CN114219778B - Data depth enhancement method based on WGAN-GP data generation and poisson fusion - Google Patents
Data depth enhancement method based on WGAN-GP data generation and poisson fusion Download PDFInfo
- Publication number
- CN114219778B CN114219778B CN202111482780.4A CN202111482780A CN114219778B CN 114219778 B CN114219778 B CN 114219778B CN 202111482780 A CN202111482780 A CN 202111482780A CN 114219778 B CN114219778 B CN 114219778B
- Authority
- CN
- China
- Prior art keywords
- data
- wgan
- network
- image
- gradient
- 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
- 230000004927 fusion Effects 0.000 title claims abstract description 34
- 238000000034 method Methods 0.000 title claims abstract description 32
- 201000010099 disease Diseases 0.000 claims abstract description 48
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 claims abstract description 48
- 238000012549 training Methods 0.000 claims abstract description 40
- 238000001514 detection method Methods 0.000 claims abstract description 24
- 238000009826 distribution Methods 0.000 claims abstract description 14
- 238000005516 engineering process Methods 0.000 claims abstract description 7
- 230000006870 function Effects 0.000 claims description 31
- 230000004913 activation Effects 0.000 claims description 10
- 238000013500 data storage Methods 0.000 claims description 8
- 238000005457 optimization Methods 0.000 claims description 7
- 238000005286 illumination Methods 0.000 claims description 6
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 5
- 238000013528 artificial neural network Methods 0.000 claims description 5
- 230000000694 effects Effects 0.000 claims description 5
- 238000005520 cutting process Methods 0.000 claims description 4
- 238000010586 diagram Methods 0.000 claims description 4
- 230000006872 improvement Effects 0.000 claims description 4
- 238000010606 normalization Methods 0.000 claims description 4
- 238000013527 convolutional neural network Methods 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 2
- 239000013598 vector Substances 0.000 claims description 2
- 238000012795 verification Methods 0.000 claims description 2
- 238000013135 deep learning Methods 0.000 description 5
- 238000012423 maintenance Methods 0.000 description 4
- 238000007500 overflow downdraw method Methods 0.000 description 4
- 230000015572 biosynthetic process Effects 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 238000003786 synthesis reaction Methods 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 238000007726 management method Methods 0.000 description 2
- 239000000203 mixture Substances 0.000 description 2
- HPTJABJPZMULFH-UHFFFAOYSA-N 12-[(Cyclohexylcarbamoyl)amino]dodecanoic acid Chemical compound OC(=O)CCCCCCCCCCCNC(=O)NC1CCCCC1 HPTJABJPZMULFH-UHFFFAOYSA-N 0.000 description 1
- 230000001133 acceleration Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 238000005096 rolling process Methods 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/50—Image enhancement or restoration by the use of more than one image, e.g. averaging, subtraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20212—Image combination
- G06T2207/20221—Image fusion; Image merging
Abstract
The invention discloses a data depth enhancement method based on WGAN-GP data generation and poisson fusion, wherein WGAN-GP is a generation countermeasure network with gradient punishment, and is a generation model based on game thought, and the generation model comprises two networks, namely a generation network G and a discrimination network D. The new training data is synthesized by inserting the road surface disease image generated by WGAN-GP into the disease-free road image. When the image is inserted, the image is inserted as truly as possible, so that the edge protrusion is avoided, and the object detection model only learns the edge characteristics of the object rather than the disease characteristic quantity. Compared with the traditional data enhancement method, the method has the advantages that the WGAN-GP data generation technology and the Poisson fusion technology are utilized to carry out depth enhancement on the data, the generated pictures are brought into the training set of the pavement disease detection model, and enough data volume is provided to enable the detection model to learn possible distribution, so that the intelligent pavement disease detection accuracy is improved.
Description
Technical Field
The invention belongs to the field of deep learning image processing, and relates to a data depth enhancement method based on WGAN-GP data generation and poisson fusion. The invention is suitable for intelligent detection of road diseases with lack of data samples or unbalanced categories in target detection.
Background
Due to the influence of temperature change, repeated rolling, improper periodic maintenance and other aspects, diseases with different degrees and different types can appear on the road surface, such as road surface cracks, pits and the like, so that the travelling comfort and the safety are seriously influenced, and necessary maintenance management is important for promoting the improvement of the service level of urban road facilities and ensuring the safe operation of roads. At present, road maintenance is excessively dependent on a manual inspection mode, and the problems of high cost, strong subjectivity, low efficiency and the like exist. Therefore, by adopting an advanced technical means, the automatic detection of road diseases is realized, and the method becomes a key place for road maintenance management.
In recent years, image recognition technology by means of artificial intelligence has been increasingly applied to road surface disease detection, and the core of the technology is image recognition technology of deep learning and convolutional neural networks. Deep learning is a data driven technique that requires a sufficient amount of data so that the model learns the possible distributions. However, it is difficult to obtain enough disease samples with balanced categories in the actual collection environment to ensure the training of the neural network. Conventional data enhancement methods include random scaling, contrast enhancement, random cropping, horizontal flipping, vertical flipping, etc., but the characteristics of road surface defects are often related to the color distribution and contrast of images, so conventional data enhancement methods may not be well suited for use in road surface pictures. And when the number of images is small, the operations such as random clipping and overturning are difficult to supplement the potential distribution rules implied by the data.
Aiming at the problems, the invention provides a data depth enhancement method combining WGAN-GP data generation and Poisson fusion, which is used for artificially generating road disease pictures under different shades and different illumination conditions and bringing the road disease pictures into a training set of a target detection model so as to improve the accuracy of intelligent detection of road diseases.
Disclosure of Invention
The technical scheme adopted by the invention is a data depth enhancement method based on combination of generating a countermeasure network WGAN-GP (Wasserstein Generative Adversarial Network with Gradient Penalty) and poisson fusion, which comprises two parts of WGAN-GP data generation and poisson fusion, wherein the implementation flow of the method is shown in a figure 1, and the specific steps are as follows:
step one: WGAN-GP data generation
WGAN-GP is a generation countermeasure Network with gradient penalty, and is a generation model based on game ideas, and the generation model includes two networks, namely a generation Network G (Generator Network) and a discrimination Network D (Network). As shown in fig. 2, the used countermeasure generation network framework adds randomly distributed noise (a) to an original image (b), generates data (c), inputs the generated data (c) to a generation network G, the generation network G extracts characteristic information (D) of pits and generates fitting data (e) to deceptively judge a network D, judges whether the network D judges whether the generated network outputs fitting data, trains a model by continuously and alternately feeding back, and finally, when a discriminator cannot distinguish whether a sample is true or false, the training of the generator model is completed.
The WGAN-GP data generation algorithm steps are as follows:
firstly, cutting out a road disease boundary frame from an original road picture as an input image generated by WGAN-GP data, wherein the cut-out picture is a pit under different shades under different illumination conditions. And then adjusting the cut picture to 80x80 pixel size to generate an HDF5 data storage file, and bringing the HDF5 data storage file into a WGAN-GP countermeasure generation network.
Secondly, the WGAN-GP network reads in the HDF5 data storage file and sets network training parameters including Batch training batch_size, picture pixel size, training band number Epoch, discriminator parameters and discriminator parameters.
Next, a generator and arbiter network structure is set. The WGAN-GP generating network G and the judging network D are both five-layer convolutional neural networks, the generating network consists of a ConvTrans2D deconvolution layer, a BN (BatchNorm) normalization layer, a Relu and a Tanh activation function, the judging network consists of a Conv2D convolution layer, a IN (InstanceNorm) normalization layer and a LeakyRelu activation function, and the network settings are shown in table 1.
Table 1 WGAN-GP challenge-generating network settings
The Relu activation function used in the invention is as shown in formula (1):
the LeakyRelu activation function used is as shown in equation (2):
the Tanh activation function used in the invention is as shown in formula (3):
in the formulas (1), (2) and (3), x is input data of each layer of neural network, and f (x) is output function of each layer of neural network.
