| CN110555842A (en) * | 2019-09-10 | 2019-12-10 | 太原科技大学 | Silicon wafer image defect detection method based on anchor point set optimization |
| CN110853015A (en) * | 2019-11-12 | 2020-02-28 | 中国计量大学 | Aluminum profile defect detection method based on improved Faster-RCNN |
-
2020
- 2020-05-11 CN CN202010391589.8A patent/CN111696077A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108765389A (en) * | 2018-05-18 | 2018-11-06 | 浙江大学 | A kind of microcosmic wafer surface defects image detecting method |
| CN110555842A (en) * | 2019-09-10 | 2019-12-10 | 太原科技大学 | Silicon wafer image defect detection method based on anchor point set optimization |
| CN110853015A (en) * | 2019-11-12 | 2020-02-28 | 中国计量大学 | Aluminum profile defect detection method based on improved Faster-RCNN |
Non-Patent Citations (4)
| Title |
|---|
| DONGLIANG ZHANG 等: "Design Patterns of Soft Products Using Surface Flattening", 《COMPUTING AND INFORMATION SCIENCE IN ENGINEERING》 * |
| MINGXING TAN 等: "EfficientDet: Scalable and Efficient Object Detection", 《HTTPS://ARXIV.ORG/ABS/1911.09070V4》 * |
| ZHU QIANQIAN 等: "Research on Vehicle Appearance Component Recognition Based on Mask R-CNN", 《PHYSICS: CONFERENCE SERIES》 * |
| 邡鑫 等: "基于卷积神经网络的晶圆缺陷检测与分类算法", 《计算机工程》 * |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112180903A (en) * | 2020-10-19 | 2021-01-05 | 江苏中讯通物联网技术有限公司 | Vehicle state real-time detection system based on edge calculation |
| CN112487862B (en) * | 2020-10-28 | 2021-07-02 | 南京云牛智能科技有限公司 | Garage pedestrian detection method based on improved EfficientDet model |
| CN112487862A (en) * | 2020-10-28 | 2021-03-12 | 南京云牛智能科技有限公司 | Garage pedestrian detection method based on improved EfficientDet model |
| WO2022105143A1 (en) * | 2020-11-18 | 2022-05-27 | 淮阴工学院 | Lightweight fire-det flame detection method and system |
| CN112528810A (en) * | 2020-12-04 | 2021-03-19 | 北京中科慧眼科技有限公司 | Semantic segmentation method, system and equipment suitable for mobile terminal deployment |
| CN112819748B (en) * | 2020-12-16 | 2023-09-19 | 机科发展科技股份有限公司 | Training method and device for strip steel surface defect recognition model |
| CN112819748A (en) * | 2020-12-16 | 2021-05-18 | 机科发展科技股份有限公司 | Training method and device for strip steel surface defect recognition model |
| CN112733686A (en) * | 2020-12-31 | 2021-04-30 | 武汉兴图新科电子股份有限公司 | Target object identification method and device used in image of cloud federation |
| CN112991267A (en) * | 2021-02-08 | 2021-06-18 | 上海工程技术大学 | Strip steel surface defect detection method based on improved efficientNet-RCNN |
| CN112884744A (en) * | 2021-02-22 | 2021-06-01 | 深圳中科飞测科技股份有限公司 | Detection method and device, detection equipment and storage medium |
| CN112884769A (en) * | 2021-04-12 | 2021-06-01 | 深圳中科飞测科技股份有限公司 | Image processing method, image processing apparatus, optical system, and computer-readable storage medium |
| CN112884769B (en) * | 2021-04-12 | 2021-09-28 | 深圳中科飞测科技股份有限公司 | Image processing method, image processing apparatus, optical system, and computer-readable storage medium |
| CN113313668A (en) * | 2021-04-19 | 2021-08-27 | 石家庄铁道大学 | Subway tunnel surface disease feature extraction method |
| CN113222967A (en) * | 2021-05-28 | 2021-08-06 | 长江存储科技有限责任公司 | Wafer detection method and system |
| KR102516199B1 (en) * | 2021-12-22 | 2023-03-30 | 호서대학교 산학협력단 | Apparatus and method for vision inspection using artificial neural network including filter for feature detection |
| CN115035082A (en) * | 2022-06-24 | 2022-09-09 | 西安电子科技大学芜湖研究院 | YOLOv4 improved algorithm-based aircraft transparency defect detection method |
| CN115035082B (en) * | 2022-06-24 | 2024-03-29 | 西安电子科技大学芜湖研究院 | Method for detecting defects of transparent parts of aircraft based on YOLOv4 improved algorithm |
| CN114998324A (en) * | 2022-07-19 | 2022-09-02 | 深圳新视智科技术有限公司 | Training method and device for semiconductor wafer defect detection model |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111696077A (en) | Wafer defect detection method based on wafer Det network | |
| US20220101517A1 (en) | Substrate defect inspection apparatus, substrate defect inspection method, and storage medium | |
| EP4280153A1 (en) | Defect detection method, apparatus and system | |
| CN113570605B (en) | Defect detection method and system based on liquid crystal display panel | |
| CN107966454A (en) | A kind of end plug defect detecting device and detection method based on FPGA | |
| US9171364B2 (en) | Wafer inspection using free-form care areas | |
| CN110490842B (en) | Strip steel surface defect detection method based on deep learning | |
| WO2021168733A1 (en) | Defect detection method and apparatus for defect image, and computer-readable storage medium | |
| CN109255787A (en) | Silk ingot scratch detection system and method based on deep learning and image processing techniques | |
| CN107358599A (en) | Use the defects of specific infomation detection chip of defect | |
| CN107084991A (en) | The detection of quartz pushrod bubble and quality grading method based on machine vision | |
| CN116559183B (en) | Method and system for improving defect judging efficiency | |
| CN112700442A (en) | Die-cutting machine workpiece defect detection method and system based on Faster R-CNN | |
| CN116359233B (en) | Square battery appearance defect detection method and device, storage medium and electronic equipment | |
| KR101867015B1 (en) | Device and method for inspecting defect of glass and inspection system | |
| CN110426395B (en) | Method and device for detecting surface of solar EL battery silicon wafer | |
| CN117392112A (en) | Method, device and system for detecting wafer surface defects | |
| CN109387527A (en) | On a kind of film in film defect detection method | |
| CN114612418A (en) | Method, device and system for detecting surface defects of mouse shell and electronic equipment | |
| CN113808104A (en) | Block-based metal surface defect detection method and system | |
| CN117115171A (en) | Slight bright point defect detection method applied to subway LCD display screen | |
| CN116051541B (en) | Bearing end face gentle abrasion detection method and device based on stroboscopic light source | |
| Wang | RETRACTED ARTICLE: Application of deep learning to detect defects on the surface of steel balls in an IoT environment | |
| CN110874837B (en) | Defect automatic detection method based on local feature distribution | |
| CN114092396A (en) | Method and device for detecting corner collision flaw of packaging box |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200922 |