CN111415329B - Workpiece surface defect detection method based on deep learning - Google Patents
Workpiece surface defect detection method based on deep learning Download PDFInfo
- Publication number
- CN111415329B CN111415329B CN202010105860.7A CN202010105860A CN111415329B CN 111415329 B CN111415329 B CN 111415329B CN 202010105860 A CN202010105860 A CN 202010105860A CN 111415329 B CN111415329 B CN 111415329B
- Authority
- CN
- China
- Prior art keywords
- workpiece
- convolution
- feature
- prediction
- size
- 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
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
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Quality & Reliability (AREA)
- Probability & Statistics with Applications (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a workpiece surface defect detection method based on deep learning, which specifically comprises the following steps: collecting workpiece images under different backgrounds and illumination conditions; preprocessing the collected workpiece image; constructing a deep convolutional neural network model to obtain feature maps of 6 different layers; carrying out multi-scale feature fusion prediction by adopting a feature pyramid feature map, obtaining a boundary frame for generating 4 anchor box prediction targets by using a K-means clustering algorithm, and predicting categories by using a cross entropy loss function; removing redundant prediction boundary boxes through a non-maximum suppression algorithm; and outputting the position information and the category of the surface defects of the workpiece. The invention solves the problems of low detection efficiency and poor precision of manual detection and physical detection methods, overcomes the problem of poor adaptability of traditional machine vision defect detection, improves the detection efficiency and accuracy of workpiece surface defects, reduces labor cost, can rapidly adapt to the surface defect detection of novel products, shortens development period and improves flexibility.
Description
Technical Field
The invention relates to the technical field of image processing, in particular to a workpiece surface defect detection method based on deep learning.
Background
The workpiece is a component part of a precise instrument, and the surface quality and the shape structure of the workpiece have great influence on the performance of the instrument. In the process of processing, transporting and assembling the workpiece, the factors of change and abrasion of the cutter stroke, the self material characteristics of the workpiece, friction collision among the workpieces and the like can cause the defects of scratch, convex powder, bump, concave, crack and the like on the surface of the workpiece, so that the attractiveness of the workpiece is influenced, the service performance of the workpiece is also influenced, and even important potential safety hazards are brought to instruments. Therefore, development of a rapid and accurate workpiece surface defect detection method is needed, which has important practical application value for popularization of product quality detection technology.
Currently, the workpiece quality detection methods mainly comprise a manual detection method, an instrument-based physical detection method and a machine vision detection method. The manual detection method is to observe whether defects exist on the surface of the workpiece by manual visual observation, judge the category of the defects by using experience knowledge, and has strong subjective consciousness, time and labor consumption and difficult detection accuracy guarantee. The physical detection method based on the instrument comprises ultrasonic detection, magnetic particle detection, eddy current detection, X-ray detection and the like, and can meet the surface defect detection of common workpieces, but the instrument has larger volume, high price, complex maintenance, lower detection efficiency and poorer accuracy. Because the machine vision detection method has the advantages of no damage, accuracy, rapidness, reliability and the like, the machine vision technology is used for carrying out nondestructive detection on the surface defects of the workpiece, so that subjective differences and visual fatigue of people can be eliminated, the labor cost is reduced, the detection efficiency and accuracy are improved, and the detection grading error is reduced. Compared with the traditional machine vision method, the deep learning method can directly learn the features from the bottom data, has higher complex structure expression capability, completely replaces the manual design of the features by an automatic learning process, can be suitable for different products, shortens the development period, improves the flexibility, and is fast suitable for surface defect detection of novel products.
In summary, the prior art has the following disadvantages:
(1) The manual detection method has strong subjective consciousness, consumes time and labor, and is difficult to ensure detection precision.
(2) The physical detection method based on the instrument has the advantages of larger instrument and equipment volume, high price, complex maintenance, lower detection efficiency and poor accuracy.
(3) The traditional machine vision detection method needs to manually extract the characteristics, and has poor product adaptability.
Disclosure of Invention
In view of the above, in order to solve the above problems in the prior art, the present invention provides a method for detecting surface defects of a workpiece based on deep learning, which improves the detection efficiency and accuracy of the surface defects of the workpiece, reduces the labor cost, and can quickly adapt to the surface defect detection of a novel product, shortens the development period, and improves the flexibility.
The invention solves the problems by the following technical means:
a workpiece surface defect detection method based on deep learning comprises the following steps:
collecting workpiece images under different backgrounds and illumination conditions;
preprocessing the collected workpiece image;
constructing a deep convolutional neural network model to obtain feature maps of 6 different layers;
carrying out multi-scale feature fusion prediction by adopting a feature pyramid feature map, obtaining a boundary frame for generating 4 anchor box prediction targets by using a K-means clustering algorithm, and predicting categories by using a cross entropy loss function;
removing redundant prediction boundary boxes through a non-maximum suppression algorithm;
and outputting the position information and the category of the surface defects of the workpiece.
Further, the workpiece images under different backgrounds and illumination conditions are collected specifically as follows:
and shooting workpiece images under different backgrounds and illumination conditions by using a CCD camera, wherein the workpiece images comprise single background, complex background, dark, bright, soft light, shielding and overlapped complex environments.
Further, preprocessing the collected workpiece image specifically includes:
firstly, carrying out data enhancement on an obtained workpiece image, wherein the data enhancement comprises optical transformation and geometric transformation; the optical transformation comprises brightness random adjustment, contrast random adjustment and channel random adjustment, and the geometric transformation comprises rotation, stretching, translation, horizontal overturning, vertical overturning, random picture cutting and random scaling;
then, generating corresponding labeling information for the expanded image data set by using a labeling tool; the labeling information comprises position information and categories of surface defects of the workpiece in the sample, wherein the categories are normal, scratches, convex powder, bumps and cracks;
finally, dividing the data set into a training set and a testing set; wherein the training set is 70% of the total sample size, and the test set is 30%.
Further, constructing a deep convolutional neural network model to obtain characteristic diagrams of 6 different layers specifically comprises the following steps:
constructing a deep convolutional neural network, including constructing a basic network structure and a deep convolutional layer, and extracting high semantic features of an image; and in the deep convolutional neural network module, each convolutional layer operates by a ReLu function and normalization Batch Normalization;
(1) Basic network structure: VGG16 is adopted as a basic feature extraction module; firstly, through the first 13 convolution layers of the VGG16, and modifying, using dense conv+stride conv to replace dense conv+maxpooling in the VGG16 network structure, wherein the convolution kernel size of dense conv is 3 multiplied by 3, and the step size is 1; the convolution kernel size of stride conv is 3×3, the step size is 2; then, using two convolutions Conv6 and Conv7 to replace full connection layers FC6 and FC7 of VGG16 to further extract features, wherein Conv6 is cavity convolution, the size of convolution kernel is 3 multiplied by 3, the number of cavities is 6, and the step length is 1; conv7 has a convolution kernel size of 1×1 and a step size of 1;
(2) Depth convolution layer: adding 4 deep convolution layers on the basic network structure, and further extracting higher semantic information; the depth convolution layers are Conv8, conv9, conv10 and Conv11, respectively, each consisting of 2 convolutions: a convolution kernel of size 1 x 1, step size 1, and a convolution kernel of size 3 x 3, step size 2.
Further, the feature pyramid feature map is adopted to conduct multi-scale feature fusion prediction, a K-means clustering algorithm is used to obtain 4 anchor box prediction target bounding boxes, and cross entropy loss function prediction categories are used, specifically:
6 feature graphs, namely 63×63, 32×32, 16×16, 8×8, 4×4 and 1×1 are obtained by using a deep convolutional neural network, and a feature pyramid method is adopted to conduct multi-scale feature fusion prediction on the feature graphs of 6 different layers; 4 anchor box prediction target boundary boxes are obtained through K-means algorithm clustering, and cross entropy loss function prediction categories are used;
the FPN overall architecture comprises 4 parts, namely a top-down network, a bottom-up network, transverse connection and convolution fusion;
top-down network: is a deep convolutional neural network;
bottom-up network: carrying out convolution with the size of 1 multiplied by 1 on a characteristic diagram 1 multiplied by 1 obtained by Conv11 in a depth convolution layer to reduce the channel number to obtain F6, and then carrying out 2 times nearest neighbor up-sampling operation in sequence;
and (3) transverse connection: carrying out 1X 1 convolution operation on bottom feature graphs obtained by Conv9, conv8, conv7, conv5 and Conv4 in the depth convolution neural network respectively to fix the channel number of the bottom feature graphs to 256, and adding the bottom feature graphs with the up-sampled high semantic features element by element to obtain F5, F4, F3, F2 and F1;
convolution fusion: f5, F4, F3, F2, F1 were fused using a convolution of 3 x 3 size.
Further, redundant prediction bounding boxes are removed through a non-maximum suppression algorithm, specifically:
a modified algorithm Soft-NMS using a non-maximum suppression algorithm removes duplicate prediction bounding boxes;
the specific expression of the Soft-NMS function is as follows:
wherein s is i For each bounding box score, M is the bounding box with the highest current score, b i For some bounding box remaining, N t To set a threshold of 0.5.
Compared with the prior art, the invention has the beneficial effects that at least:
(1) The dense conv+stride conv structure is used for replacing dense conv+maxpooling in VGG16 to carry out downsampling, so that more characteristic information is reserved, the speed is higher, and the detection efficiency is effectively improved.
(2) The deep convolutional neural network is constructed to extract features, 4 deep convolutional layers are added on a network structure taking VGG16 as a base, FPN is adopted to conduct multi-scale feature fusion prediction, higher semantic information is extracted, the detection accuracy is improved, the generalization capability of a model is enhanced, and the flexibility is improved.
(3) The repeated prediction boundary boxes are removed by using a Soft-NMS algorithm, so that the problem of missed detection caused by overlapping of a plurality of workpieces can be solved, and the recall rate and the robustness of detection are improved.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly described below, and it is apparent that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is an overall flow chart of a deep learning based workpiece surface defect detection method of the present invention;
FIG. 2 is a detailed step diagram of the workpiece surface defect detection method based on deep learning of the present invention.
Detailed Description
In order to make the above objects, features and advantages of the present invention more comprehensible, the following detailed description of the technical solution of the present invention refers to the accompanying drawings and specific embodiments. It should be noted that the described embodiments are only some embodiments of the present invention, and not all embodiments, and that all other embodiments obtained by persons skilled in the art without making creative efforts based on the embodiments in the present invention are within the protection scope of the present invention.
Examples
As shown in fig. 1, the invention discloses a workpiece surface defect detection method based on deep learning, which specifically comprises the following steps: collecting workpiece images under different backgrounds and illumination conditions; preprocessing the collected workpiece image; constructing a deep convolutional neural network model to obtain feature maps of 6 different layers; carrying out multi-scale feature fusion prediction by adopting a feature pyramid feature map, obtaining a boundary frame for generating 4 anchor box prediction targets by using a K-means clustering algorithm, and predicting categories by using a cross entropy loss function; removing redundant prediction boundary boxes through a non-maximum suppression algorithm; and outputting the position information and the category of the surface defects of the workpiece.
As shown in fig. 2, the workpiece surface defect detection method based on deep learning of the invention specifically comprises the following steps:
step 1: acquisition of workpiece images
And shooting workpiece images under different backgrounds and illumination conditions by using a CCD camera, wherein the workpiece images comprise single background, complex background, dark, bright, soft light, shielding, overlapping and other complex environments.
Step 2: image preprocessing
First, data enhancement is performed on the workpiece image obtained in step 1. Data enhancement mainly includes optical transformations and geometric transformations. The optical transformation comprises brightness random adjustment, contrast random adjustment and channel random adjustment, and the geometric transformation comprises rotation, stretching, translation, horizontal overturning, vertical overturning, random picture cropping and random scaling.
And then, generating corresponding labeling information for the expanded image data set by using a labeling tool. The labeling information comprises position information and categories of surface defects of the workpiece in the sample, wherein the categories are normal, scratches, convex powder, bumps and pits and cracks.
Finally, the data set is divided into a training set and a testing set. Wherein the training set is 70% of the total sample size, and the test set is 30%.
Step 3: construction of deep convolutional neural network to extract features
The method comprises the steps of constructing a deep convolutional neural network, namely constructing a basic network structure and a deep convolutional layer, and extracting high semantic features of an image. And in the deep convolutional neural network module, each convolutional layer operates by a ReLu function and normalization Batch Normalization.
(1) Basic network structure: VGG16 is used as the base feature extraction module. Firstly, through the first 13 convolution layers of the VGG16, and modifying, using dense conv+stride conv to replace dense conv+maxpooling in the VGG16 network structure, wherein the convolution kernel size of dense conv is 3 multiplied by 3, and the step size is 1; the convolution kernel size of stride conv is 3×3, with a step size of 2. Then, using two convolutions Conv6 and Conv7 to replace full connection layers FC6 and FC7 of VGG16 to further extract features, wherein Conv6 is cavity convolution, the size of convolution kernel is 3 multiplied by 3, the number of cavities is 6, and the step length is 1; conv7 has a convolution kernel size of 1×1 and a step size of 1.
(2) Depth convolution layer: 4 deep convolution layers are added on the basic network structure, and higher semantic information is further extracted. The depth convolution layers are Conv8, conv9, conv10 and Conv11, respectively, each consisting of 2 convolutions: a convolution kernel of size 1 x 1, step size 1, and a convolution kernel of size 3 x 3, step size 2.
Step 4: multi-scale feature fusion prediction
6 feature graphs are obtained by utilizing a deep convolutional neural network, namely 63×63, 32×32, 16×16, 8×8, 4×4 and 1×1, and a feature pyramid (Feature Pyramid Network, FPN) method is adopted to conduct multi-scale feature fusion prediction on the feature graphs of 6 different layers; 4 anchor box prediction target bounding boxes are clustered through a K-means algorithm, and categories are predicted through a cross entropy loss function.
Wherein the FPN overall architecture includes 4 parts of top-down network, bottom-up network, cross connect and convolution fusion.
Top-down network: is a deep convolutional neural network
Bottom-up network: and (3) carrying out convolution with the size of 1 multiplied by 1 on the characteristic diagram 1 multiplied by 1 obtained by Conv11 in the depth convolution layer to reduce the channel number to obtain F6, and then carrying out 2 times nearest neighbor up-sampling operation in sequence.
And (3) transverse connection: and respectively carrying out 1X 1 convolution operation on the bottom layer feature graphs obtained by Conv9, conv8, conv7, conv5 and Conv4 in the depth convolution neural network to fix the channel number to 256, and adding the bottom layer feature graphs with the up-sampled high semantic features element by element to obtain F5, F4, F3, F2 and F1.
Convolution fusion: f5, F4, F3, F2, F1 were fused using a convolution of 3 x 3 size.
Step 5: non-maximum suppression algorithm deduplication
A modified algorithm Soft-NMS using a Non-maximum suppression algorithm (Non-Maximum Suppression, NMS) removes duplicate prediction bounding boxes. The specific expression of the Soft-NMS function is as follows:
wherein s is i For each bounding box score, M is the bounding box with the highest current score, b i For some bounding box remaining, N t To set a threshold of 0.5.
Step 6: detection result
And outputting the position information and the category of the surface defects of the workpiece.
The invention uses the dense+stride structure to replace dense+maxpooling in VGG16 for downsampling, so that more characteristic information is reserved, the speed is higher, and the detection efficiency is effectively improved.
According to the invention, the deep convolutional neural network is constructed to extract the characteristics, 4 deep convolutional layers are added on a network structure taking VGG16 as a base, and FPN is adopted to conduct multi-scale characteristic fusion prediction, so that higher semantic information is extracted, the detection accuracy is improved, the generalization capability of a model is enhanced, and the flexibility is improved.
The invention removes repeated prediction boundary boxes by using a Soft-NMS algorithm, can solve the problem of missed detection caused by overlapping a plurality of workpieces, and improves the recall rate and the robustness of detection.
The foregoing examples illustrate only a few embodiments of the invention and are described in detail herein without thereby limiting the scope of the invention. It should be noted that it will be apparent to those skilled in the art that several variations and modifications can be made without departing from the spirit of the invention, which are all within the scope of the invention. Accordingly, the scope of protection of the present invention is to be determined by the appended claims.
Claims (5)
1. The workpiece surface defect detection method based on deep learning is characterized by comprising the following steps of:
collecting workpiece images under different backgrounds and illumination conditions;
preprocessing the collected workpiece image;
constructing a deep convolutional neural network model to obtain feature maps of 6 different layers;
carrying out multi-scale feature fusion prediction by adopting a feature pyramid feature map, obtaining a boundary frame for generating 4 anchor box prediction targets by using a K-means clustering algorithm, and predicting categories by using a cross entropy loss function;
removing redundant prediction boundary boxes through a non-maximum suppression algorithm;
outputting position information and categories of the surface defects of the workpiece;
carrying out multi-scale feature fusion prediction by adopting a feature pyramid feature map, obtaining a boundary frame for generating 4 anchor box predictions by using a K-means clustering algorithm, and predicting the classes by using a cross entropy loss function specifically as follows:
6 feature graphs, namely 63×63, 32×32, 16×16, 8×8, 4×4 and 1×1 are obtained by using a deep convolutional neural network, and a feature pyramid method is adopted to conduct multi-scale feature fusion prediction on the feature graphs of 6 different layers; 4 anchor box prediction target boundary boxes are obtained through K-means algorithm clustering, and cross entropy loss function prediction categories are used;
the FPN overall architecture comprises 4 parts, namely a top-down network, a bottom-up network, transverse connection and convolution fusion;
top-down network: is a deep convolutional neural network;
bottom-up network: carrying out convolution with the size of 1 multiplied by 1 on a characteristic diagram 1 multiplied by 1 obtained by Conv11 in a depth convolution layer to reduce the channel number to obtain F6, and then carrying out 2 times nearest neighbor up-sampling operation in sequence;
and (3) transverse connection: carrying out 1X 1 convolution operation on bottom feature graphs obtained by Conv9, conv8, conv7, conv5 and Conv4 in the depth convolution neural network respectively to fix the channel number of the bottom feature graphs to 256, and adding the bottom feature graphs with the up-sampled high semantic features element by element to obtain F5, F4, F3, F2 and F1;
convolution fusion: f5, F4, F3, F2, F1 were fused using a convolution of 3 x 3 size.
2. The method for detecting surface defects of a workpiece based on deep learning according to claim 1, wherein the step of collecting workpiece images under different backgrounds and illumination conditions is specifically as follows:
and shooting workpiece images under different backgrounds and illumination conditions by using a CCD camera, wherein the workpiece images comprise single background, complex background, dark, bright, soft light, shielding and overlapped complex environments.
3. The method for detecting surface defects of a workpiece based on deep learning according to claim 1, wherein preprocessing the collected workpiece image specifically comprises:
firstly, carrying out data enhancement on an obtained workpiece image, wherein the data enhancement comprises optical transformation and geometric transformation; the optical transformation comprises brightness random adjustment, contrast random adjustment and channel random adjustment, and the geometric transformation comprises rotation, stretching, translation, horizontal overturning, vertical overturning, random picture cutting and random scaling;
then, generating corresponding labeling information for the expanded image data set by using a labeling tool; the labeling information comprises position information and categories of surface defects of the workpiece in the sample, wherein the categories are normal, scratches, convex powder, bumps and cracks;
finally, dividing the data set into a training set and a testing set; wherein the training set is 70% of the total sample size, and the test set is 30%.
4. The method for detecting the surface defects of the workpiece based on the deep learning as claimed in claim 1, wherein the construction of the deep convolutional neural network model to obtain the characteristic diagrams of 6 different layers is specifically as follows:
constructing a deep convolutional neural network, including constructing a basic network structure and a deep convolutional layer, and extracting high semantic features of an image; and in the deep convolutional neural network module, each convolutional layer operates by a ReLu function and normalization Batch Normalization;
(1) Basic network structure: VGG16 is adopted as a basic feature extraction module; firstly, through the first 13 convolution layers of the VGG16, and modifying, using dense conv+stride conv to replace dense conv+maxpooling in the VGG16 network structure, wherein the convolution kernel size of dense conv is 3 multiplied by 3, and the step size is 1; the convolution kernel size of stride conv is 3×3, the step size is 2; then, using two convolutions Conv6 and Conv7 to replace full connection layers FC6 and FC7 of VGG16 to further extract features, wherein Conv6 is cavity convolution, the size of convolution kernel is 3 multiplied by 3, the number of cavities is 6, and the step length is 1; conv7 has a convolution kernel size of 1×1 and a step size of 1;
(2) Depth convolution layer: adding 4 deep convolution layers on the basic network structure, and further extracting higher semantic information; the depth convolution layers are Conv8, conv9, conv10 and Conv11, respectively, each consisting of 2 convolutions: a convolution kernel of size 1 x 1, step size 1, and a convolution kernel of size 3 x 3, step size 2.
5. The method for detecting surface defects of a workpiece based on deep learning according to claim 1, wherein the removing redundant prediction bounding boxes by a non-maximum suppression algorithm is specifically:
a modified algorithm Soft-NMS using a non-maximum suppression algorithm removes duplicate prediction bounding boxes;
the specific expression of the Soft-NMS function is as follows:
wherein s is i For each bounding box score, M is the bounding box with the highest current score, b i For some bounding box remaining, N t To set a threshold of 0.5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010105860.7A CN111415329B (en) | 2020-02-20 | 2020-02-20 | Workpiece surface defect detection method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010105860.7A CN111415329B (en) | 2020-02-20 | 2020-02-20 | Workpiece surface defect detection method based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111415329A CN111415329A (en) | 2020-07-14 |
| CN111415329B true CN111415329B (en) | 2023-08-01 |
Family
ID=71492719
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010105860.7A Active CN111415329B (en) | 2020-02-20 | 2020-02-20 | Workpiece surface defect detection method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111415329B (en) |
Families Citing this family (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114076794A (en) * | 2020-08-19 | 2022-02-22 | 宝山钢铁股份有限公司 | Automatic detection device and detection method for near-surface defects of small strip steel square billet |
| CN112419237B (en) * | 2020-11-03 | 2023-06-30 | 中国计量大学 | Deep learning-based automobile clutch master cylinder groove surface defect detection method |
| CN112329861B (en) * | 2020-11-06 | 2024-05-28 | 北京工业大学 | Layered feature fusion method for mobile robot multi-target detection |
| CN112686843B (en) * | 2020-12-21 | 2023-09-15 | 福建新大陆软件工程有限公司 | Board defect detection method and system based on neural network |
| US20220222477A1 (en) * | 2021-01-14 | 2022-07-14 | Nvidia Corporation | Performing non-maximum suppression in parallel |
| CN112950584B (en) * | 2021-03-01 | 2022-07-29 | 哈尔滨工程大学 | Coating surface defect identification method based on deep learning |
| CN113111719A (en) * | 2021-03-16 | 2021-07-13 | 苏州海宸威视智能科技有限公司 | Vehicle detection method under unmanned aerial vehicle visual angle based on reconstructed brain-like convolution |
| CN112734760B (en) * | 2021-03-31 | 2021-08-06 | 高视科技(苏州)有限公司 | Semiconductor bump defect detection method, electronic device, and storage medium |
| CN113096098B (en) * | 2021-04-14 | 2024-01-05 | 大连理工大学 | Casting appearance defect detection method based on deep learning |
| CN114581806B (en) * | 2022-03-18 | 2024-03-19 | 重庆科技学院 | Industrial part empty rate calculation method based on trunk edge feature extraction |
| CN116109639B (en) * | 2023-04-13 | 2023-08-18 | 常州微亿智造科技有限公司 | Defect detection method and device based on multi-task learning network |
| CN117152577B (en) * | 2023-10-30 | 2024-02-02 | 北京机械工业自动化研究所有限公司 | Casting defect detection method, casting defect detection device, electronic equipment and storage medium |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110660040A (en) * | 2019-07-24 | 2020-01-07 | 浙江工业大学 | Industrial product irregular defect detection method based on deep learning |
-
2020
- 2020-02-20 CN CN202010105860.7A patent/CN111415329B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN111415329A (en) | 2020-07-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111415329B (en) | Workpiece surface defect detection method based on deep learning | |
| Ren et al. | State of the art in defect detection based on machine vision | |
| CN111862064B (en) | Silver wire surface flaw identification method based on deep learning | |
| CN108562589B (en) | Method for detecting surface defects of magnetic circuit material | |
| CN109460753B (en) | Method for detecting floating object on water | |
| CN108074231B (en) | Magnetic sheet surface defect detection method based on convolutional neural network | |
| CN109724984B (en) | Defect detection and identification device and method based on deep learning algorithm | |
| He et al. | Deep regression neural network for industrial surface defect detection | |
| CN112967243A (en) | Deep learning chip packaging crack defect detection method based on YOLO | |
| CN107966454A (en) | A kind of end plug defect detecting device and detection method based on FPGA | |
| CN111223093A (en) | AOI defect detection method | |
| CN109191421B (en) | Visual detection method for pits on circumferential surface of cylindrical lithium battery | |
| CN114022420A (en) | Detection method for automatically identifying defects of photovoltaic cell EL assembly | |
| CN112308826A (en) | Bridge structure surface defect detection method based on convolutional neural network | |
| CN112614105B (en) | Depth network-based 3D point cloud welding spot defect detection method | |
| Mao et al. | Automatic image detection of multi-type surface defects on wind turbine blades based on cascade deep learning network | |
| CN114973032A (en) | Photovoltaic panel hot spot detection method and device based on deep convolutional neural network | |
| CN115063725B (en) | Aircraft skin defect identification system based on multi-scale self-adaptive SSD algorithm | |
| CN115035081B (en) | Industrial CT-based metal internal defect dangerous source positioning method and system | |
| Zhou et al. | DeepInspection: Deep learning based hierarchical network for specular surface inspection | |
| CN112991271A (en) | Aluminum profile surface defect visual detection method based on improved yolov3 | |
| CN116385430A (en) | Machine vision flaw detection method, device, medium and equipment | |
| CN116051586A (en) | Detection method suitable for detecting outline of light color glue | |
| CN114549489A (en) | Carved lipstick quality inspection-oriented instance segmentation defect detection method | |
| CN114663382A (en) | Surface defect detection method for electronic component based on YOLOv5 convolutional neural network |
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 |