Then, a loss function and an optimization function of the arbiter are set. For the loss function, the first two terms are the game losses of the arbiter, and the last term is the introduced Lipschitz constrained gradient penalty. For game losses, training to generate the countermeasure network maximizes the discrimination accuracy through training the discrimination network D while training the generation network G minimizes log (1-D (G (z))), i.e., training the generation network G and the discrimination network D is a very small and very large game with respect to the function V (G, D), the adopted game function is shown in formula (4):
the used arbiter loss function is as follows (5):
wherein: x is the true data distribution, maximizing D (x) means that the discriminator will learn as much as possible the true sample data distribution, P z (z) is the noise distribution of the input, G (z) represents the sample data distribution generated by the generator, D (G (z)) represents that the discriminator will try to discriminate the sample data generated by the generator isAnd if not, the method is true.
For the optimization function, model parameters are optimized using Adam's method. The Adam method is a simple random objective function gradient optimization algorithm with high calculation efficiency, and has two advantages in the aspects of processing sparse gradients and processing non-stationary targets.
Adam maintains the average square gradient v in the past t In the trend of exponential decay, there is a past gradient m of exponential decay trend at the same time t And has a flat minimum preference over the error plane. Then, the past attenuation average value and the past square gradient m are calculated t And v t Accordingly, as in formulas (6), (7):
m t ＝β 1 m t - 1 +(1-β 1 )g t (6)
wherein m is t And v t The first moment (mean) and the second moment (no center variance) of the gradient, respectively. The algorithm keeps the random gradient descent of the image matrix and the single learning rate, and updates and generates all weights in the countermeasure network.
Due to m t And v t Vectors initialized to 0, which are biased toward 0, are calculated as shown in equations (8), (9):
these t are then used and the update parameters see equation (10):
wherein beta is 1 Default value of 0.9, beta 2 Default for e is 0.999 and default for e is 10 -8 。
Finally, the discriminator calculates the difference gradient M between the real sample and the generated sample, sets Lipschitz constraint not to exceed M (M is set as 1 here), and realizes continuous updating optimization of the network through iteration, wherein the Loss constraint is maximally close to M.
Step two: poisson fusion
In the present invention, new training data is synthesized by inserting a road surface disease image generated by WGAN-GP into a disease-free road image. When the image is inserted, the image is inserted as truly as possible, so that the edge protrusion is avoided, and the object detection model only learns the edge characteristics of the object rather than the disease characteristic quantity. Poisson fusion (Poisson blend) is used as an image fusion method, which can better fuse a source image into a target scene, and the tone and illumination of the Poisson fusion (Poisson blend) are consistent with those of the target scene.
The poisson fusion method for solving the optimal value of the pixel based on the poisson equation can better fuse the source image and the target image while preserving the gradient information of the source image. According to the method, a poisson equation is solved according to the appointed boundary condition, and continuity on the gradient domain is achieved, so that seamless fusion at the boundary is achieved.
Poisson fusion algorithm steps:
firstly, preparing a source image source and a background image destination, wherein an original image is road pit diseases generated by WGAN-GP data, and a background image is a disease-free road image shot by a vehicle-mounted mobile phone.
Second, the setpoint P specifies where the source map is placed on the background map, where the setpoint P is where the center point of the source map is located.
Then, the gradient fields of the source map and the background map are calculated, and the gradient field of the fusion image is obtained. In the invention, a fusion algorithm is used, and the pixel value of an unknown region is calculated through a poisson equation by using a graph gradient and boundary conditions, as shown in a formula (11):
wherein,is a gradient operator; v is a gradient region in the image; f unknown function of target area; f (f) * A known function of the source region; omega is the target region; />Is the boundary between the source region and the target region.
Finally, the pixel values of the fused image are determined by solving the poisson equation.
Step three: effect verification
And verifying the improvement effect of the data depth enhancement technology on the target detection performance by using the Yolov5 target detection, and quantitatively evaluating the performance by using the F1-Score quantitative index.
The precision is defined as the proportion of correctly detected objects among all detected objects, as in equation (12):
recall refers to the proportion of correctly detected objects in all positive samples detected, as in formula (13):
where TP is the number of diseases detected correctly, FP is the number of non-diseases considered to be diseases, and FN is the number of diseases considered to be non-diseases.
f1-Score is a harmonic mean of precision and recall that combines the yield results of precision and recall, as in equation (14):
compared with the prior art, the invention has the following technical advantages. Compared with the traditional data enhancement method, the method can mine the implicit distribution rule of the data and generate road disease pictures under different shades and different illumination conditions. The generated pictures are brought into a training set of the pavement disease detection model, and enough data volume is provided to enable the detection model to learn possible distribution, so that the intelligent pavement disease detection accuracy is improved.
Drawings
Fig. 1 is a data depth enhancement calculation step.
Fig. 2 is a WGAN-GP algorithm architecture diagram.
Fig. 3 is a WGAN-GP training procedure.
FIG. 4 is a pothole disease generated based on WGAN-GP data.
FIG. 5 shows a pot hole disease data synthesis scheme.
Fig. 6 is a road picture synthesized using a Mixed fusion approach.
FIG. 7 is a photograph of a road synthesized using Normal fusion.
FIG. 8 is a comparison of training accuracy before and after data depth enhancement.
Fig. 9 is a comparison of training loss before and after data depth enhancement.
Detailed Description
The data set for training the pavement disease data enhancement algorithm comes from Global Road Damage Detection Challenge, is a high-definition pavement picture shot by a vehicle-mounted intelligent mobile phone, has the running speed of 40km/h, and is a non-overlapping image with the acquired size of 600 x 600 pixels.
Firstly, cutting out pit and groove boundary frames of pavement diseases from a data set as input images generated by data, cutting out 300 pit and groove diseases under different shielding objects under different illumination conditions, adjusting the pit and groove boundary frames to 80x80 pixels, generating an HDF5 data storage file, and bringing the HDF5 data storage file into a WGAN-GP network for training, wherein the training process is shown in figure 3. The WGAN-GP challenge-generating network parameter settings are shown in table 2, where K (Kernel size) represents the convolution Kernel size, n_out represents the number of convolution kernels, and S (Stride) represents the step size.
Table 2 WGAN-GP challenge-generating network parameter settings
After 100000 times of training on the generated countermeasure network, generating a generator.pkl weight file, namely, loading the generated weight after training the lot of pit slot pictures, generating 600 pit slot disease pictures, and randomly generating a result shown in figure 4.
And then, data synthesis is carried out on the generated pit diseases by using poisson fusion, and the flow is shown in figure 5. Taking 600 disease-free road pictures in the Japanese dataset as background pictures, and 600 road pit diseases generated by the WGAN-GP generating network trained in the first step as source pictures, wherein 1200 pictures are taken as data sets for data synthesis. Then, setting the placement point of pit slots in the background picture, wherein the position of the midpoint P is set at the position 3/5 of the height and 1/2 of the width of the picture. The road pictures synthesized by using the Normal fusion method are shown in FIG. 6, and the road pictures synthesized by using the Mixed fusion method are shown in FIG. 7. As can be seen from the figure, compared with the Normal fusion mode, the Mixed fusion mode can better fuse the background of the source image and the target image while preserving the gradient information of the source image. Therefore, the invention adopts a Mixed fusion mode.
Finally, the improvement effect of the data depth enhancement method combining WGAN-GP data generation and Poisson fusion on the performance of the target detection algorithm is verified, the synthesized 600 pavement pictures with pit diseases are added into Japanese original public data sets, and the front and rear detection accuracy is enhanced by using the Yolov5 target detection algorithm to compare the data. In the invention, experimental training is based on windows10 operating system, and a Pytorch framework is used, and the deep learning hardware environment is configured as follows: intel (R) Core (TM) i7-8850H CPU processor; memory: 64GB; display card: quadro P4000. The computer language Python 3.8 is loaded with the parallel computing architecture CUDA10.2 version and the CUDA-based deep learning GPU acceleration library CUDNN8.1 version. The training uses a random gradient descent algorithm with an attenuation coefficient of 0.937 to optimize the network, and the parameter settings of the learning rate, the batch training, the training rounds and the like are shown in table 3.
Table 3 experimental parameter settings
The training accuracy curve is shown in fig. 8, and the training loss curve is shown in fig. 9, wherein the solid line represents the training process after data enhancement, and the broken line represents the original data training process. As can be seen from FIG. 8, after the data depth enhancement, the accuracy of the target detection algorithm is obviously higher than the training accuracy of the original data, and the F1-Score is improved by 3.6%. As can be seen from fig. 9, after the data depth enhancement, the loss of the target detection algorithm is lower than the training loss of the original data, and the effectiveness of the method provided by the invention in solving the problems of unbalanced road surface disease types and fewer samples is verified.
Claims (4)
1. The data depth enhancement method based on WGAN-GP data generation and poisson fusion is characterized by comprising two major parts of WGAN-GP data generation and poisson fusion, and specifically comprises the following steps:
step one: WGAN-GP data generation;
firstly, cutting out a road disease boundary frame from an original road picture as an input image generated by WGAN-GP data, wherein the cut picture is a pit under different shielding objects under different illumination conditions; then adjusting the cut picture to 80x80 pixel size to generate an HDF5 data storage file, and bringing the HDF5 data storage file into a WGAN-GP countermeasure generation network;
secondly, the WGAN-GP network reads in the HDF5 data storage file and sets network training parameters, including Batch training batch_size, picture pixel size, training band number Epoch, discriminator parameters and discriminator parameters;
then, setting a network structure of a generator and a discriminator; the WGAN-GP generating network G and the judging network D are both five-layer convolutional neural networks, the generating network consists of a ConvTrans2D deconvolution layer, a BN normalization layer, a Relu and a Tanh activation function, and the judging network consists of a Conv2D convolution layer, an IN normalization layer and a LeakyRelu activation function;
finally, the discriminator calculates a difference gradient M between the real sample and the generated sample, the Lipschitz constraint is set to be not more than M, the Loss constraint is maximally close to M, and the continuous updating optimization of the network is realized through iteration;
step two: poisson fusion
The new training data is synthesized by inserting the road surface disease image generated by the WGAN-GP into the road image without the disease; the method for solving the optimal value of the pixel by poisson fusion introduced based on poisson equation is used for fusing the source image and the target image while keeping the gradient information of the source image; according to the method, a poisson equation is solved according to the appointed boundary condition, and continuity on the gradient domain is achieved, so that seamless fusion at the boundary is achieved.
2. The method for data depth enhancement based on WGAN-GP data generation and poisson fusion according to claim 1, wherein the Relu activation function used is as follows:
the LeakyRelu activation function used is as shown in equation (2):
the Tanh activation function used is as in equation (3):
in the formulas (1), (2) and (3), x is input data of each layer of neural network, and f (x) is an output function of each layer of neural network;
setting a loss function and an optimization function of the discriminator; for the loss function, the first two terms are game losses of the arbiter, and the last term is the introduced Lipschitz constraint gradient penalty; for game losses, training to generate the countermeasure network maximizes the discrimination accuracy through training the discrimination network D while training the generation network G minimizes log (1-D (G (z))), i.e., training the generation network G and the discrimination network D is a very small and very large game with respect to the function V (G, D), the adopted game function is shown in formula (4):
the used arbiter loss function is as follows (5):
wherein: x is the true data distribution, maximizing D (x) means that the discriminator will learn as much as possible the true sample data distribution, P z (z) is the noise distribution of the input, G (z) represents the sample data distribution generated by the generator, and D (G (z)) represents whether the discriminator will try to discriminate whether the sample data generated by the generator is true.
3. The data depth enhancement method based on WGAN-GP data generation and poisson fusion of claim 1, wherein model parameters are optimized for an optimization function using Adam method;
adam maintains the average square gradient v in the past t In the trend of exponential decay, there is a past gradient m of exponential decay trend at the same time t And has a flat minimum preference in the error plane; then, the past attenuation average value and the past square gradient m are calculated t And v t Accordingly, as in formulas (6), (7):
m t ＝β 1 m t-1 +(1-β 1 )g t (6)
wherein m is t And v t Estimated values of the first and second moments of the gradient, respectively; maintaining the random gradient descent of the image matrix, maintaining a single learning rate, and updating and generating all weights in the countermeasure network;
due to m t And v t Vectors initialized to 0, which are biased toward 0, are calculated as shown in equations (8), (9):
these t are then used and the update parameters see equation (10):
wherein beta is 1 Default value of 0.9, beta 2 Default for e is 0.999 and default for e is 10 -8 。
4. The data depth enhancement method based on WGAN-GP data generation and poisson fusion according to claim 1, wherein the poisson fusion algorithm step:
firstly, preparing a source image source and a background image destination, wherein an original image is road pit diseases generated by WGAN-GP data, and a background image is a disease-free road image shot by a vehicle-mounted mobile phone;
secondly, a set point P designates the position of the source diagram to be placed on the background diagram, wherein the P point is the position of the center point of the source diagram;
then, calculating gradient fields of the source image and the background image, and further obtaining a gradient field of the fusion image; in the invention, a fusion algorithm is used, and the pixel value of an unknown region is calculated through a poisson equation by using a graph gradient and boundary conditions, as shown in a formula (11):
wherein,is a gradient operator; v is a gradient region in the image; f unknown function of target area; f (f) * A known function of the source region; omega is the target region; />Is the boundary between the source region and the target region;
finally, determining the pixel value of the fusion image by solving a poisson equation;
step three: effect verification
Verifying the improvement effect of the data depth enhancement technology on the target detection performance by using Yolov5 target detection, and quantitatively evaluating the performance by using F1-Score quantitative indexes;
the precision is defined as the proportion of correctly detected objects among all detected objects, as in equation (12):
recall refers to the proportion of correctly detected objects in all positive samples detected, as in formula (13):
wherein TP is the number of diseases correctly detected, FP is the number of non-diseases considered to be diseases, and FN is the number of diseases considered to be non-diseases;
F1-Score is a harmonic mean of precision and recall, and the outcome of precision and recall is synthesized as shown in formula (14):
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111482780.4A CN114219778B (en) | 2021-12-07 | 2021-12-07 | Data depth enhancement method based on WGAN-GP data generation and poisson fusion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111482780.4A CN114219778B (en) | 2021-12-07 | 2021-12-07 | Data depth enhancement method based on WGAN-GP data generation and poisson fusion |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114219778A CN114219778A (en) | 2022-03-22 |
| CN114219778B true CN114219778B (en) | 2024-04-02 |
Family
ID=80699973
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111482780.4A Active CN114219778B (en) | 2021-12-07 | 2021-12-07 | Data depth enhancement method based on WGAN-GP data generation and poisson fusion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114219778B (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110827216A (en) * | 2019-10-23 | 2020-02-21 | 上海理工大学 | Multi-generator generation countermeasure network learning method for image denoising |
| CN112767344A (en) * | 2021-01-16 | 2021-05-07 | 北京工业大学 | Disease enhancement method based on vehicle-mounted camera shooting and coupling tradition and machine learning |
| CN113034411A (en) * | 2020-12-19 | 2021-06-25 | 北京工业大学 | Road disease picture enhancement method for resisting generation network by coupling traditional method and deep convolution |
| WO2021169292A1 (en) * | 2020-02-24 | 2021-09-02 | 上海理工大学 | Adversarial optimization method for training process of generative adversarial neural network |
-
2021
- 2021-12-07 CN CN202111482780.4A patent/CN114219778B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110827216A (en) * | 2019-10-23 | 2020-02-21 | 上海理工大学 | Multi-generator generation countermeasure network learning method for image denoising |
| WO2021169292A1 (en) * | 2020-02-24 | 2021-09-02 | 上海理工大学 | Adversarial optimization method for training process of generative adversarial neural network |
| CN113034411A (en) * | 2020-12-19 | 2021-06-25 | 北京工业大学 | Road disease picture enhancement method for resisting generation network by coupling traditional method and deep convolution |
| CN112767344A (en) * | 2021-01-16 | 2021-05-07 | 北京工业大学 | Disease enhancement method based on vehicle-mounted camera shooting and coupling tradition and machine learning |
Non-Patent Citations (1)
| Title |
|---|
| 肖儿良 ; 周莹 ; 简献忠 ; .迁移学习与GAN结合的医学图像融合模型.小型微型计算机系统.2020,(09),全文. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114219778A (en) | 2022-03-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109272048B (en) | Pattern recognition method based on deep convolutional neural network | |
| CN108648191B (en) | Pest image recognition method based on Bayesian width residual error neural network | |
| CN112801902B (en) | Traffic image defogging method based on improved generation countermeasure network | |
| CN109614968B (en) | License plate detection scene picture generation method based on multi-scale mixed image stylization | |
| CN108074244B (en) | Safe city traffic flow statistical method integrating deep learning and background difference method | |
| WO2021203795A1 (en) | Pancreas ct automatic segmentation method based on saliency dense connection expansion convolutional network | |
| CN112232450B (en) | Multi-stage comprehensive difficult sample mining method and target detection method | |
| CN113344849A (en) | Microemulsion head detection system based on YOLOv5 | |
| CN111489301B (en) | Image defogging method based on image depth information guide for migration learning | |
| CN111325711A (en) | Chromosome split-phase image quality evaluation method based on deep learning | |
| CN110245587B (en) | Optical remote sensing image target detection method based on Bayesian transfer learning | |
| CN110866872B (en) | Pavement crack image preprocessing intelligent selection method and device and electronic equipment | |
| US20230260628A1 (en) | Classification based on characterization analysis methods and systems | |
| CN109671055B (en) | Pulmonary nodule detection method and device | |
| CN111507183A (en) | Crowd counting method based on multi-scale density map fusion cavity convolution | |
| CN111833322B (en) | Garbage multi-target detection method based on improved YOLOv3 | |
| CN110363218A (en) | A kind of embryo's noninvasively estimating method and device | |
| CN114742758A (en) | Cell nucleus classification method in full-field digital slice histopathology picture | |
| CN114219778B (en) | Data depth enhancement method based on WGAN-GP data generation and poisson fusion | |
| CN114140469A (en) | Depth hierarchical image semantic segmentation method based on multilayer attention | |
| CN111275642B (en) | Low-illumination image enhancement method based on significant foreground content | |
| CN113420793A (en) | Improved convolutional neural network ResNeSt 50-based gastric ring cell carcinoma classification method | |
| CN115690100B (en) | Semi-supervised signal point detection model training method, signal point detection method and device | |
| CN115424032A (en) | Target detection model training method, device and computer readable storage medium | |
| CN114171119A (en) | Triple negative breast cancer tumor mutation load prediction method and system based on nuclear fraction and histopathology full-glass image |
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 | ||
| CB03 | Change of inventor or designer information |
Inventor after: Chen Ning Inventor after: Zhang Huiting Inventor after: Hou Yue Inventor after: Liu Zhuo Inventor after: Chen Yanyan Inventor before: Hou Yue Inventor before: Zhang Huiting Inventor before: Chen Ning Inventor before: Liu Zhuo Inventor before: Chen Yanyan |
|
| CB03 | Change of inventor or designer information | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |