CN112257627A - Overwater image data set expansion method - Google Patents
Overwater image data set expansion method Download PDFInfo
- Publication number
- CN112257627A CN112257627A CN202011179545.5A CN202011179545A CN112257627A CN 112257627 A CN112257627 A CN 112257627A CN 202011179545 A CN202011179545 A CN 202011179545A CN 112257627 A CN112257627 A CN 112257627A
- Authority
- CN
- China
- Prior art keywords
- image data
- water
- data set
- data
- aquatic
- 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.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- 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
- G06T3/00—Geometric image transformation in the plane of the image
- G06T3/40—Scaling the whole image or part thereof
- G06T3/4038—Scaling the whole image or part thereof for image mosaicing, i.e. plane images composed of plane sub-images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2200/00—Indexing scheme for image data processing or generation, in general
- G06T2200/32—Indexing scheme for image data processing or generation, in general involving image mosaicing
Abstract
The invention relates to a method for expanding an aquatic image data set, which comprises the following steps: s1, acquiring an existing overwater image data set as a first data subset; s2, performing expansion processing on the image data in the existing overwater image data set to obtain a second data subset; s3, constructing and training a generative confrontation network, and generating water simulation image data by adopting the generative confrontation network to obtain a third data subset; and S4, combining the first data subset, the second data subset and the third data subset into a water image data expansion data set after standardization processing. Compared with the prior art, the method can effectively increase the coverage area of the overwater image data set, so that the data set has a larger number of images with wider scenes, and the cost for constructing the data set is reduced.
Description
Technical Field
The invention relates to the technical field of image processing, in particular to an overwater image data set expansion method.
Background
In recent years, deep learning has been rapidly developed and widely applied to various fields with different sizes, and the model quality of a model constructed in the deep learning is closely related to the size of a data set. The large number and high quality of data sets can solve a series of problems such as overfitting in network training, and the accuracy of the data sets is highly dependent on the size and quality of the data sets. However, acquiring a large and high quality data set requires a large amount of labor and financial costs, which limits the effective application of deep learning in many contexts.
In the fields of target detection and automatic driving related to water, deep learning technology is widely applied. Compared with deep learning model data of a common scene, the deep learning model data is easier to obtain, such as ImageNet, CIFAR-100, VOC and other open-source public data sets, the overwater image data is more difficult to obtain, and open-source data meeting requirements are difficult to find for model training. And the quantity of the existing aquatic image data is far less than that of other life image data, which also increases the difficulty of generating the aquatic data set.
In the field of offshore automatic driving, the change speed of the offshore weather is high, and the marine weather is difficult to predict, so the influence of the weather on the picture quality must be considered in the offshore automatic driving and other technologies, and the data volume of the existing data set under the extreme weather condition is too small to meet the requirement of effective training of a model. Meanwhile, the lens of the water equipment under long-term operation is easy to crack or is polluted but cannot be maintained in time, so that the acquired image data is not clear enough, the image quality can be seriously influenced, and the model has problems in practical application. Therefore, the problem of insufficient aquatic image data sets in the deep learning of the related art becomes a technical problem to be solved urgently at present.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provide a method for expanding an aquatic image data set.
The purpose of the invention can be realized by the following technical scheme:
a method of aquatic image dataset expansion, the method comprising the steps of:
s1, acquiring an existing overwater image data set as a first data subset;
s2, performing expansion processing on the image data in the existing overwater image data set to obtain a second data subset;
s3, constructing and training a generative confrontation network, and generating water simulation image data by adopting the generative confrontation network to obtain a third data subset;
and S4, combining the first data subset, the second data subset and the third data subset into a water image data expansion data set after standardization processing.
Preferably, the method further comprises updating the extended data set of the water image data, and the specific method is as follows: the method comprises the steps of using a water image data expansion data set for training a water equipment application model, collecting water image real-time data, inputting the water image real-time data into the application model, judging whether the water image data expansion data set for training the application model needs to be updated or not according to an output error of the application model, and periodically adding the collected water image real-time data into the water image data expansion data set to finish updating if the water image data expansion data set needs to be updated.
Preferably, the real-time data of the aquatic image is classified when the real-time data of the aquatic image is added to the aquatic image data expansion data set, and is added to the corresponding category of the aquatic image data expansion data set.
Preferably, the manner of the expansion processing in step S2 includes processing the image data into images in different weather, and processing the image data into images in different influence factors of the photographing lens.
Preferably, the different weather conditions include heavy fog, cloudy days and heavy rain.
Preferably, the different shooting lens influencing factors comprise different lens cracks and lens pollution.
Preferably, the generative countermeasure network comprises:
a first generative confrontation network: the system is used for automatically generating water simulation image data according to random input noise;
second generation countermeasure networks: the method is used for automatically generating water simulation image data according to the water splicing image data, and the water splicing image data is obtained by splicing the existing water image data and the water obstacle picture.
Preferably, the first generative countermeasure network and the second generative countermeasure network have the same structure, and both comprise:
generating a network model, wherein the generated network model comprises an input layer, a convolution layer, an activation layer and a deconvolution layer which are sequentially cascaded, and the generated network model is used for generating water simulation image data according to the input image data;
and judging the network model, wherein the judging network model comprises a convolution layer, an activation layer and a full connection layer which are sequentially cascaded, and the judging network model is used for judging the truth of the water simulation image data generated by the generating network model and the input image data.
Preferably, when training the generative confrontation network: the existing overwater image data is used as the input of a first generating type confrontation network, the overwater splicing image data obtained by splicing the existing overwater image data and the water surface obstacle picture is used as the input of a second generating type confrontation network, and the first generating type confrontation network and the second generating type confrontation network are trained respectively;
when the aquatic simulation image data is generated by adopting the generating type countermeasure network: randomly input noise is used as the input of a first generation type countermeasure network, the existing overwater image data and the overwater spliced image data obtained by splicing the water surface obstacle pictures are used as the input of a second generation type countermeasure network, and the two generation type countermeasure networks are used for respectively obtaining overwater simulation image data.
Preferably, the acquisition mode of the water mosaic image data is as follows: firstly, separating the foreground and the background of a water surface obstacle picture, deleting the background, and then overlapping and splicing the obtained picture and the existing water image data to obtain water splicing image data.
Compared with the prior art, the invention has the following advantages:
(1) the invention provides a data set extension method for learning and training of an underwater deep learning model, and further can solve the difficult problems of insufficient data set quantity and low quality of the underwater technical work based on deep learning.
(2) The method aims at the special conditions that the change of the marine weather condition is complex, the lens is easy to corrode and have stains or cracks after the equipment is operated for a long time, and the image data is difficult to collect on water under the conditions, the data in the data set is expanded in a targeted mode, and the training effect of the deep learning model can be effectively improved.
(3) The invention utilizes the existing data set training generation model to automatically generate high-quality simulation data based on the generation type confrontation network, and is applied to the expansion of the water data set, and simultaneously provides a method for splicing and combining the water surface environment image and the water surface obstacle image and inputting the spliced and combined images into the generation type confrontation network, so that the obtained combined image effect is more real and natural, the cost for constructing the data set is reduced, and the number of the data set is increased.
(5) The invention combines data set expansion with an application model of the water equipment as an updating and supplementing method of the data set, a data acquisition system is arranged on the water equipment, the water data is acquired in different time periods, the data is acquired when the error in the model application is larger, the data set is updated regularly, so that the quality of the data set generated by the method is higher, for example, an application model of the water equipment such as an unmanned ship water surface garbage recognition model, namely, the water surface garbage recognition is carried out through a water picture shot on an unmanned ship, a corresponding unmanned ship water surface garbage recognition model needs to be trained through a water image data expansion data set, if the data in the water image data expansion data set is insufficient, the output error of the application model is larger, and at the moment, the water image real-time data obtained in real time through the water equipment is added into the water image data expansion data set for expansion, therefore, the application model is trained again at regular intervals, so that the quality of data contained in the water image data set is higher, and the actual application error of the application model is reduced.
Drawings
FIG. 1 is a block flow diagram of a method for aquatic image dataset expansion according to the present invention;
fig. 2 is a block diagram of a flow chart of updating an extended data set of water image data according to the present invention.
Detailed Description
The invention is described in detail below with reference to the figures and specific embodiments. Note that the following description of the embodiments is merely a substantial example, and the present invention is not intended to be limited to the application or the use thereof, and is not limited to the following embodiments.
Examples
As shown in fig. 1, a method for expanding a water image data set includes the following steps:
s1, acquiring an existing overwater image data set as a first data subset;
s2, performing expansion processing on the image data in the existing overwater image data set to obtain a second data subset;
s3, constructing and training a generative confrontation network, and generating water simulation image data by adopting the generative confrontation network to obtain a third data subset;
and S4, combining the first data subset, the second data subset and the third data subset into a water image data expansion data set after standardization processing.
As shown in fig. 2, the method further includes updating the extended data set of the aquatic image data, and the specific method includes: the method comprises the steps of using a water image data expansion data set for training a water equipment application model, collecting water image real-time data, inputting the water image real-time data into the application model, judging whether the water image data expansion data set for training the application model needs to be updated or not according to an output error of the application model, and periodically adding the collected water image real-time data into the water image data expansion data set to finish updating if the water image data expansion data set needs to be updated. And classifying the real-time data of the aquatic image when the real-time data of the aquatic image is added to the aquatic image data expansion data set, and adding the classified real-time data of the aquatic image to the corresponding classification of the aquatic image data expansion data set. Specifically, an automatic data acquisition system is installed on the water equipment and used for acquiring high-quality image information, and the acquired data is automatically classified by using a target recognition algorithm, so that the specific sub-steps of obtaining the expansion data are as follows:
firstly, a scoring rule is set for judging the quality of the acquired data, and the scoring rule mainly comprises the weighted sum of the image resolution, the error of the network obtained by inputting the trained model network and other influencing factors.
Then, a data acquisition time period is set in the system to enable the system to acquire data, and when the error of the system during working is larger than a set value, the data acquisition is carried out.
And finally, inputting the acquired data into a scoring system to screen high-quality photos, and automatically classifying the photos by using a target recognition algorithm to obtain an expanded data set.
It should be noted here that, the application model of the water equipment is, for example, an unmanned ship water surface garbage recognition model, that is, water surface garbage recognition is performed through a water picture shot on an unmanned ship, and the corresponding unmanned ship water surface garbage recognition model needs to be trained through a water image data expansion data set, if the data in the water image data expansion data set is insufficient, the output error of the application model is large, so that the water image real-time data obtained by the water equipment in real time is added to the water image data expansion data set for expansion, and the application model is trained again, so that the error of the application model is reduced.
The manner of the expansion processing in step S2 includes processing the image data into images in different weather, and processing the image data into images in different influence factors of the photographing lens. Different weather includes fog, cloudy days and heavy rain, and different shooting lens influencing factors include different lens cracks and lens pollution. Considering that the processing workload of all data in the data set is large, in the embodiment, fifteen percent of images in the data set are randomly adopted and are atomized by using a matlab picture atomization algorithm, the brightness, the hue and the saturation of the images are reduced to generate cloudy effect images, and a heavy rain image layer is added to the cloudy effect images to be processed into heavy rain images. And randomly selecting five percent of image superposition stains and crack image layers to simulate image data obtained under lens cracks and pollution.
The generative confrontation network comprises:
a first generative confrontation network: the system is used for automatically generating water simulation image data according to random input noise;
second generation countermeasure networks: the method is used for automatically generating water simulation image data according to the water splicing image data, and the water splicing image data is obtained by splicing the existing water image data and the water obstacle picture.
The first generative confrontation network and the second generative confrontation network have the same structure and both comprise:
generating a network model, wherein the generated network model comprises an input layer, a convolution layer, an activation layer and a deconvolution layer which are sequentially cascaded, and the generated network model is used for generating water simulation image data according to the input image data;
and judging the network model, wherein the judging network model comprises a convolution layer, an activation layer and a full connection layer which are sequentially cascaded, and the judging network model is used for judging the truth of the water simulation image data generated by the generating network model and the input image data.
Training the generative confrontation network: the existing overwater image data is used as the input of a first generation type countermeasure network, the overwater splicing image data obtained by splicing the existing overwater image data and a water surface obstacle image is used as the input of a second generation type countermeasure network, the first generation type countermeasure network and the second generation type countermeasure network are trained respectively, a generation network model and a discrimination network model are trained alternately according to a loss function in the training process, and finally, a balance state is reached, and the loss function is as follows:
wherein p (z) is an a priori noise signal input to the network model, g (z) is generative data for the generative model for which minimization is sought, D (g (z)) represents the discriminant model's discriminant result for the simulation data generated by the generative model, D (x) represents the discriminant model for which maximization is sought,
and Ez～p(z)Representing expected values of the distribution function, qdata(x)The distribution of real samples is shown, p (z) shows the noise distribution, G shows the generated model, and D shows the discriminant model.
When the aquatic simulation image data is generated by adopting the generating type countermeasure network: randomly input noise is used as the input of a first generation type countermeasure network, the existing overwater image data and the overwater spliced image data obtained by splicing the water surface obstacle pictures are used as the input of a second generation type countermeasure network, and the two generation type countermeasure networks are used for respectively obtaining overwater simulation image data.
The acquisition mode of the water splicing image data is as follows: firstly, separating the foreground and the background of a water surface obstacle picture, deleting the background, and then overlapping and splicing the obtained picture and the existing water image data to obtain water splicing image data.
The above embodiments are merely examples and do not limit the scope of the present invention. These embodiments may be implemented in other various manners, and various omissions, substitutions, and changes may be made without departing from the technical spirit of the present invention.
Claims (10)
1. A method for extending an aquatic image dataset, the method comprising the steps of:
s1, acquiring an existing overwater image data set as a first data subset;
s2, performing expansion processing on the image data in the existing overwater image data set to obtain a second data subset;
s3, constructing and training a generative confrontation network, and generating water simulation image data by adopting the generative confrontation network to obtain a third data subset;
and S4, combining the first data subset, the second data subset and the third data subset into a water image data expansion data set after standardization processing.
2. The method for expanding the aquatic image data set according to claim 1, further comprising updating the aquatic image data expansion data set by: the method comprises the steps of using a water image data expansion data set for training a water equipment application model, collecting water image real-time data, inputting the water image real-time data into the application model, judging whether the water image data expansion data set for training the application model needs to be updated or not according to an output error of the application model, and periodically adding the collected water image real-time data into the water image data expansion data set to finish updating if the water image data expansion data set needs to be updated.
3. The aquatic image data set expansion method according to claim 2, wherein the aquatic image real-time data is classified when the aquatic image real-time data is added to the aquatic image data expansion data set, and is added to a corresponding category of the aquatic image data expansion data set.
4. The method for expanding the aquatic image data set according to claim 1, wherein the expanding process in step S2 includes processing the image data into images under different weather conditions and processing the image data into images under different influence factors of the shooting lens.
5. The method of claim 4, wherein the different weather conditions include fog, cloudy days, and heavy rain.
6. The method according to claim 4, wherein the different camera lens influencing factors comprise different lens cracks and lens pollution.
7. The method of claim 1, wherein the generative confrontation network comprises:
a first generative confrontation network: the system is used for automatically generating water simulation image data according to random input noise;
second generation countermeasure networks: the method is used for automatically generating water simulation image data according to the water splicing image data, and the water splicing image data is obtained by splicing the existing water image data and the water obstacle picture.
8. The method according to claim 7, wherein the first generative confrontation network and the second generative confrontation network are identical in structure and each comprises:
generating a network model, wherein the generated network model comprises an input layer, a convolution layer, an activation layer and a deconvolution layer which are sequentially cascaded, and the generated network model is used for generating water simulation image data according to the input image data;
and judging the network model, wherein the judging network model comprises a convolution layer, an activation layer and a full connection layer which are sequentially cascaded, and the judging network model is used for judging the truth of the water simulation image data generated by the generating network model and the input image data.
9. The method of claim 8, wherein the image data set on water is expanded,
training the generative confrontation network: the existing overwater image data is used as the input of a first generating type confrontation network, the overwater splicing image data obtained by splicing the existing overwater image data and the water surface obstacle picture is used as the input of a second generating type confrontation network, and the first generating type confrontation network and the second generating type confrontation network are trained respectively;
when the aquatic simulation image data is generated by adopting the generating type countermeasure network: randomly input noise is used as the input of a first generation type countermeasure network, the existing overwater image data and the overwater spliced image data obtained by splicing the water surface obstacle pictures are used as the input of a second generation type countermeasure network, and the two generation type countermeasure networks are used for respectively obtaining overwater simulation image data.
10. The method for expanding the aquatic image data set according to any one of claims 7 to 9, wherein the aquatic mosaic image data is obtained by: firstly, separating the foreground and the background of a water surface obstacle picture, deleting the background, and then overlapping and splicing the obtained picture and the existing water image data to obtain water splicing image data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011179545.5A CN112257627B (en) | 2020-10-29 | 2020-10-29 | Overwater image data set expansion method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011179545.5A CN112257627B (en) | 2020-10-29 | 2020-10-29 | Overwater image data set expansion method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112257627A true CN112257627A (en) | 2021-01-22 |
| CN112257627B CN112257627B (en) | 2022-10-25 |
Family
ID=74261230
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011179545.5A Active CN112257627B (en) | 2020-10-29 | 2020-10-29 | Overwater image data set expansion method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112257627B (en) |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108052940A (en) * | 2017-12-17 | 2018-05-18 | 南京理工大学 | SAR remote sensing images waterborne target detection methods based on deep learning |
| CN108205659A (en) * | 2017-11-30 | 2018-06-26 | 深圳市深网视界科技有限公司 | Face occluder removes and its method, equipment and the medium of model construction |
| CN108875511A (en) * | 2017-12-01 | 2018-11-23 | 北京迈格威科技有限公司 | Method, apparatus, system and the computer storage medium that image generates |
| CN109598241A (en) * | 2018-12-05 | 2019-04-09 | 武汉大学 | Satellite image marine vessel recognition methods based on Faster R-CNN |
| CN110458019A (en) * | 2019-07-10 | 2019-11-15 | 哈尔滨工程大学 | The waterborne target detection method for excluding inverted image interference under rare cognition sample conditions |
| CN110728628A (en) * | 2019-08-30 | 2020-01-24 | 南京航空航天大学 | Face de-occlusion method for generating confrontation network based on condition |
| CN111145116A (en) * | 2019-12-23 | 2020-05-12 | 哈尔滨工程大学 | Sea surface rainy day image sample augmentation method based on generation of countermeasure network |
| CN111833277A (en) * | 2020-07-27 | 2020-10-27 | 大连海事大学 | Marine image defogging method with non-paired multi-scale hybrid coding and decoding structure |
-
2020
- 2020-10-29 CN CN202011179545.5A patent/CN112257627B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108205659A (en) * | 2017-11-30 | 2018-06-26 | 深圳市深网视界科技有限公司 | Face occluder removes and its method, equipment and the medium of model construction |
| CN108875511A (en) * | 2017-12-01 | 2018-11-23 | 北京迈格威科技有限公司 | Method, apparatus, system and the computer storage medium that image generates |
| CN108052940A (en) * | 2017-12-17 | 2018-05-18 | 南京理工大学 | SAR remote sensing images waterborne target detection methods based on deep learning |
| CN109598241A (en) * | 2018-12-05 | 2019-04-09 | 武汉大学 | Satellite image marine vessel recognition methods based on Faster R-CNN |
| CN110458019A (en) * | 2019-07-10 | 2019-11-15 | 哈尔滨工程大学 | The waterborne target detection method for excluding inverted image interference under rare cognition sample conditions |
| CN110728628A (en) * | 2019-08-30 | 2020-01-24 | 南京航空航天大学 | Face de-occlusion method for generating confrontation network based on condition |
| CN111145116A (en) * | 2019-12-23 | 2020-05-12 | 哈尔滨工程大学 | Sea surface rainy day image sample augmentation method based on generation of countermeasure network |
| CN111833277A (en) * | 2020-07-27 | 2020-10-27 | 大连海事大学 | Marine image defogging method with non-paired multi-scale hybrid coding and decoding structure |
Non-Patent Citations (1)
| Title |
|---|
| 王海文，邱晓晖: "一种基于生成式对抗网络的图像数据扩充方法", 《计算机技术与发展》 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112257627B (en) | 2022-10-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Hu et al. | Depth-attentional features for single-image rain removal | |
| CN111310862B (en) | Image enhancement-based deep neural network license plate positioning method in complex environment | |
| CN110570371B (en) | Image defogging method based on multi-scale residual error learning | |
| Fan et al. | Dual refinement underwater object detection network | |
| CN108537824B (en) | Feature map enhanced network structure optimization method based on alternating deconvolution and convolution | |
| CN110599445A (en) | Target robust detection and defect identification method and device for power grid nut and pin | |
| CN111783551B (en) | Countermeasure sample defense method based on Bayesian convolutional neural network | |
| CN109492596B (en) | Pedestrian detection method and system based on K-means clustering and regional recommendation network | |
| CN110059539A (en) | A kind of natural scene text position detection method based on image segmentation | |
| CN111915525A (en) | Low-illumination image enhancement method based on improved depth separable generation countermeasure network | |
| CN110084837B (en) | Target detection and tracking method based on unmanned aerial vehicle video | |
| CN109410144A (en) | A kind of end-to-end image defogging processing method based on deep learning | |
| CN111260687B (en) | Aerial video target tracking method based on semantic perception network and related filtering | |
| CN110310305A (en) | A kind of method for tracking target and device based on BSSD detection and Kalman filtering | |
| CN111008979A (en) | Robust night image semantic segmentation method | |
| CN111223087A (en) | Automatic bridge crack detection method based on generation countermeasure network | |
| CN113284061A (en) | Underwater image enhancement method based on gradient network | |
| CN115861756A (en) | Earth background small target identification method based on cascade combination network | |
| CN115424072A (en) | Unmanned aerial vehicle defense method based on detection technology | |
| CN111914938A (en) | Image attribute classification and identification method based on full convolution two-branch network | |
| CN111598793A (en) | Method and system for defogging image of power transmission line and storage medium | |
| CN115115863A (en) | Water surface multi-scale target detection method, device and system and storage medium | |
| CN112734739A (en) | Visual building crack identification method based on attention mechanism and ResNet fusion | |
| CN112257627B (en) | Overwater image data set expansion method | |
| CN112348762A (en) | Single image rain removing method for generating confrontation network based on multi-scale fusion |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |