CN109740608A - A kind of image partition method based on deep learning - Google Patents
A kind of image partition method based on deep learning Download PDFInfo
- Publication number
- CN109740608A CN109740608A CN201811627300.7A CN201811627300A CN109740608A CN 109740608 A CN109740608 A CN 109740608A CN 201811627300 A CN201811627300 A CN 201811627300A CN 109740608 A CN109740608 A CN 109740608A
- Authority
- CN
- China
- Prior art keywords
- image
- data
- network
- depth
- feature
- 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
Landscapes
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The present invention discloses a kind of image partition method based on deep learning, first with image data enhancing, extractive technique, to data set dilatation;And considers subterranean depth information when Image Acquisition, carry out 5 folding processing.Improved sorter network is finally used, as decoder, the data of TGS open source to be used to carry out model training as encoder and improved FPN network structure.
Description
Technical field
The invention belongs to technical field of computer vision more particularly to some image procossings, image, semantic dividing method,
Deep learning image partition method etc..
Background technique
With the development of artificial intelligence technology, the application of computer vision is more universal.In computer vision application, figure
It is essential link as dividing, image, semantic is divided the foundation stone technology that can be described as image understanding, ground in medical image
Study carefully, there is very important meaning in geologic image research, automated driving system, the fields such as modernization industry.For example, big on earth
The area of amount oil and natural gas aggregation often forms huge salt deposit below earth's surface, and, these salt surface sediments,
Exist in the form of high-temperature liquid state in underground.Before exploitation, the position in order to determine them can carry out rigorous survey to geology
It visits.By seismic imaging technology, some salt deposits of underground, rock stratum can be reacted in image.And then by being ground to image
Study carefully the specific location to identify them.But different geology knots regrettably, are partitioned into the geologic image label of these professions
The specific location of structure is very difficult.In addition, these geologic images for being fed back by sound wave, generally require the people of profession
Member is labeled it.This also causes result to have very strong subjectivity, carries out 3D rendering to the image of seismic imaging and brings
Very big difficulty.More seriously, if it is determined that inaccuracy is blindly exploited, it may cause the outflow of salt deposit, spray, this will give
The driller and drilling equipment of Koolaj-Es Foldgazbanyaszati Vallalat bring potential danger, and cause huge economic loss.
Image, semantic segmentation is the very important research direction of computer vision field, with the development of deep learning, is divided
Generic task and semantic segmentation task have also obtained full progress, and semantic segmentation may be considered image based on pixel scale
Divide and carrys out task.
1985, Hinton etc. proposed Back Propagation Algorithm, and the training of neural network is made to become simple possible.Classifying
Aspect, 1998, LeNet5 indicated the real appearance of CNN.It is proposed AlexNet in 2012, achieves year ImageNet
Classify the champion to compete.2014, GoogLeNet used Inception structure, deepened web results not only, Er Qiebian
" width ".2015, He Kaiming etc. proposed depth residual error network ResNet, had used residual unit, and the network number of plies breaks through 1000.
2016, the DenseNet of yellow high proposition used dense connection, under the premise of reaching nicety of grading suitable with ResNet,
Parameter and calculation amount only have the half of ResNet.2017, (the Squeeze- that domestic automatic Pilot company Momenta is proposed
And-Excitation Networks, SENet) " feature recalibration " strategy is used, and the same year obtains ImageNet image
The champion of classification task.In terms of segmentation, 2014, the full convolutional network FCN of the propositions such as Long, so that convolutional neural networks are not
Need full articulamentum that can realize the classification of intensive pixel scale.The U-Net network proposed in 2015, has used coding
Device-decoder structure, and in encoder to quick connection is introduced between decoder, to preferably restore the details of object
Information, the segmentation task especially on medical image, achieves preferable achievement.2017, Tsung-Yi Lin et al. was mentioned
Feature pyramid network (Feature Pyramid Networks, FPN) out, while utilizing low-level feature high-resolution and height
The high semantic information of layer feature, the characteristic information by merging these different layers make prediction achieve preferable effect.These bases
In the sorter network and segmentation network of deep learning, preferable achievement is achieved on natural image and on medical image.So
And segmentation task is done with these networks in geology salt deposit image domains, it is difficult to obtain preferable effect, main cause is ground
Matter image is influenced by depth of stratum, and the geology imaging results of different depth are also inconsistent, and followed by image data is few, to deep
Degree e-learning brings difficulty, and then causes segmentation effect low.
Summary of the invention
The technical problem to be solved by the present invention is to, a kind of geology salt deposit image partition method based on deep learning is provided,
First with image data enhancing, extractive technique, to data set dilatation.Finally using sorter network as encoder, FPN network
Structure carries out model training as decoder, using the data that TGS (geoscience data company advanced in the world) increases income.
The present invention proposes a kind of based on deep learning SENet154 and FPN towards a small amount of geology salt deposit image data
Network salt deposit dividing method.In data preprocessing phase, using image enhancement technique, to original image carry out left and right overturning, with
Machine changes the modes such as brightness, to achieve the purpose that carry out dilatation to raw data set.Then according to the underground of pictorial data representation
Depth distribution situation carries out data extraction.According to geology depth, divided 5 sections data, each section according to area into
Row subdivision, that is, carry out the training of 5 foldings.The data that can make to obtain every time are done so in subterranean depth distribution, are similar to normal state point
Cloth, the generalization ability of further lift scheme.In the design of network structure, using the U- for being similar to encoder-decoder
Net network structure, is different from, in coding stage using the stronger depth network of classification capacity.In coding stage
The depth sorting network that can be used has SENet154, SE-Resnet101, Resnet152 etc..Using deeper sorter network
Make every effort to practise from geology salt tomographic image middle school to more information, in addition, at the beginning of carrying out parameter to these Web vector graphic pre-training models
Beginningization accelerates the study of the low semantic information in image, and then acceleration model is restrained.FPN network, FPN are used in decoding stage
Network utilizes the high semantic information of low-level feature high-resolution and high-level characteristic simultaneously, and the feature by merging these different layers reaches
To the effect of prediction.And predict individually to be carried out on each fused characteristic layer, i.e., in each layer of feature, individually
It is predicted, is then carrying out information fusion.Generally speaking, by consider geology salt tomographic image in this factor of subterranean depth,
In each repetitive exercise, data is made to tend to just be distributed very much, is repeatedly reused using deeper network model and feature in addition
Network structure.So while geologic image data set is smaller, the information into image can also be learnt as far as possible, it can be with this
Efficiently and accurately realize the semantic segmentation task to geology salt deposit image.
For a realization above-mentioned purpose, the invention adopts the following technical scheme: in order to preferably realize entire method, it is preferred
Based on Python.On image enhancement, left and right overturning is carried out to raw image data using the library OpenCV, is changed at random
Then crop, translation, random Gaussian etc. again after degree of brightening, contrast, random scale export the data after image enhancement
Collection.And basic row herein, according to when Image Acquisition in underground depth (when seismic imaging, can be to the geology of underground different depth
It is imaged), it carries out dividing 5 sections, for the training of 5 foldings.Then it using open source deep learning frame PyTorch, realizes and compiles
The SENet154 sorter network in code stage, and borrow forefathers' trained SENet154 network parameter on ImageNet and carry out just
Beginningization.Equally, the FPN network of decoding stage is realized using PyTorch frame, so far completes building for model.By to data
Collection carries out the training of 5 foldings, and final choice goes out optimal model parameter.
A kind of geology salt deposit image partition method based on deep learning the following steps are included:
Step 1, the geology salt deposit image data set for obtaining related fields, and these data are cleaned.
Step 2, using image enhancement technique, enhancing processing is carried out to initial data, and increase the quantity of sample, and abundant
Data content.
Step 3, according to data set, in acquisition, subterranean depth carries out statistics and is divided into 5 sections, and in each section
5 equal portions are further divided into according to area.
Step 4, model buildings, coding stage first choice SENet154 network, decoding stage use FPN network.
Step 5, the 5 parts of data sets obtained according to step 3 carry out the training of 5 foldings, and vote and select optimal result.
Preferably, step 2 specifically includes the following steps:
Step 2.1, data shape enhancing, mark original image and its mask, according to a certain percentage to its length and width into
Row scaling, and carry out mirror image (left and right overturning).Translation is carried out to original image and its mask mark, is mended using edge pixel
Flush the image-region for moving and generating;
Step 2.2, the enhancing of data space domain.Lesser degree of random change brightness and contrast, guarantee the object of image
Meaning is managed, being found to be effective data by experiment enhances.Median filtering, gaussian filtering etc.;
Preferably, step 3 specifically includes the following steps:
Step 3.1, the principle being imaged according to geologic image, the image of different depth acquisition have certain in spatial domain
Data set is divided into five parts according to geology depth by correlation;
Step 3.2, the generalization ability in order to improve depth model divide step 3.1 to the data for 5 parts, in every portion again
It is divided into 5 parts by the area of salt deposit area marking;
The data of step 3.3, in summary two steps make each folding all comprising five depth intervals and 5 salt deposits
The data in region area section;
Preferably, step 4 specifically includes the following steps:
The coding stage of step 4.1, depth model uses SENet154 as basic structure.SENet154 has used SE-
Block structure, prominent validity feature, and the deep network number of plies, can provide better feature;
Step 4.2, the design of decoding stage modify routine FPN network, the semantic segmentation task for this geologic image
In.The feature that multiresolution has been merged using FPN improves for the segmentation tool of zonule salt deposit;
Step 4.3, on the basis of FPN, introduce hypercolumn module, further merged the spy of multiresolution
Sign;
Step 4.4 is eventually adding global average pond layer and classification head, in segmentation network, introducing point in encoder
Class auxiliary loss;
Step 4.5, in the level of each resolution ratio of decoder, introduce segmentation auxiliary loss, further adjust each level
The training of parameter;
Opposite with the prior art, the present invention has following clear superiority:
Deep learning model generally requires a large amount of data set and goes to train, and seeks best model with this, and, geology salt deposit
Image data set is smaller compared to scale with respect to data sets such as other natural images, medical images, and method of the invention passes through consideration
Geology salt tomographic image makes trained data be similar to just be distributed very much in the depth factor of underground, and the model for obtaining training more connects
It is bordering on nature situation, increases the generalization ability of model.Secondly in the coding stage of model, the deeper sorter network of use, example
Such as 154 layers of SENet network, in addition, it is logical to get each feature by way of study using " feature recalibration " strategy
Then the significance level in road goes to promote useful feature and inhibits the spy little to current task use according to this significance level
It levies, explicitly the relation of interdependence between Modelling feature channel.It, can by loading pre-training model in parameter initialization
To reach the low semantic information of Fast Learning, accelerate the convergence of model.Decoding stage, further by more points of the FPN network integration
The feature of resolution is significantly improved for the segmentation tool of zonule salt deposit, and FPN structure is effectively reduced decoding stage
Parameter amount, reduce the use of video memory, improve training speed.For generally, method of the invention can be relatively efficiently more quasi-
True is split geology salt deposit image.
Detailed description of the invention:
Fig. 1 is the flow chart of method involved in the present invention;
Fig. 2 is that 5 foldings of the present invention in data handling statistically analyze figure;
Fig. 3 is the semantic segmentation network structure that the present invention designs;
Fig. 4 is SENet Principles of Network figure according to the present invention;
Fig. 5 is FPN Principles of Network figure according to the present invention;
Specific embodiment
Yi Xiajiehejutishishili,Bing Canzhaofutu,Dui Benfamingjinyibuxiangxishuoming.
Hardware device used in the present invention has PC machine 1,1080 video card 1;
As shown in Figure 1, the present invention provides a kind of geology salt deposit image partition method based on deep learning, specifically include with
Lower step:
Step 1, the geology salt deposit image data set of related fields is obtained, and first time cleaning (example is carried out to these data
Such as, dirty data is deleted).
Step 2, using image enhancement technique, enhancing processing is carried out to initial data, increases the quantity of sample with this, and
The content of abundant data collection.
Step 2.1, form enhances, and marks to original image and its mask, contracts according to a certain percentage to its length and width
It puts, then intercepts out the size of semantic segmentation network needs, original sample is distorted to a certain extent, and is predicting
When, it can merge multiple dimensioned;
Step 2.2, form enhances, and marks to original image and its mask, carries out mirror image processing.Because geologic image reflects
Practical geological state under locality has certain depth, there is very much an actual meaning.In addition, translation is carried out,
In, the image-region that the translation of edge pixel polishing generates can be used;
Step 2.3, image space domain enhances the experiment has found that lesser degree of random change brightness and contrast, it can be with
Guarantee the physical significance of image, enhances for effective data;
Step 3, according to data set, in acquisition, subterranean depth carries out statistics and is divided into 5 sections, and in each section
It carries out being randomly divided into 5 equal portions.
Step 3.1, the principle being imaged according to geologic image, the image of different depth acquisition have certain in spatial domain
Correlation, so data set is first divided into five equal portions according to the depth in underground;
Step 3.2, in order to improve the generalization ability of depth model, divide step 3.1 to the data for 5 parts, in every portion again
It is divided into 5 parts by the area of salt deposit area marking.
Step 3.3, in summary two steps make each folding all comprising five depth intervals and 5 salt deposit area surfaces
The data in product section.
As shown in Fig. 2, statisticalling analyze figure for 5 foldings of the present invention in data handling, abscissa indicates training set and test
Collect the subterranean depth of acquisition, ordinate indicates the frequency that a certain depth occurs, and obtains data according to 5 folding of distribution situation, makes entirety
Distribution is similar to just be distributed very much.
Step 4, model buildings, coding stage first choice SENet154 network, decoding stage use FPN network.
As shown in figure 3, be the structure chart of the semantic segmentation network entirely based on geology salt tomographic image, it is whole by two big modules
Composition, coding module and decoder module.
Step 4.1, the coding stage of depth model uses SENet154 as basic structure.SENet has used SE-
Block structure, prominent validity feature, and the deep network number of plies, can provide better feature.
As shown in figure 4, the SE-block structure principle chart of SENet.SENet Web vector graphic " Squeeze " and
A kind of completely new " feature recalibration " strategy is realized in " Excitation " two operations.That is, being obtained automatically by way of study
The significance level in each feature channel is got, then go to promote useful feature according to this significance level and is inhibited to as predecessor
The little feature of use of being engaged in.
Step 4.2, the design of decoding stage, using the FPN network structure with fusion multilayer feature ability, for current
Semantic segmentation task in.FPN network can merge the feature of multiresolution, have centainly for the segmentation of zonule salt deposit
It helps, furthermore FPN structure can effectively reduce the parameter amount of decoding stage and the usage amount of video memory.
Step 4.3, in addition, introducing hypercolumn module in FPN network, multiresolution has further been merged
Feature.
Step 4.4, finally, be eventually adding global average pond layer and classifier in encoder draw in segmentation network
Enter classification auxiliary loss.In the level of each resolution ratio of decoder, segmentation auxiliary loss is introduced, further adjusts each level ginseng
Several training.
It is FPN Principles of Network figure according to the present invention shown in Fig. 5.It can be seen from the figure that FPN network is using a kind of
Then first bottom-up and then top-down structure carries out horizontal line attended operation, has not only merged the multiresolution in bottom
Feature, also study arrived high semantic information in deep layer.Furthermore FPN structure effectively reduce decoding stage parameter amount and
The use of video memory improves training speed.
Step 5, the training of 5 foldings is carried out, and votes and exports optimal result.
Above embodiments are only exemplary embodiment of the present invention, are not used in the limitation present invention, protection scope of the present invention
It is defined by the claims.Those skilled in the art can within the spirit and scope of the present invention make respectively the present invention
Kind modification or equivalent replacement, this modification or equivalent replacement also should be regarded as being within the scope of the present invention.
Claims (4)
1. a kind of image partition method based on deep learning, purpose is solving a small amount of geologic image segmentation problem, including
Following steps:
Step 1 obtains geology salt deposit image data set, and cleans to these data;
Step 2, using image enhancement technique, enhancing processing is carried out to initial data;
Step 3, according to data set acquisition when subterranean depth carry out statistics be divided into 5 sections, and in each section according to
Area is further divided into 5 equal portions;
Step 4, model buildings, coding stage first choice SENet154 network, decoding stage use FPN network;
Step 5, the 5 parts of data sets obtained according to step 3 carry out the training of 5 foldings, and vote and select optimal result.
2. as described in claim 1 based on the image partition method of deep learning, which is characterized in that step 2 specifically: data
Form enhancing, marks original image and its mask, zooms in and out according to predetermined ratio to its length and width, and carries out mirror image;To original
Beginning image and its mask mark carry out translation, the image-region generated using the translation of edge pixel polishing.
3. as described in claim 1 based on the image partition method of deep learning, which is characterized in that step 3 specifically include with
Lower step:
The image of step 3.1, the principle being imaged according to geologic image, different depth acquisition has certain correlation in spatial domain
Property, data set is divided into five parts according to geology depth;
Step 3.1 is divided the data for 5 parts by step 3.2, is divided into 5 parts by the area of salt deposit area marking again in every portion;
The data of step 3.3, in summary two steps make each folding all comprising five depth intervals and 5 salt layer regions
The data in area section.
4. as described in claim 1 based on the image partition method of deep learning, which is characterized in that step 4 specifically include with
Lower step:
The coding stage of step 4.1, depth model uses SENet154 as basic structure.SENet154 has used SE-
Block structure, prominent validity feature, and the deep network number of plies, can provide better feature;
Step 4.2, the design of decoding stage modify routine FPN network, in the semantic segmentation task of this geologic image.
The feature that multiresolution has been merged using FPN improves for the segmentation tool of zonule salt deposit;
Step 4.3, on the basis of FPN, introduce hypercolumn module, further merged the feature of multiresolution;
Step 4.4 is eventually adding global average pond layer and classification head in encoder, and in segmentation network, introducing classification is auxiliary
Help loss;
Step 4.5, in the level of each resolution ratio of decoder, introduce segmentation auxiliary loss, further adjust each level parameter
Training.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811627300.7A CN109740608B (en) | 2018-12-28 | 2018-12-28 | Image segmentation method based on deep learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811627300.7A CN109740608B (en) | 2018-12-28 | 2018-12-28 | Image segmentation method based on deep learning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109740608A true CN109740608A (en) | 2019-05-10 |
CN109740608B CN109740608B (en) | 2021-02-02 |
Family
ID=66361988
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811627300.7A Active CN109740608B (en) | 2018-12-28 | 2018-12-28 | Image segmentation method based on deep learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109740608B (en) |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110189342A (en) * | 2019-06-27 | 2019-08-30 | 中国科学技术大学 | Glioma region automatic division method |
CN110276402A (en) * | 2019-06-25 | 2019-09-24 | 北京工业大学 | A kind of salt body recognition methods based on the enhancing of deep learning semanteme boundary |
CN110443801A (en) * | 2019-08-23 | 2019-11-12 | 电子科技大学 | A kind of salt dome recognition methods based on improvement AlexNet |
CN110517278A (en) * | 2019-08-07 | 2019-11-29 | 北京旷视科技有限公司 | Image segmentation and the training method of image segmentation network, device and computer equipment |
CN110751011A (en) * | 2019-05-23 | 2020-02-04 | 北京嘀嘀无限科技发展有限公司 | Driving safety detection method, driving safety detection device and vehicle-mounted terminal |
CN111340064A (en) * | 2020-02-10 | 2020-06-26 | 中国石油大学(华东) | Hyperspectral image classification method based on high-low order information fusion |
CN112541926A (en) * | 2020-12-15 | 2021-03-23 | 福州大学 | Ambiguous pixel optimization segmentation method based on improved FCN and Densenet |
CN113158802A (en) * | 2021-03-22 | 2021-07-23 | 安徽理工大学 | Smart scene segmentation technique |
CN113392875A (en) * | 2021-05-20 | 2021-09-14 | 广东工业大学 | Method, system and equipment for classifying fine granularity of image |
CN113657480A (en) * | 2021-08-13 | 2021-11-16 | 江南大学 | Clothing analysis method based on feature fusion network model |
CN116977253A (en) * | 2022-12-29 | 2023-10-31 | 开立生物医疗科技(武汉)有限公司 | Cleanliness detection method and device for endoscope, electronic equipment and medium |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108921123A (en) * | 2018-07-17 | 2018-11-30 | 重庆科技学院 | A kind of face identification method based on double data enhancing |
CN109009110A (en) * | 2018-06-26 | 2018-12-18 | 东北大学 | Axillary lymphatic metastasis forecasting system based on MRI image |
CN109063710A (en) * | 2018-08-09 | 2018-12-21 | 成都信息工程大学 | Based on the pyramidal 3D CNN nasopharyngeal carcinoma dividing method of Analysis On Multi-scale Features |
-
2018
- 2018-12-28 CN CN201811627300.7A patent/CN109740608B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109009110A (en) * | 2018-06-26 | 2018-12-18 | 东北大学 | Axillary lymphatic metastasis forecasting system based on MRI image |
CN108921123A (en) * | 2018-07-17 | 2018-11-30 | 重庆科技学院 | A kind of face identification method based on double data enhancing |
CN109063710A (en) * | 2018-08-09 | 2018-12-21 | 成都信息工程大学 | Based on the pyramidal 3D CNN nasopharyngeal carcinoma dividing method of Analysis On Multi-scale Features |
Non-Patent Citations (2)
Title |
---|
BERNHARD MUELLER: ""Machine Learning-based Image Segmentation"", 《HTTPS://PUBLICATIONS.CMS.BGU.TUM.DE/THESES/2018_MUELLER_ML_BRAUN.PDF》 * |
LIANG-CHIEH CHEN等: ""MaskLab:Instance Segmetation by Refining Object Detection with Semantic and Direction Features"", 《2018 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION》 * |
Cited By (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110751011A (en) * | 2019-05-23 | 2020-02-04 | 北京嘀嘀无限科技发展有限公司 | Driving safety detection method, driving safety detection device and vehicle-mounted terminal |
CN110276402B (en) * | 2019-06-25 | 2021-06-11 | 北京工业大学 | Salt body identification method based on deep learning semantic boundary enhancement |
CN110276402A (en) * | 2019-06-25 | 2019-09-24 | 北京工业大学 | A kind of salt body recognition methods based on the enhancing of deep learning semanteme boundary |
CN110189342A (en) * | 2019-06-27 | 2019-08-30 | 中国科学技术大学 | Glioma region automatic division method |
CN110189342B (en) * | 2019-06-27 | 2022-03-01 | 中国科学技术大学 | Automatic segmentation method for brain glioma region |
CN110517278B (en) * | 2019-08-07 | 2022-04-29 | 北京旷视科技有限公司 | Image segmentation and training method and device of image segmentation network and computer equipment |
CN110517278A (en) * | 2019-08-07 | 2019-11-29 | 北京旷视科技有限公司 | Image segmentation and the training method of image segmentation network, device and computer equipment |
CN110443801A (en) * | 2019-08-23 | 2019-11-12 | 电子科技大学 | A kind of salt dome recognition methods based on improvement AlexNet |
CN111340064A (en) * | 2020-02-10 | 2020-06-26 | 中国石油大学(华东) | Hyperspectral image classification method based on high-low order information fusion |
CN112541926A (en) * | 2020-12-15 | 2021-03-23 | 福州大学 | Ambiguous pixel optimization segmentation method based on improved FCN and Densenet |
CN112541926B (en) * | 2020-12-15 | 2022-07-01 | 福州大学 | Ambiguous pixel optimization segmentation method based on improved FCN and Densenet |
CN113158802A (en) * | 2021-03-22 | 2021-07-23 | 安徽理工大学 | Smart scene segmentation technique |
CN113392875A (en) * | 2021-05-20 | 2021-09-14 | 广东工业大学 | Method, system and equipment for classifying fine granularity of image |
CN113657480A (en) * | 2021-08-13 | 2021-11-16 | 江南大学 | Clothing analysis method based on feature fusion network model |
CN116977253A (en) * | 2022-12-29 | 2023-10-31 | 开立生物医疗科技(武汉)有限公司 | Cleanliness detection method and device for endoscope, electronic equipment and medium |
CN116977253B (en) * | 2022-12-29 | 2024-03-19 | 开立生物医疗科技(武汉)有限公司 | Cleanliness detection method and device for endoscope, electronic equipment and medium |
Also Published As
Publication number | Publication date |
---|---|
CN109740608B (en) | 2021-02-02 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109740608A (en) | A kind of image partition method based on deep learning | |
Liu et al. | Seismic facies classification using supervised convolutional neural networks and semisupervised generative adversarial networks | |
Karimpouli et al. | Segmentation of digital rock images using deep convolutional autoencoder networks | |
US11119235B2 (en) | Automated seismic interpretation using fully convolutional neural networks | |
CN110276402B (en) | Salt body identification method based on deep learning semantic boundary enhancement | |
CN109709603A (en) | Seismic horizon identification and method for tracing, system | |
CN102918423A (en) | Method for seismic hydrocarbon system analysis | |
CN109671071A (en) | A kind of underground piping defect location and grade determination method based on deep learning | |
CN113703045B (en) | Seismic facies identification method based on lightweight network | |
dos Anjos et al. | Deep learning for lithological classification of carbonate rock micro-CT images | |
Patel et al. | The seismic analyzer: Interpreting and illustrating 2d seismic data | |
Jung et al. | Training image-based scenario modeling of fractured reservoirs for flow uncertainty quantification | |
Wang et al. | Deep-learning-based workflow for boundary and small target segmentation in digital rock images using UNet++ and IK-EBM | |
CN109727270A (en) | The movement mechanism and analysis of texture method and system of Cardiac Magnetic Resonance Images | |
Sen et al. | SaltNet: A production-scale deep learning pipeline for automated salt model building | |
Ravasi et al. | A joint inversion-segmentation approach to assisted seismic interpretation | |
Alfarhan et al. | Robust concurrent detection of salt domes and faults in seismic surveys using an improved UNet architecture | |
CN112698398A (en) | Deep fracture system space depicting method | |
Wu et al. | Variable seismic waveforms representation: Weak-supervised learning based seismic horizon picking | |
Wu et al. | Improved seismic well tie by integrating variable-size window resampling with well-tie net | |
Hou et al. | Reconstructing Three-dimensional geological structures by the Multiple-point statistics method coupled with a deep neural network: A case study of a metro station in Guangzhou, China | |
Yamada et al. | Instance segmentation of piled rock particles based on mask R-CNN | |
CN113688901B (en) | Reservoir discontinuous boundary line identification method based on expansion convolutional neural network | |
US11802984B2 (en) | Method for identifying subsurface features | |
Liu et al. | Using deep-learning to predict Dunham textures and depositional facies of carbonate rocks from thin sections |
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 |