CN114998580A - Fracture semantic segmentation method based on deep LabV3+ network model - Google Patents
Fracture semantic segmentation method based on deep LabV3+ network model Download PDFInfo
- Publication number
- CN114998580A CN114998580A CN202210434313.2A CN202210434313A CN114998580A CN 114998580 A CN114998580 A CN 114998580A CN 202210434313 A CN202210434313 A CN 202210434313A CN 114998580 A CN114998580 A CN 114998580A
- Authority
- CN
- China
- Prior art keywords
- convolution
- detected
- fracture
- crack
- network model
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Abstract
The invention provides a method for semantically segmenting cracks based on a DeepLabV3+ network model, which can effectively improve the semantically segmenting accuracy of the cracks of the asphalt pavement and improve the overall crack prediction performance of the network. The fracture semantic segmentation method specifically comprises the following steps: coding the crack image to be detected by adopting parallel common convolution, cavity convolution and deformable convolution; decoding the encoded crack image to be detected through bilinear interpolation; performing key pixel point reconstruction on the decoded crack image to be detected through pixel point rendering; and classifying and predicting the reconstructed crack image to be detected to obtain a crack segmentation result.
Description
Technical Field
The invention relates to the technical field of image semantic segmentation, in particular to a fracture semantic segmentation method based on a DeepLabV3+ network model.
Background
The asphalt pavement cracking can accelerate the pavement degradation process, and the occurrence and severity of cracking are important indexes of a pavement maintenance decision scheme. Therefore, crack evaluation is a key task for making timely maintenance decisions. Traditionally, fracture evaluation is performed by manual field investigation. However, these manual measurement methods are poor in reproducibility and reproducibility, require excessive time, consume a large amount of labor, and place the surveyors in a dangerous environment. Furthermore, due to subjectivity, the data collected may vary from one scorer to another. In order to overcome the defects of manual measurement, a great deal of research work is carried out at present, and with the development of artificial intelligence deep learning, the development of an automatic crack detection and measurement method becomes a research hotspot.
At present, crack detection can be divided into three types of tasks, namely crack classification, target detection and semantic segmentation. The segmentation task needs to classify the picture pixel points from the viewpoint of the pixel points, and the picture pixel points are divided into backgrounds and cracks. For a general semantic segmentation task, coding is carried out through a deep neural network to extract features, and then decoding is carried out through linear interpolation to restore the original image size. In the early research of investigation, it is found that different backbone networks in the existing model have certain differences on the segmentation results of the cracks, and part of the cracks cannot be accurately identified, which has great influence on the accuracy of maintenance data and the reliability formulated by the maintenance scheme.
Disclosure of Invention
In order to solve the problems in the prior art, the invention provides the method for semantically segmenting the cracks based on the DeepLabV3+ network model, which can effectively improve the accuracy of semantically segmenting the cracks of the asphalt pavement and improve the overall crack prediction performance of the network.
In order to achieve the purpose, the invention provides the following technical scheme:
a method for semantically segmenting cracks based on a DeepLabV3+ network model comprises the following steps:
coding the crack image to be detected by adopting parallel common convolution, cavity convolution and deformable convolution;
decoding the coded crack image to be detected through bilinear interpolation;
performing key pixel point reconstruction on the decoded crack image to be detected through pixel point rendering;
and classifying and predicting the reconstructed crack image to be detected to obtain a crack segmentation result.
Preferably, before encoding the crack image to be detected by using the parallel common convolution, the void convolution and the deformable convolution, the method further includes:
firstly, carrying out primary feature extraction on a crack image to be detected by adopting a Backbone network, and then carrying out multi-scale convolution on the primary extracted features to complete further feature extraction.
Preferably, the deformable convolution comprises adding a learnable two-dimensional offset to each sample position on the basis of a normal convolution.
Preferably, the encoding of the crack image to be detected by using parallel common convolution, hole convolution and deformable convolution further includes:
and after the features of the crack image to be detected are extracted by adopting parallel common convolution, cavity convolution and deformable convolution, fusing the extracted features of different convolutions.
Preferably, the reconstructing key pixel points of the decoded crack image to be detected through pixel point rendering comprises:
and selecting N uncertain pixel points on the grid of the crack image to be detected after decoding, and reconstructing point-by-point characteristics on the selected uncertain pixel points by combining the low-layer characteristics and the high-layer characteristics of the crack image to be detected.
Preferably, after the reconstruction of the key pixel points, the method further includes:
and carrying out multilayer perceptron learning on the pixel points of the counterweight.
A fracture semantic segmentation system based on a DeepLabV3+ network model comprises:
the encoding module is used for encoding the crack image to be detected by adopting parallel common convolution, cavity convolution and deformable convolution;
the decoding module is used for decoding the coded crack image to be detected through bilinear interpolation;
the pixel point rendering module is used for performing key pixel point reconstruction on the decoded crack image to be detected through pixel point rendering;
and the fracture segmentation result output module is used for carrying out classification prediction on the reconstructed to-be-detected fracture image to obtain a fracture segmentation result.
Preferably, the encoding module further comprises a backhaul network module,
the backsbone network module is used for performing primary feature extraction on the crack image to be detected and then performing convolution on the primary extracted features to complete further feature extraction.
Preferably, the encoding module comprises a deformable convolution module,
the deformable convolution module is used for adding a learnable two-dimensional offset to each sampling position on the basis of the ordinary convolution.
Preferably, the pixel point rendering module further comprises an MLP module,
the MLP module is used for carrying out multi-layer perceptron learning on the pixel points of the counterweight.
Compared with the prior art, the invention has the following beneficial effects:
the invention provides a method for splitting a crack semantic based on a DeepLabV3+ network model, which is based on the DeepLabV3+ network model and further improves the network performance by applying a void convolution, a deformable convolution and a pixel point rendering reconstruction mode. Compared with the common convolution adopted in the traditional network model, the parallel-connected cavity convolution and deformable convolution adopted in the invention have larger receptive field, and meanwhile, key crack edge regions can be mainly studied, thereby realizing the accuracy of crack pixel point identification. Then decoding is carried out through bilinear interpolation, the feature graph of the to-be-detected crack image after coding is amplified to the size of an original image, finally, key pixel point reconstruction is carried out through pixel point rendering (PointRend), on the basis that the bilinear interpolation carries out up-sampling on the previously predicted segmentation of the to-be-detected crack image, N most uncertain points are selected on the feature graph grid of the to-be-detected crack image which is obtained more densely, point-by-point features are constructed on the selected points through combining low-layer features and high-layer features, then classification prediction is carried out on each selected point, fine-grained features in the low-layer and the high-layer can enable the pixel point rendering (PointRend) to present fine segmentation details, rough segmentation prediction of the high-layer features can be supplemented, more global information is provided, and a better prediction result is obtained after the two are combined. Therefore, compared with the existing network model, the method has the advantages that the hole convolution, the deformable convolution and the pixel point rendering network are combined, so that the characteristics can be better extracted, the expression capability of the characteristics is improved, the semantic segmentation accuracy of the asphalt pavement cracks can be effectively improved, and the overall crack prediction performance of the network can be effectively improved.
Drawings
FIG. 1 is a schematic diagram of a network model based on DeepLabV3 +;
FIG. 2 is a flow chart of the steps of the fracture semantic segmentation method of the present invention;
FIG. 3 is a graph comparing network model segmentation results according to an embodiment of the present invention;
FIG. 4 is a graph comparing network model segmentation results according to an embodiment of the present invention;
fig. 5 is a comparison graph of network model segmentation results according to an embodiment of the present invention.
Detailed Description
In order to make the technical solutions of the present invention better understood, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
It should be noted that the terms "first," "second," and the like in the description and claims of the present invention and in the drawings described above are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the invention described herein are capable of operation in sequences other than those illustrated or described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
The invention is described in further detail below with reference to the accompanying drawings:
as shown in fig. 2, the method for splitting the fracture semantics based on the deep lab v3+ network model of the present invention includes the following steps:
coding the crack image to be detected by adopting parallel common convolution, cavity convolution and deformable convolution;
decoding the encoded crack image to be detected through bilinear interpolation;
performing key pixel point reconstruction on the decoded crack image to be detected through pixel point rendering;
and classifying and predicting the reconstructed crack image to be detected to obtain a crack segmentation result.
The invention provides a method for splitting a crack semantic based on a DeepLabV3+ network model, which is based on the DeepLabV3+ network model and further improves the network performance by applying a void convolution, a deformable convolution and a pixel point rendering reconstruction mode. Compared with the common convolution adopted in the traditional network model, the method has the advantages that the cavity convolution and the deformable convolution which are connected in parallel have larger receptive field, and meanwhile, key crack edge regions can be mainly learned, so that the accuracy of crack pixel point identification is realized. Then decoding is carried out through bilinear interpolation, the feature graph of the to-be-detected crack image after coding is amplified to the size of an original image, finally, key pixel point reconstruction is carried out through pixel point rendering (PointRend), on the basis that the bilinear interpolation carries out up-sampling on the previously predicted segmentation of the to-be-detected crack image, N most uncertain points are selected on the feature graph grid of the to-be-detected crack image which is obtained more densely, point-by-point features are constructed on the selected points through combining low-layer features and high-layer features, then classification prediction is carried out on each selected point, fine-grained features in the low-layer and the high-layer can enable the pixel point rendering (PointRend) to present fine segmentation details, rough segmentation prediction of the high-layer features can be supplemented, more global information is provided, and a better prediction result is obtained after the two are combined. Therefore, compared with the existing network model, the method has the advantages that the hole convolution, the deformable convolution and the pixel point rendering network are combined, so that the characteristics can be better extracted, the expression capability of the characteristics is improved, the semantic segmentation accuracy of the asphalt pavement cracks can be effectively improved, and the overall crack prediction performance of the network can be effectively improved.
Further, before encoding the crack image to be detected by using the parallel common convolution, the hole convolution and the deformable convolution, the method further includes:
firstly, a backsbone network in a deep LabV3+ network model is adopted to perform primary feature extraction on a crack image to be detected, then the primary extracted features are convoluted to complete further feature extraction, and the backsbone network can be replaced by other network structures in the actual use process.
The hole convolution introduces holes in the traditional common convolution, the introduction of the hole convolution can increase the reception fields under the condition of not reducing the resolution, the information of different reception fields can be obtained by using the hole convolution with different hole rates, and the capability of a network for capturing multi-scale context information is improved.
The deformable convolution comprises the step of adding a learnable two-dimensional offset to each sampling position on the basis of common convolution, and because the shape of a sampling grid of the deformable convolution is dynamic and learnable, in a deep convolution neural network, high-level semantic information is coded in a deeper network layer, and spatial information is weaker, the characteristic extraction performance of the network can be further improved by adopting the deformable convolution.
Further, the encoding of the crack image to be detected by using the parallel common convolution, the hole convolution and the deformable convolution further includes:
after the parallel common convolution, cavity convolution and deformable convolution are adopted to extract the features of the crack image to be detected, the features of the crack image to be detected extracted by different convolutions are fused, the features containing semantic information obtained by the parallel different convolution extraction are fused, multi-scale context information can be captured, the semantic information of the crack image feature diagram to be detected is enhanced, and further the target response is enhanced.
Further, the pixel rendering (PointRend) network upsamples the previously predicted segmentation by using bilinear interpolation, then selects N most uncertain points on the denser grid, constructs point-by-point characteristics on the selected points by combining low-layer characteristics and high-layer characteristics, and then performs classification prediction on each selected point by using a multilayer perceptron (MLP).
The fine-grained characteristic of the lower layer can enable PointRend to present fine segmentation details, the rough segmentation prediction of the upper layer can be supplemented, more global information is provided, and a better prediction result is obtained after the PointRend and the rough segmentation prediction are combined.
The invention also provides a fracture semantic segmentation system based on the DeepLabV3+ network model, which is used for realizing the fracture semantic segmentation method based on the DeepLabV3+ network model, and comprises the following steps:
the encoding module is used for encoding the crack image to be detected by adopting parallel common convolution, cavity convolution and deformable convolution;
the decoding module is used for decoding the coded crack image to be detected through bilinear interpolation;
the pixel point rendering module is used for performing key pixel point reconstruction on the decoded crack image to be detected through pixel point rendering;
and the fracture segmentation result output module is used for carrying out classification prediction on the reconstructed to-be-detected fracture image to obtain a fracture segmentation result.
Further, the coding module also comprises a backhaul network module,
the backsbone network module is used for performing primary feature extraction on the crack image to be detected and then performing convolution on the primary extracted features to complete further feature extraction. .
Wherein the encoding module comprises a deformable convolution module,
the deformable convolution module is used for adding a learnable two-dimensional offset to each sampling position on the basis of the ordinary convolution.
Further, the pixel point rendering module further comprises an MLP module,
the MLP module is used for carrying out multilayer perceptron learning on each selected uncertain pixel point (namely a key pixel point).
Examples
The model of the present invention is further described below with reference to schematic structural diagram of DeepLabV3+ neural network.
As shown in FIG. 1, the present invention is composed of three modules, an encoding module, a decoding module and a PointRent module.
The coding module firstly utilizes a backhaul network to extract features, and after the features are extracted, different convolutions are connected in parallel to further extract the features.
The different convolutions are 1 traditional convolution 1 × 1, 3 cavity convolutions with a cavity rate {6,12,18} and 1 deformable convolution, and the global average pooling is also applied to the input features to capture the context features with the global receptive field.
Further, the features extracted by the different convolutions are fused.
Furthermore, the fused features are input into a decoding module, and are expanded by 4 times after linear interpolation, so that feature fusion is carried out on the features and the convolutional layer of the size corresponding to the feature map of the backhaul module.
And further, inputting the fused features and the features of the convolution layer corresponding to the original backhaul into a PointRend network, and performing NLP (non line segment) perception machine learning on pixel points which are difficult to distinguish. And further improve the resolution learning of the network hard-to-distinguish points.
For comparison, the same Crack500 data set is used for learning and comparison with the current popular network, and the comparison result is shown in table 1.
TABLE 1 comparison of Crack semantic segmentation indexes of Crack500 asphalt pavement by different networks
As can be seen from Table 1, compared with popular network architectures, the model of the present invention has higher evaluation indexes in terms of accuracy (Acc), recall (Re), F1-score and mIoU, etc. than other network models. Particularly, the model greatly improves the recall rate (Re), so that the crack prediction performance of the whole network is improved.
In order to further visually evaluate the semantic segmentation performance of the network of the present invention, as shown in fig. 3, 4 and 5, the semantic segmentation results of different cracks input by the present invention and other network models are listed (fig. 3, 4 and 5 are comparison results after 3 different crack images are input in the embodiment, respectively).
Wherein, fig. 3a, 4a and 5a are respectively the images of the cracks to be detected input in 3 embodiments, fig. 3b, 4b and 5b are respectively the standard crack results in 3 embodiments, fig. 3c, 4c and 5c are respectively the crack semantic segmentation results of the FCN network model in 3 embodiments, fig. 3d, 4d and 5d are respectively the crack semantic segmentation results of the UNet network model in 3 embodiments, fig. 3e, 4e and 5e are respectively the crack semantic segmentation results of the denseas network model in 3 embodiments, fig. 3f, 4f and 5f are respectively the crack semantic segmentation results of the PSPNet network model in 3 embodiments, fig. 3g, 4g and 5g are respectively the crack semantic segmentation results of the Fast-SCNN network model in 3 embodiments, fig. 3h, 4h and 5h are respectively the crack semantic segmentation results of the DFANet network model in 3 embodiments, fig. 3i, 4i, and 5i are the fracture semantic segmentation results of the deplab v3 network model in 3 embodiments, respectively, fig. 3j, 4j, and 5j are the fracture semantic segmentation results of the deplab v3+ network model in 3 embodiments, respectively, and fig. 3k, 4k, and 5k are the fracture semantic segmentation results of the network model in 3 embodiments, respectively.
3-5, compared with the semantic segmentation results of the existing DeepLabV3+ network model and other network models, the semantic segmentation method provided by the improved DeepLabV3+ network model (FIGS. 3k, 4k and 5k) of the invention can more effectively extract crack information. As shown in fig. 5, for a crack with a complex structure or a crack with a relatively light color pair, the difference between the segmentation result (as shown in fig. 5 c-j) of other network models and the target result is large, and the semantic segmentation method (as shown in fig. 5k) provided by the present invention has a better segmentation effect on the crack.
Finally, it should be noted that: although the present invention has been described in detail with reference to the above embodiments, it should be understood by those skilled in the art that: modifications and equivalents may be made to the embodiments of the invention without departing from the spirit and scope of the invention, which is to be covered by the claims.
Claims (10)
1. A method for semantically segmenting cracks based on a DeepLabV3+ network model is characterized by comprising the following steps of:
coding the crack image to be detected by adopting parallel common convolution, cavity convolution and deformable convolution;
decoding the coded crack image to be detected through bilinear interpolation;
performing key pixel point reconstruction on the decoded crack image to be detected through pixel point rendering;
and classifying and predicting the reconstructed crack image to be detected to obtain a crack segmentation result.
2. The method for splitting the fracture semantics based on the deep lab v3+ network model according to claim 1, wherein before encoding the image of the fracture to be detected by using the parallel normal convolution, the hole convolution and the deformable convolution, the method further comprises:
firstly, carrying out primary feature extraction on a crack image to be detected by adopting a Backbone network, and then carrying out multi-scale convolution on the primary extracted features to complete further feature extraction.
3. The method for splitting the crack semantics of claim 1 based on the deep lab v3+ network model, wherein the deformable convolution comprises adding a learnable two-dimensional offset to each sampling position based on a common convolution.
4. The method for fracture semantic segmentation based on the DeepLabV3+ network model according to claim 1, wherein the method for coding the fracture image to be detected by parallel normal convolution, hole convolution and deformable convolution further comprises:
and after the features of the crack image to be detected are extracted by adopting parallel common convolution, cavity convolution and deformable convolution, fusing the extracted features of different convolutions.
5. The method for splitting the fracture semantics based on the DeepLabV3+ network model according to claim 1, wherein the performing of the key pixel point reconstruction on the decoded fracture image to be detected through pixel point rendering comprises:
and selecting N uncertain pixel points on the grid of the crack image to be detected after decoding, and reconstructing point-by-point characteristics on the selected uncertain pixel points by combining the low-layer characteristics and the high-layer characteristics of the crack image to be detected.
6. The method for splitting the crack semantics based on the deep lab v3+ network model according to claim 1, wherein after the key pixel points are reconstructed, the method further comprises:
and carrying out multilayer perceptron learning on the pixel points of the counterweight.
7. A fracture semantic segmentation system based on a DeepLabV3+ network model, which is characterized in that the fracture semantic segmentation method based on any one of claims 1-6 comprises the following steps:
the encoding module is used for encoding the crack image to be detected by adopting parallel common convolution, cavity convolution and deformable convolution;
the decoding module is used for decoding the coded crack image to be detected through bilinear interpolation;
the pixel point rendering module is used for performing key pixel point reconstruction on the decoded crack image to be detected through pixel point rendering;
and the fracture segmentation result output module is used for carrying out classification prediction on the reconstructed to-be-detected fracture image to obtain a fracture segmentation result.
8. The deep LabV3+ network model-based fracture semantic segmentation system according to claim 7, wherein the encoding module further comprises a Backbone network module,
the backhaul network module is used for performing primary feature extraction on the crack image to be detected, and then performing convolution on the primary extracted features to complete further feature extraction.
9. The deep LabV3+ network model-based fracture semantic segmentation system according to claim 7, wherein the encoding module comprises a deformable convolution module,
the deformable convolution module is used for adding a learnable two-dimensional offset to each sampling position on the basis of the ordinary convolution.
10. The dehp LabV3+ network model-based fracture semantic segmentation system according to claim 7, wherein the pixel point rendering module further comprises an MLP module,
the MLP module is used for carrying out multi-layer perceptron learning on the pixel points of the counterweight.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210434313.2A CN114998580A (en) | 2022-04-24 | 2022-04-24 | Fracture semantic segmentation method based on deep LabV3+ network model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210434313.2A CN114998580A (en) | 2022-04-24 | 2022-04-24 | Fracture semantic segmentation method based on deep LabV3+ network model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114998580A true CN114998580A (en) | 2022-09-02 |
Family
ID=83024605
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210434313.2A Pending CN114998580A (en) | 2022-04-24 | 2022-04-24 | Fracture semantic segmentation method based on deep LabV3+ network model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114998580A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117576405A (en) * | 2024-01-17 | 2024-02-20 | 深圳汇医必达医疗科技有限公司 | Tongue picture semantic segmentation method, device, equipment and medium |
-
2022
- 2022-04-24 CN CN202210434313.2A patent/CN114998580A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117576405A (en) * | 2024-01-17 | 2024-02-20 | 深圳汇医必达医疗科技有限公司 | Tongue picture semantic segmentation method, device, equipment and medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111461114A (en) | Multi-scale feature pyramid text detection method based on segmentation | |
| CN111415329A (en) | Workpiece surface defect detection method based on deep learning | |
| CN113034444A (en) | Pavement crack detection method based on MobileNet-PSPNet neural network model | |
| CN110826429A (en) | Scenic spot video-based method and system for automatically monitoring travel emergency | |
| CN113221688B (en) | Knife switch state identification method, device and storage medium | |
| CN116110036B (en) | Electric power nameplate information defect level judging method and device based on machine vision | |
| CN113392849A (en) | R-CNN-based complex pavement crack identification method | |
| CN114998580A (en) | Fracture semantic segmentation method based on deep LabV3+ network model | |
| CN116309485A (en) | Pavement crack detection method for improving UNet network structure | |
| CN114596316A (en) | Road image detail capturing method based on semantic segmentation | |
| CN115147383A (en) | Insulator state rapid detection method based on lightweight YOLOv5 model | |
| CN115761300A (en) | Method, system and detection device for dividing safety exit abnormity | |
| CN114612803A (en) | Transmission line insulator defect detection method for improving CenterNet | |
| CN113971666A (en) | Power transmission line machine inspection image self-adaptive identification method based on depth target detection | |
| CN117422990A (en) | Bridge structure classification and evaluation method based on machine learning | |
| Zhang et al. | Pavement crack detection based on deep learning | |
| CN115830302A (en) | Multi-scale feature extraction and fusion power distribution network equipment positioning identification method | |
| CN115170783A (en) | Expressway pavement crack detection method using street view image | |
| CN114581780A (en) | Tunnel surface crack detection method for improving U-Net network structure | |
| CN115861948A (en) | Lane line detection method, lane line detection device, lane line early warning method, lane line early warning system and medium | |
| CN112444522B (en) | Method for detecting defects of insulator string of power system | |
| Ashraf et al. | Efficient Pavement Crack Detection and Classification Using Custom YOLOv7 Model | |
| CN117315354B (en) | Insulator anomaly detection method based on multi-discriminant composite coding GAN network | |
| Liu | Road Crack Detection System Using Image Segmentation Algorithm | |
| CN115272808A (en) | Defect image generation method based on generation countermeasure network and image fusion facility |
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 |