CN116128825A - Meibomian gland morphology analysis method based on deep learning - Google Patents
Meibomian gland morphology analysis method based on deep learning Download PDFInfo
- Publication number
- CN116128825A CN116128825A CN202211720980.3A CN202211720980A CN116128825A CN 116128825 A CN116128825 A CN 116128825A CN 202211720980 A CN202211720980 A CN 202211720980A CN 116128825 A CN116128825 A CN 116128825A
- Authority
- CN
- China
- Prior art keywords
- meibomian
- morphology analysis
- model
- gland
- image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 210000004175 meibomian gland Anatomy 0.000 title claims abstract description 47
- 238000004458 analytical method Methods 0.000 title claims abstract description 41
- 238000013135 deep learning Methods 0.000 title claims abstract description 19
- 210000004907 gland Anatomy 0.000 claims abstract description 38
- 238000012549 training Methods 0.000 claims description 17
- 238000000605 extraction Methods 0.000 claims description 4
- 238000012545 processing Methods 0.000 claims description 4
- 238000013528 artificial neural network Methods 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 3
- 238000006243 chemical reaction Methods 0.000 claims description 2
- 230000002708 enhancing effect Effects 0.000 claims description 2
- 238000000034 method Methods 0.000 abstract description 8
- 230000011218 segmentation Effects 0.000 description 13
- 238000004364 calculation method Methods 0.000 description 9
- 238000013527 convolutional neural network Methods 0.000 description 6
- 208000003556 Dry Eye Syndromes Diseases 0.000 description 5
- 206010013774 Dry eye Diseases 0.000 description 5
- 230000000694 effects Effects 0.000 description 4
- 238000005070 sampling Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 206010065062 Meibomian gland dysfunction Diseases 0.000 description 2
- 238000013459 approach Methods 0.000 description 2
- 230000037237 body shape Effects 0.000 description 2
- 238000003759 clinical diagnosis Methods 0.000 description 2
- 238000003745 diagnosis Methods 0.000 description 2
- 230000000762 glandular Effects 0.000 description 2
- 238000002372 labelling Methods 0.000 description 2
- 230000015572 biosynthetic process Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 210000000744 eyelid Anatomy 0.000 description 1
- 230000033822 gland development Effects 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000005461 lubrication Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000012800 visualization Methods 0.000 description 1
Images
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
- G06T7/0012—Biomedical image inspection
-
- 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
- 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
- G06T2207/20084—Artificial neural networks [ANN]
-
- 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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30041—Eye; Retina; Ophthalmic
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Quality & Reliability (AREA)
- Radiology & Medical Imaging (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Life Sciences & Earth Sciences (AREA)
- Medical Informatics (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Processing (AREA)
- Eye Examination Apparatus (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a meibomian gland morphology analysis method based on deep learning, which comprises the following steps: the method comprises the steps that an eye surface image is enhanced through an image quality enhancement module to obtain an eye surface image; marking a meibomian range and a gland range on the eye surface image to obtain a meibomian coordinate and a gland coordinate; data enhancement is carried out by combining the meibomian coordinates and the gland coordinates with the eye surface image, and the enhanced meibomian image, the enhanced gland image and the enhanced eye surface image are trained to obtain a meibomian gland morphology analysis primary model; inputting a real picture GT in the data into a meibomian gland morphology analysis preliminary model to obtain a model prediction picture P; and (3) calculating the intersection ratio of the real picture GT and the gland area relative position of the model prediction picture P, optimizing the meibomian gland morphology analysis preliminary model through the intersection ratio to obtain a meibomian gland morphology analysis model, and analyzing the meibomian gland morphology through the meibomian gland morphology analysis model.
Description
Technical Field
The invention relates to the field of meibomian gland morphology analysis, in particular to a meibomian gland morphology analysis method based on deep learning.
Background
Dry eye is a common disease, which occurs when tears do not provide enough lubrication for eyes, and the detection rate of dry eye in China is about 6.1% -52.4% according to statistics, and the incidence rate of dry eye in China is kept high along with the increase of related dangerous factors. The main cause of dry eye is meibomian gland dysfunction (meibomian gland dysfunction, MGD), and in clinical diagnosis, analysis of meibomian gland morphology to diagnose and grade the extent of MGD is an important means of dry eye diagnosis.
The current clinical diagnosis means of MGD mainly relies on professional eye surface photographing analysis instruments, high-definition pictures are photographed through the instruments, traditional indexes such as white balance and saturation are adjusted, gland parts are marked manually, and various indexes are calculated, such as: uneven expansion results in deformation coefficient (DI), glandular Torsion (TI), etc. According to the method, various indexes are required to be marked and calculated manually, the workload is high, the subjective experience of marking personnel is relied on, errors are large, and the patients feel painful when the eyelids are shot, so that the imaging is easy to be fuzzy, and diagnosis is influenced. And the index between the single glands cannot be extracted and analyzed, and the related index of the development intensity of the gland picture cannot be accurately obtained.
Disclosure of Invention
The invention provides a meibomian gland morphology analysis method based on deep learning.
A meibomian gland morphology analysis method based on deep learning comprises the following steps:
1) Shooting an eye table picture through an eye table shooting instrument, and enhancing the eye table picture through a picture quality enhancement module to obtain an eye table image;
2) Marking a meibomian range and a gland range on the eye surface image to obtain a meibomian coordinate and a gland coordinate;
3) Data enhancement is carried out on the meibomian coordinates and the gland coordinates combined with the eye surface image obtained in the step 1) to obtain an enhanced meibomian image, an enhanced gland image and an enhanced eye surface image, and training is carried out on the enhanced meibomian image, the enhanced gland image and the enhanced eye surface image to obtain a meibomian gland morphology analysis preliminary model;
4) Inputting a real picture GT in the data into a meibomian gland morphology analysis preliminary model to obtain a model prediction picture P;
5) And (3) calculating the intersection ratio of the real picture GT and the gland area relative position of the model prediction picture P, optimizing the meibomian gland morphology analysis preliminary model through the intersection ratio to obtain a meibomian gland morphology analysis model, and analyzing the meibomian gland morphology through the meibomian gland morphology analysis model.
In step 1), the image quality enhancement module enhances the surface-eye image, which specifically includes:
the picture downsampling module is used for downsampling the eye surface picture;
the neural network convolution layer is used for carrying out convolution processing on the eye surface image after downsampling;
the 1D LUT color conversion module is used for carrying out color extraction on the eye table data obtained after convolution processing;
and the VGG network structure is used for carrying out color mapping on the color eye table data obtained after the color extraction to obtain an eye table image.
The LUT itself does not operate, but a series of input and output data are listed, and for R channels, there is r_out=lut (r_in), and for G, B channels, these data are corresponding to each other, so that the Gamma value, gray field, and contrast of the image can be changed by using 1DLUT, and the white field and black field can be redefined. In deep learning neural networks, 1DLUT is sampled discretely as a complete color transform function and embedded in the network structure as a fully connected operational layer. When training the quality enhancement module, sampling results
Stored in an output three-channel one-dimensional matrix, can be represented by input color coordinates +.>
Making a query, where N s Is the number of sampling coordinates along each of the three dimensions, +.>
Represents {0,1, … N s -1}. Thus in the complete 1D LUT colorA total of N3 s sampling points are defined on the transformation function. During reasoning, the sampled eye table data passes through the full connection layer, the input pixel searches the nearest sampling point according to the color of the input pixel, and the converted output of the input pixel is calculated through tri-linear interpolation. Predictive image output value T epsilon [0,1 ]]The 3×ns×ns can enhance picture quality by learning several basic 1D LUTs and fusing with image-dependent weights predicted from the downsampled input image by the CNN model.
In step 5), calculating the intersection ratio of the gland positions and the relative areas of the real picture GT and the model predicted picture P, specifically including:
In step 5), I ranges from [0-1 ].
In the step 5), optimizing the meibomian gland morphology analysis preliminary model through the intersection ratio to obtain the meibomian gland morphology analysis model, which specifically comprises the following steps:
optimizing the meibomian gland morphology analysis preliminary model through the intersection ratio, wherein the intersection ratio of the trained meibomian gland morphology analysis preliminary model is the maximum meibomian gland morphology analysis model.
Wherein the gland position intersection area operation of the predicted pictures P and GT, i.e., GT n p= Σ i,j GT (i,j) ∧P (i,j) The union area is calculated as GT U.P= Σ i,j GT (i,j) ∨P (i,j) The calculation of the cross-over ratio replaces the original classification index map by the calculation of the relative area of the gland pixel position, and in the training iteration process, the model with the best performance of gland segmentation is preferentially selected instead of the model with the highest classification accuracy. Compared with the prior art, the invention has the following advantages:
the meibomian gland body shape analysis capability provided by the invention completely depends on a deep learning algorithm, and the quality enhancement module can optimize the quality of the eye surface picture in a 1D color space, so that the picture quality is more in line with the viewing of human eyes, the gland characteristics are more obvious, manual labeling of personnel is not needed after model training is completed, and the meibomian gland body shape analysis capability is accurate and objective enough, and can accurately describe the meibomian range and the meibomian gland range.
The invention can calculate the index between single glands, can calculate regional index, and can calculate the gland development intensity related index according to the accurate numerical value of the color space of the photographed original image.
The invention can iterate the version, the more the data is accumulated, the stronger the model capability is, and the better the effect is.
Drawings
FIG. 1 is a schematic diagram of a training portion of the present invention.
Fig. 2 is a schematic diagram of the inferencing portion of the present invention.
Fig. 3 is a schematic diagram of a quality enhancement module according to the present invention.
Detailed Description
The invention aims at analyzing the meibomian gland morphology by a deep learning method and can be continuously and iteratively optimized along with the data quantity, so the method is mainly divided into two parts, training and reasoning.
As shown in fig. 1, the training part mainly provides training of a deep learning CNN model before reasoning, and includes five modules of data enhancement, picture quality enhancement, deep learning meibomian segmentation training, deep learning gland segmentation training and IOU calculation index. After the eye surface image shot by the ophthalmic equipment is obtained, the eye surface image with higher subjective quality and better glandular identification can be obtained through the image quality enhancement module, and then data are input into the data enhancement module to obtain training data which is at least ten times that of source data. Thereafter, two parts: 1. after the meibomian segmentation training module is input to obtain the optimal CNN model, the effect of the model is judged by calculating the IOU, and the training iteration can be continued until an ideal result is obtained. 2. After the gland segmentation training module is input to obtain the optimal CNN model, the IOU judgment model effect is calculated, and the iteration can be performed to an ideal result.
As shown in fig. 2, the reasoning part mainly comprises model reasoning and various index calculations after reasoning, including four modules of image quality enhancement, meibomian segmentation reasoning, gland segmentation reasoning, index calculation and output. The database eye surface images are input and then preprocessed through the quality enhancement module to obtain high-quality eye surface images, the meibomian images are input into the meibomian segmentation reasoning module to obtain meibomian images, the meibomian images are input into the gland segmentation reasoning module to obtain gland segmentation images, and finally the meibomian images and the gland images are input into index calculation to calculate various indexes and output results.
The training process inputs not less than ten pictures at a time.
The quality enhancement module is a subjective quality enhancement structure MGSE (Meibomian Gland Subjective Enhancement) provided by the invention aiming at meibomian glands, as shown in fig. 3), takes an image to be enhanced as input, takes a high-quality image as output, takes the output image and the high-quality image subjected to manual adjustment as target images to calculate MSE loss, and completes an end-to-end supervised learning process to realize subjective quality enhancement of the eye surface pictures. After inputting a picture, downsampling the CNN structure to fix the size of the image, extracting features to provide global understanding of the image, and fusing the color coefficient T { x ] by a full-connection layer based on 1DLUT R ,x G ,x B And finally, the VGG-8 structure color mapping outputs the processed high-quality image.
Deep learning meibomian segmentation into Mask-rcnn convolutional neural network modules, mask and BG of the image are separated by example segmentation, i.e., the meibomian portion and the ocular surface portion other than the meibomian, using res net50 as the classification network. The loss function is designed as l=lcls (meibomian and other eye surfaces) +lbox (circumscribed frame position of meibomian part) +lmask (two-dimensional position of meibomian part), and the segmentation task is completed by reducing the loss value through iterative learning.
The deep learning gland is divided into a Unet convolutional neural module, VGG16 is used as a backstone, characteristics are extracted in depth by utilizing characteristics of a VGG network, and the gland and the non-gland are classified according to thought of an encoder-decoder.
The IOU calculates the position relative area of the real picture GT and the model predicted picture P to do cross ratio calculation (0-1), namely:
the larger the number, the better the model prediction effect.
The model only needs to be iteratively upgraded, the eye table picture data can be fitted after training is finished, the model can be retrained to be iteratively upgraded, and the more the data is, the stronger the model capacity is.
The index calculation module can calculate the index of the single gland, and can output all the indexes related to the single gland in a self-defined way, such as: total variation, individual gland formation coefficients, etc.
And the index calculation module outputs a complete visualization result and can synthesize a result picture.
The reasoning process is a pipeline, manual labeling is not needed, and input and output are needed.
Implementing picture quality enhancement example:
referring to fig. 3, a certain number of high quality eye surface pictures (glands are clearly visible) are manually adjusted and selected in advance to be GT, and Input into an end-to-end model of Input-output, the MSE is made to approach an optimal value through iteration, so that the color distribution of a predictive picture Pred approaches to GT, the model is fixedly stored, and the pictures are directly Input as GT during reasoning, so that the enhanced Pred can be obtained.
Claims (5)
1. The meibomian gland morphology analysis method based on deep learning is characterized by comprising the following steps of:
1) Shooting an eye table picture through an eye table shooting instrument, and enhancing the eye table picture through a picture quality enhancement module to obtain an eye table image;
2) Marking a meibomian range and a gland range on the eye surface image to obtain a meibomian coordinate and a gland coordinate;
3) Data enhancement is carried out on the meibomian coordinates and the gland coordinates combined with the eye surface image obtained in the step 1) to obtain an enhanced meibomian image, an enhanced gland image and an enhanced eye surface image, and training is carried out on the enhanced meibomian image, the enhanced gland image and the enhanced eye surface image to obtain a meibomian gland morphology analysis preliminary model;
4) Inputting a real picture GT in the data into a meibomian gland morphology analysis preliminary model to obtain a model prediction picture P;
5) And (3) calculating the position relative area of the real picture GT and the model prediction picture P, optimizing a meibomian gland morphology analysis primary model through the intersection ratio to obtain a meibomian gland morphology analysis model, and analyzing the meibomian gland morphology through the meibomian gland morphology analysis model.
2. The deep learning-based meibomian gland morphology analysis method of claim 1, wherein in step 1), the image quality enhancement module enhances an eye surface image, specifically comprising:
the picture downsampling module is used for downsampling the eye surface picture;
the neural network convolution layer is used for carrying out convolution processing on the eye surface image after downsampling;
the 1D LUT color conversion module is used for carrying out color extraction on the eye table data obtained after convolution processing;
and the VGG network structure is used for carrying out color mapping on the color eye table data obtained after the color extraction to obtain an eye table image.
3. The deep learning-based meibomian gland morphology analysis method of claim 1, wherein in step 5), the position relative area of the real picture GT and the model predictive picture P is calculated by a cross ratio, specifically comprising:
4. The deep learning-based meibomian gland morphology analysis method of claim 3, wherein in step 5), I ranges from [0-1 ].
5. The deep learning-based meibomian gland morphology analysis method of claim 1, wherein in step 5), the meibomian gland morphology analysis model is obtained by optimizing the meibomian gland morphology analysis preliminary model by means of an intersection ratio, specifically comprising:
optimizing the meibomian gland morphology analysis preliminary model through the intersection ratio, wherein the intersection ratio of the trained meibomian gland morphology analysis preliminary model is the maximum meibomian gland morphology analysis model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211720980.3A CN116128825A (en) | 2022-12-30 | 2022-12-30 | Meibomian gland morphology analysis method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211720980.3A CN116128825A (en) | 2022-12-30 | 2022-12-30 | Meibomian gland morphology analysis method based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116128825A true CN116128825A (en) | 2023-05-16 |
Family
ID=86309510
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211720980.3A Pending CN116128825A (en) | 2022-12-30 | 2022-12-30 | Meibomian gland morphology analysis method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116128825A (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109087310A (en) * | 2018-07-24 | 2018-12-25 | 深圳大学 | Dividing method, system, storage medium and the intelligent terminal of Meibomian gland texture region |
| CN111127431A (en) * | 2019-12-24 | 2020-05-08 | 杭州求是创新健康科技有限公司 | Dry eye disease grading evaluation system based on regional self-adaptive multitask neural network |
| CN112885456A (en) * | 2021-01-20 | 2021-06-01 | 武汉爱尔眼科医院有限公司 | Meibomian gland quantitative analysis based on deep learning and application thereof in MGD diagnosis and treatment |
| CN113962978A (en) * | 2021-10-29 | 2022-01-21 | 北京富通东方科技有限公司 | Eye movement damage detection and film reading method and system |
| CN114343563A (en) * | 2021-12-31 | 2022-04-15 | 温州医科大学附属眼视光医院 | Method, device and system for assisting dry eye diagnosis and typing through multi-modal fusion |
-
2022
- 2022-12-30 CN CN202211720980.3A patent/CN116128825A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109087310A (en) * | 2018-07-24 | 2018-12-25 | 深圳大学 | Dividing method, system, storage medium and the intelligent terminal of Meibomian gland texture region |
| CN111127431A (en) * | 2019-12-24 | 2020-05-08 | 杭州求是创新健康科技有限公司 | Dry eye disease grading evaluation system based on regional self-adaptive multitask neural network |
| CN112885456A (en) * | 2021-01-20 | 2021-06-01 | 武汉爱尔眼科医院有限公司 | Meibomian gland quantitative analysis based on deep learning and application thereof in MGD diagnosis and treatment |
| CN113962978A (en) * | 2021-10-29 | 2022-01-21 | 北京富通东方科技有限公司 | Eye movement damage detection and film reading method and system |
| CN114343563A (en) * | 2021-12-31 | 2022-04-15 | 温州医科大学附属眼视光医院 | Method, device and system for assisting dry eye diagnosis and typing through multi-modal fusion |
Non-Patent Citations (2)
| Title |
|---|
| 朱敏颖 等: "基于卷积神经网络的睑板腺形态人工智能分析系统的构建", 《浙江医学》, vol. 43, no. 18, pages 1946 - 1952 * |
| 金阳: "《锂离子电池储能电站早期安全预警及防护 第1版》", 机械工业出版社, pages: 202 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Bhalla et al. | A fuzzy convolutional neural network for enhancing multi-focus image fusion | |
| CN110287846B (en) | Attention mechanism-based face key point detection method | |
| CN109255758A (en) | Image enchancing method based on full 1*1 convolutional neural networks | |
| CN111489324A (en) | Cervical cancer lesion diagnosis method fusing multi-modal prior pathology depth features | |
| CN110675462A (en) | Gray level image colorizing method based on convolutional neural network | |
| CN113420794B (en) | Binaryzation Faster R-CNN citrus disease and pest identification method based on deep learning | |
| CN106157249A (en) | Based on the embedded single image super-resolution rebuilding algorithm of optical flow method and sparse neighborhood | |
| CN118097372B (en) | Crop growth visual prediction method based on artificial intelligence | |
| CN113486894A (en) | Semantic segmentation method for satellite image feature component | |
| CN118379288B (en) | Embryo prokaryotic target counting method based on fuzzy rejection and multi-focus image fusion | |
| CN112508814A (en) | Image tone restoration type defogging enhancement method based on unmanned aerial vehicle at low altitude view angle | |
| CN114897742A (en) | Image restoration method with texture and structural features fused twice | |
| CN116543386A (en) | Agricultural pest image identification method based on convolutional neural network | |
| Zheng et al. | Overwater image dehazing via cycle-consistent generative adversarial network | |
| CN110992301A (en) | Gas contour identification method | |
| CN113810683A (en) | No-reference evaluation method for objectively evaluating underwater video quality | |
| CN113643297A (en) | Computer-aided age analysis method based on neural network | |
| CN111881924B (en) | Dark-light vehicle illumination identification method combining illumination invariance and short-exposure illumination enhancement | |
| CN113128517A (en) | Tone mapping image mixed visual feature extraction model establishment and quality evaluation method | |
| CN115187982B (en) | Algae detection method and device and terminal equipment | |
| CN116128825A (en) | Meibomian gland morphology analysis method based on deep learning | |
| CN116543414A (en) | Tongue color classification and tongue redness and purple quantification method based on multi-model fusion | |
| CN115690047A (en) | Prostate ultrasound image segmentation method and device based on abnormal point detection | |
| CN114463192A (en) | Infrared video distortion correction method based on deep learning | |
| CN111179224A (en) | Joint learning-based reference-free evaluation method for aerial image restoration quality |
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 |