CN115965568A - Bathroom ceramic surface defect detection method and system - Google Patents
Bathroom ceramic surface defect detection method and system Download PDFInfo
- Publication number
- CN115965568A CN115965568A CN202111181870.XA CN202111181870A CN115965568A CN 115965568 A CN115965568 A CN 115965568A CN 202111181870 A CN202111181870 A CN 202111181870A CN 115965568 A CN115965568 A CN 115965568A
- Authority
- CN
- China
- Prior art keywords
- defect
- defect detection
- training
- ceramic
- bathroom
- 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
-
- 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
- Image Analysis (AREA)
Abstract
The invention discloses a bathroom ceramic surface defect detection method, which comprises the following steps: step 1, collecting a surface defect image of the sanitary ware, and marking the defect type and the defect position in the surface defect image to obtain a training data set of the surface defect of the sanitary ware; step 2, constructing a defect detection model by taking the SSD network model as a reference; step 3, training the defect detection model by using the training data set of the ceramic defects on the surface of the bathroom ceramic; and 4, inputting the surface defect image of the bathroom ceramic to be detected into the defect detection model obtained by training in the step 3, and outputting the defect type and the defect position of the surface defect image of the bathroom ceramic to be detected.
Description
Technical Field
The invention relates to the technical field of defect detection, in particular to a bathroom ceramic surface defect detection method and system.
Background
The bathroom ceramic is a necessary product in daily life, and is used by each household, and the quality of the bathroom ceramic directly influences the daily experience of the user, so that it is very important to keep the factory quality of bathroom ceramic products. At the present stage, the detection of the surface defects of the bathroom ceramics is mainly completed manually, the defects on the surface of a workpiece are identified by visual observation of quality inspection workers, the efficiency is low, the experience of the workers is seriously depended, and the quality inspection result is extremely easily influenced by the personal state of the workers.
With the rapid development of deep learning techniques, more and more industries are beginning to adopt deep learning-based image processing techniques for defect detection. Nowadays, many companies and organizations begin to apply deep learning technology to surface defect detection of sanitary ware ceramics, and all achieve good results.
At present, the existing one-stage deep learning target detection model has the characteristics of high detection precision and high speed, has good universality and is suitable for beat sensitive production work. The production beat of the bathroom ceramic is fast, the surface defect types are many, and the characteristics are complex. In view of the above, the present inventors have provided a method for detecting surface defects of a sanitary ware through a large number of experimental researches to solve the above problems.
Disclosure of Invention
Aiming at the defects of the prior art, the invention aims to provide a method and a system for detecting the surface defects of the bathroom ceramics, which can obviously improve the detection efficiency and the detection precision.
In order to achieve the above object, the present invention provides a method for detecting surface defects of a sanitary ceramic, comprising:
step 1, collecting a surface defect image of the sanitary ware, and labeling the defect type and the defect position in the surface defect image to obtain a training data set of the surface defect of the sanitary ware;
and 4, inputting the surface defect image of the bathroom ceramic to be detected into the defect detection model obtained by training in the step 3, and outputting the defect type and the defect position of the surface defect image of the bathroom ceramic to be detected.
In the method for detecting surface defects of sanitary ware, the defect type includes at least one of the following: cracks, lack of glaze, rust spots, green spots and streaky glaze shortages.
In the method for detecting the surface defects of the sanitary ceramic, the backbone network in the defect detection model established based on the SSD network model is replaced by a ResNet-50 network having a jump connection structure.
In the method for detecting surface defects of bathroom ceramics, the average pooling layer and the full-connection layer at the end of the ResNet-50 network are changed into convolution layers, the Dropout layer is removed, and a plurality of residual convolution modules are added.
In the method for detecting the surface defects of the sanitary ware, a Mish function is used as an activation function of the defect detection model.
The method for detecting the surface defects of the bathroom ceramics comprises the following steps of 3:
scaling the size of an image input to the defect detection model to 300 × 300;
and selecting feature maps with the sizes of 38 multiplied by 38, 19 multiplied by 19, 10 multiplied by 10, 5 multiplied by 5, 3 multiplied by 3 and 1 multiplied by 1 from the defect detection model to be responsible for the prediction task of the result.
The method for detecting surface defects of bathroom ceramics comprises the following steps of (3):
the defect detection model sets different numbers of prior frames for each feature point of the feature map of each size, and the number of the prior frames corresponding to each feature point of the feature map is respectively 4, 6, 4 and 4;
a priori box scale s of the feature map k 30, 60, 111, 162, 213, 264 respectively, and for general aspect ratios, the ratio is selected
Or the width and height of the prior box are calculated for a particular aspect ratio as follows:
wherein aspect ratios of 3 and 1 are not used for the feature maps of 38 × 38, 3 × 3, and 1 × 1 dimensions
A priori block(s).
In the method for detecting surface defects of sanitary ceramics, for the sanitary ceramics surface defect image used for training, a bounding box of a single real target may correspond to a plurality of the prior frames, and each prior frame corresponds to a bounding box of one real target.
In the method for detecting the surface defects of the sanitary ware, the loss function of the defect detection model is a weighted sum of the position loss and the confidence coefficient loss, and the formula is as follows:
wherein, N is the number of positive samples of the prior frame, c is the predicted value of the category confidence coefficient, l is the predicted value of the position of the corresponding boundary frame of the prior frame, and g is the position parameter of the boundary frame.
In the method for detecting the surface defects of the sanitary ware, a Smooth L1Loss is adopted for the position Loss, and the formula is as follows:
for confidence Loss, softmax Loss is used, the formula is as follows:
in the method for detecting the surface defects of the bathroom ceramics, a label smoothing technology is adopted in the step 3 to prevent the network model from being over-fitted.
In the method for detecting the defects of the bathroom ceramics, the migration learning is adopted in the step 3, the training is performed in two stages, the batch size of the first stage is 16, the learning rate of 0.001 is adopted for training a plurality of rounds, the batch size of the second stage is 8, the learning rate of 0.0001 is adopted for training a plurality of rounds, and a learning rate reduction strategy and a training termination strategy in advance are adopted during the training of the second stage.
In order to achieve the above object, the present invention further provides a system for detecting surface defects of a sanitary ware, comprising:
the data acquisition module is used for acquiring a sanitary ceramic surface defect image, marking the defect type and the defect position in the surface defect image and obtaining a training data set of the sanitary ceramic surface defect;
the model construction module is used for constructing a defect detection model by taking the SSD network model as a reference;
the model training module is used for training the defect detection model by utilizing the training data set of the ceramic defects on the surface of the sanitary ware;
and the defect detection module is used for inputting the surface defect image of the bathroom ceramic to be detected into the defect detection model obtained by training in the step 3 and outputting the defect type and the defect position of the surface defect image of the bathroom ceramic to be detected.
In the bathroom ceramic surface defect detection system, the backbone network in the defect detection model constructed based on the SSD network model is replaced with a ResNet-50 network having a jumper connection structure.
In the bathroom ceramic surface defect detection system, the average pooling layer and the full-connection layer at the end of the ResNet-50 network are changed into convolution layers, the Dropout layer is removed, and a plurality of residual convolution modules are added.
According to the scheme, the invention has the advantages that:
firstly, the technical scheme of the invention adopts the target detection technology of deep learning as a basic frame, so that the influence of factors such as illumination, brightness, photo exposure and the like on the detection result can be effectively reduced, and the defect detection precision is greatly improved. Secondly, the ResNet-50 network is used as the backbone network of the SSD model, so that the expression capability of the model is improved, the depth of the network is deepened, the model parameters are reduced, and the detection effect of the model can be effectively improved. Thirdly, a real-time data enhancement algorithm is adopted during model training, training data can be expanded in real time, the problem of training data shortage is effectively relieved, and the generalization capability of the model is improved. Fourth, the method is suitable for detecting the defects of cracks, glaze shortage, rusty spots, green spots, strip-shaped glaze shrinkage and the like on the surface of the bathroom ceramic, is high in detection precision and wide in application range, and has a good detection effect on the surface defects of the typical bathroom ceramic.
The invention is described in detail below with reference to the drawings and specific examples, but the invention is not limited thereto.
Drawings
Fig. 1 is a flowchart of a bathroom ceramic surface defect detection method according to an embodiment of the invention.
Fig. 2 is a schematic diagram of a backbone network ResNet-50 network structure.
Fig. 3 is a schematic diagram of the activation function Mish.
FIG. 4 is a schematic diagram of real-time data enhancement by the Mosaic algorithm.
FIG. 5 is a schematic diagram of the detection result of the surface defect of the sanitary ware.
FIG. 6 is a block diagram of a sanitary ceramic surface defect detection system according to an embodiment of the present invention
Detailed Description
In order to make the aforementioned features and advantages of the present invention more comprehensible, embodiments accompanied with figures are described in detail below.
Referring to fig. 1, an embodiment of the invention provides a method 100 for detecting surface defects of a sanitary ceramic, including steps S110-S140, wherein:
step S110, collecting a surface defect image of the sanitary ware, and labeling the defect type and the defect position in the surface defect image to obtain a training data set of the surface defect of the sanitary ware.
Specifically, the actually acquired image data of the surface defects of the sanitary ware at the production site can be used in the example, wherein the image data includes the following defects: cracks, lack of glaze, rusty spots, green spots and strip-shaped glaze shrinkage defects.
In this embodiment, when labeling the acquired original image data, specifically, for example, labelImg open-source auxiliary image calibration software is used to label all images including defects, mark corresponding defect labels, and mark defect types and positions of the defects (defect positions) in the images, so as to form a label file in an XML format.
In addition, a data set of the surface defects of the manufactured sanitary ware is divided, and the data set is randomly divided into a training set, a verification set and a test set according to the proportion of 8.
And step S120, constructing a defect detection model by taking the SSD network model as a reference.
Specifically, for example, a Tensorflow and Keras open source deep learning framework is utilized, and a defect detection model is established by taking an SSD network model as a reference; however, in the present embodiment, a certain improvement is made to the network model, and the whole network model can be mainly divided into two parts: a backbone network and a detection network.
The backbone network part adopts a ResNet-50 network, the network structure is shown in figure 2, and the backbone network is used as a feature extraction network and mainly responsible for extracting features of input images. The network layer number of ResNet-50 is deeper, the parameters of the model are reduced, and the jump connection structure is greatly adopted in the network model, so that the problem that the gradient disappears when the deep neural network model is trained is solved. In addition, in the invention, the average pooling layer and the full-link layer at the tail part of the backbone network ResNet-50 are removed, the convolution layer is changed, the Dropout layer is also removed, and a plurality of residual convolution modules (for example, 4 residual convolution modules) are additionally arranged at the tail end of the model to extract feature maps with different scales.
It should be further noted that, for the activation functions in the backbone network ResNet-50 of the present invention, all use the Mish activation function, as shown in fig. 3, whose mathematical expression is as follows:
Mish＝x×tanh(ln(1+e x ))
and S130, training the defect detection model by using the training data set of the ceramic surface defects of the bathroom ceramics.
The input image size of the detection network for the defect detection model is uniformly scaled to 300 × 300, and the feature pyramid structure can be used to improve the accuracy of the model, and feature maps of 6 sizes, namely 38 × 38, 19 × 19, 10 × 10, 5 × 5, 3 × 3 and 1 × 1, are selected from different levels of the backbone network to be responsible for the prediction task of the result.
In addition, the defect detection model sets different numbers of prior frames for each feature point of the feature map of each size, and for the 6 feature maps, the number of the prior frames corresponding to each feature point of each feature map is 4, 6, 4; prior frame scale s of six feature maps k 30, 60, 111, 162, 213, 264 respectively, and for general aspect ratios, the ratio is selected
The width and height of the prior box are calculated for a particular aspect ratio as follows:
wherein for feature maps of 3 sizes 38 × 38, 3 × 3, and 1 × 1, only 4 prior blocks are used, which do not use aspect ratios of 3 and 1
A priori block(s).
The relation between the real target (Ground Truth) and the prior frame in the training image is that the bounding box of a single real target may correspond to a plurality of prior frames, but each prior frame may only correspond to the bounding box of one real target.
The loss function of the defect detection model is a weighted sum of the position loss and the confidence loss, wherein the weight coefficient α is set to 1 by cross validation:
wherein, N is the number of positive samples of the prior frame, c is the category confidence prediction value, l is the position prediction value of the corresponding boundary frame of the prior frame, and g is the position parameter of the real target (Ground Truth).
For position Loss, smooth L1Loss is used, defined as follows:
for confidence Loss, softmax Loss is used, defined as follows:
the model training adopts transfer learning and two-stage training, the batch size of the first stage is 16, the learning rate of 0.001 is adopted for training a plurality of (for example, 100) rounds, the batch size of the second stage is 8, the learning rate of 0.0001 is adopted for training a plurality of (for example, 400) rounds, and the learning rate reduction strategy and the early termination training strategy are adopted during the second stage training.
The learning rate reduction strategy is that after the network model is trained for a certain number of times, the loss function of the network does not change any more, and the learning rate is reduced, so that a better training effect is obtained. The early termination training strategy is that after a certain round of training, the loss function of the network does not change any more, the network is considered to be converged at the moment, and the training is terminated in advance.
In the embodiment of the present invention, the tolerance value of the learning rate decreasing strategy is set to 3, the momentum parameter is 0.1, and the tolerance value of the early termination training strategy is set to 10. It shows that when 3 rounds of training are performed, the model loss function does not decrease, and the learning rate decreases to one tenth of the previous one. When 10 rounds of training are carried out, the model loss function does not decrease, the model is considered to reach a convergence state at the moment, and the training is terminated early.
In the defect detection model training, embodiments of the present invention may, for example, use Label Smoothing technology (Label Smoothing) to add a little penalty to accurately classified classes and slightly reduce the classification precision of the models for these classes, thereby preventing the network model from being over-fitted.
Optionally, when the defect detection model is trained, for example, the training data is extended by using a method of Mosaic online data enhancement, as shown in fig. 4, for example, four images are randomly read each time, the four images are respectively subjected to operations such as flipping, zooming, color gamut change, and the like, and are well placed according to four directions, and image combination and frame combination are performed, so that the training data amount is improved by using the method, and the model training effect is ensured.
Step S140, inputting the image of the surface defect of the sanitary ware to be detected into the defect detection model obtained by training in the step S130, and outputting the defect type and the defect position of the image of the surface defect of the sanitary ware to be detected.
In this embodiment, after the defect detection model trained completely through the above process is obtained, a defect detection task may be performed. Specifically, a defect detection model to be detected is input into a defect detection model, 6 feature maps with different sizes are extracted through a ResNet-50 network in the model, and then result prediction is carried out according to information in the feature maps.
In the result prediction process, for each prediction frame, firstly, determining the category and the confidence value of the prediction frame according to the category confidence, and filtering the prediction frames belonging to the background; then, a prediction frame with a lower threshold is filtered according to the confidence threshold, the remaining prediction frames are decoded, the true position parameters of the prediction frames are obtained according to the prior frames, generally, the prediction frames need to be arranged in a descending order according to the confidence, and then only the prediction frames such as top-400 are reserved.
After the prediction frames are obtained, a non-maximum suppression algorithm is performed, the prediction frames with a large overlap degree are filtered, and finally the remaining prediction frames are the detection results, which are shown in fig. 5 and are a schematic diagram of the crack defect detection results.
Referring to fig. 6, based on the same inventive concept, an embodiment of the invention provides a bathroom ceramic surface defect detection system 200, and it should be noted that, for convenience and simplicity of description, specific working processes of modules of the system may refer to corresponding processes in the foregoing method embodiments, and are not repeated herein. The detection system 200 specifically includes:
the data acquisition module 210 is configured to acquire a sanitary ceramic surface defect image, and label a defect type and a defect position in the surface defect image to obtain a training data set of a sanitary ceramic surface defect;
a model construction module 220, configured to construct a defect detection model based on the SSD network model;
a model training module 230, configured to train the defect detection model with a training data set of the ceramic defect on the surface of the sanitary ware;
the defect detection module 240 is configured to input the to-be-detected surface defect image of the sanitary ceramic into the defect detection model obtained by training of the model training module 230, and output the defect type and the defect position of the to-be-detected surface defect image of the sanitary ceramic.
In an embodiment of the present invention, the backbone network in the defect detection model constructed based on the SSD network model is replaced with a ResNet-50 network including a hop-and-connect structure.
In one embodiment of the present invention, the average pooling layer and the fully connected layer at the end of the ResNet-50 network are changed to convolutional layers, dropout layers are removed, and several residual convolution modules are added.
The present invention is capable of other embodiments, and various changes and modifications can be made by one skilled in the art without departing from the spirit and scope of the invention.
Claims (15)
1. A method for detecting surface defects of bathroom ceramics is characterized by comprising the following steps:
step 1, collecting a surface defect image of the sanitary ware, and labeling the defect type and the defect position in the surface defect image to obtain a training data set of the surface defect of the sanitary ware;
step 2, constructing a defect detection model by taking the SSD network model as a reference;
step 3, training the defect detection model by using the training data set of the ceramic surface defects of the bathroom ceramics;
and 4, inputting the surface defect image of the bathroom ceramic to be detected into the defect detection model obtained by training in the step 3, and outputting the defect type and the defect position of the surface defect image of the bathroom ceramic to be detected.
2. The bathroom ceramic surface defect detection method of claim 1, wherein the defect category comprises at least one of: cracks, lack of glaze, rusty spots, green spots and streak-like glaze reduction.
3. The method for detecting surface defects of sanitary ceramics according to claim 1, wherein the backbone network in the defect detection model constructed based on the SSD network model is replaced with a ResNet-50 network comprising a jumper connection structure.
4. The bathroom ceramic surface defect detection method of claim 3, wherein the average pooling layer and the full connection layer at the end of the ResNet-50 network are changed into convolution layers, and Dropout layers are removed and a plurality of residual convolution modules are added.
5. The method of claim 1, wherein a Mish function is used as an activation function of the defect detection model.
6. The method for detecting surface defects of bathroom ceramics of claim 1, wherein the step 3 comprises:
scaling an image input to the defect detection model to a size of 300 × 300;
and selecting feature maps with the sizes of 38 multiplied by 38, 19 multiplied by 19, 10 multiplied by 10, 5 multiplied by 5, 3 multiplied by 3 and 1 multiplied by 1 from the defect detection model to be responsible for the prediction task of the result.
7. The method for detecting surface defects of bathroom ceramics of claim 6, wherein the step 3 further comprises:
the defect detection model sets different numbers of prior frames for each feature point of the feature map of each size, and the number of the prior frames corresponding to each feature point of the feature map is respectively 4, 6, 4 and 4;
a priori box scale s of the feature map k 30, 60, 111, 162, 213, 264 respectively, and for general aspect ratios, are selected
Or the width and height of the prior box are calculated for a particular aspect ratio as follows:
wherein aspect ratios of 3 and 1 are not used for the feature maps of 38 × 38, 3 × 3, and 1 × 1 dimensions
A priori block(s).
8. The method of claim 1, wherein for the sanitary ceramic surface defect image used for training, a bounding box of a single real target may correspond to a plurality of the prior boxes, each of the prior boxes corresponding to a bounding box of the real target.
9. The method of claim 1, wherein the loss function of the defect detection model is a weighted sum of position loss and confidence loss, and the formula is as follows:
wherein, N is the number of positive samples of the prior frame, c is the predicted value of the category confidence coefficient, l is the predicted value of the position of the corresponding boundary frame of the prior frame, and g is the position parameter of the boundary frame.
10. The method for detecting surface defects of bathroom ceramics of claim 9, wherein for position Loss, a Smooth L1Loss is adopted, and the formula is as follows:
for confidence Loss, softmax Loss is used, the formula is as follows:
11. the method for detecting surface defects of sanitary ceramics according to claim 1, wherein a label smoothing technique is used in the step 3 to prevent overfitting of the network model.
12. The bathroom ceramic defect detection method of claim 1, wherein in the step 3, transfer learning is adopted, and training is performed in two stages, the first stage has a batch size of 16, a plurality of rounds of training are performed with a learning rate of 0.001, the second stage has a batch size of 8, a plurality of rounds of training are performed with a learning rate of 0.0001, and a learning rate reduction strategy and a training termination strategy are adopted during the second stage training.
13. A sanitary ware ceramic surface defect detection system, comprising:
the data acquisition module is used for acquiring a sanitary ceramic surface defect image, marking the defect type and the defect position in the surface defect image and obtaining a training data set of the sanitary ceramic surface defect;
the model construction module is used for constructing a defect detection model by taking the SSD network model as a reference;
the model training module is used for training the defect detection model by utilizing the training data set of the ceramic defects on the surface of the sanitary ware;
and the defect detection module is used for inputting the surface defect image of the bathroom ceramic to be detected into the defect detection model obtained by training in the step 3 and outputting the defect type and the defect position of the surface defect image of the bathroom ceramic to be detected.
14. The sanitary ceramic surface defect detection system of claim 13, wherein the backbone network in the defect detection model constructed based on the SSD network model is replaced with a ResNet-50 network comprising a jumper structure.
15. The sanitary ceramic surface defect detection system of claim 14 wherein the ResNet-50 net end averaging pooling layer and full connection layer are changed to convolution layers and Dropout layers are removed and residual convolution modules are added.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111181870.XA CN115965568A (en) | 2021-10-11 | 2021-10-11 | Bathroom ceramic surface defect detection method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111181870.XA CN115965568A (en) | 2021-10-11 | 2021-10-11 | Bathroom ceramic surface defect detection method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115965568A true CN115965568A (en) | 2023-04-14 |
Family
ID=87362273
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111181870.XA Pending CN115965568A (en) | 2021-10-11 | 2021-10-11 | Bathroom ceramic surface defect detection method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115965568A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117726628A (en) * | 2024-02-18 | 2024-03-19 | 青岛理工大学 | Steel surface defect detection method based on semi-supervised target detection algorithm |
-
2021
- 2021-10-11 CN CN202111181870.XA patent/CN115965568A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117726628A (en) * | 2024-02-18 | 2024-03-19 | 青岛理工大学 | Steel surface defect detection method based on semi-supervised target detection algorithm |
| CN117726628B (en) * | 2024-02-18 | 2024-04-19 | 青岛理工大学 | Steel surface defect detection method based on semi-supervised target detection algorithm |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110570396B (en) | Industrial product defect detection method based on deep learning | |
| CN110059694B (en) | Intelligent identification method for character data in complex scene of power industry | |
| CN108960245B (en) | Tire mold character detection and recognition method, device, equipment and storage medium | |
| CN109509187B (en) | Efficient inspection algorithm for small defects in large-resolution cloth images | |
| CN110853015A (en) | Aluminum profile defect detection method based on improved Faster-RCNN | |
| CN111402226A (en) | Surface defect detection method based on cascade convolution neural network | |
| CN106295653B (en) | Water quality image classification method | |
| CN110472636B (en) | Deep learning-based water gauge E-shaped scale identification method | |
| CN108960135B (en) | Dense ship target accurate detection method based on high-resolution remote sensing image | |
| CN108710919A (en) | A kind of crack automation delineation method based on multi-scale feature fusion deep learning | |
| CN108961230B (en) | Identification and extraction method for structural surface crack characteristics | |
| CN110751644B (en) | Road surface crack detection method | |
| CN112330593A (en) | Building surface crack detection method based on deep learning network | |
| CN108985337A (en) | A kind of product surface scratch detection method based on picture depth study | |
| CN110245697B (en) | Surface contamination detection method, terminal device and storage medium | |
| CN111222519B (en) | Construction method, method and device of hierarchical colored drawing manuscript line extraction model | |
| CN114663436A (en) | Cross-scale defect detection method based on deep learning | |
| CN111369526B (en) | Multi-type old bridge crack identification method based on semi-supervised deep learning | |
| CN107622276B (en) | Deep learning training method based on combination of robot simulation and physical sampling | |
| CN111161224A (en) | Casting internal defect grading evaluation system and method based on deep learning | |
| CN109685030A (en) | A kind of mug rim of a cup defects detection classification method based on convolutional neural networks | |
| CN106340007A (en) | Image processing-based automobile body paint film defect detection and identification method | |
| CN114255223A (en) | Deep learning-based method and equipment for detecting surface defects of two-stage bathroom ceramics | |
| CN115965568A (en) | Bathroom ceramic surface defect detection method and system | |
| CN112967271B (en) | Casting surface defect identification method based on improved DeepLabv3+ network model |
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 |