CN110378916B - TBM image slag segmentation method based on multitask deep learning - Google Patents
TBM image slag segmentation method based on multitask deep learning Download PDFInfo
- Publication number
- CN110378916B CN110378916B CN201910601332.8A CN201910601332A CN110378916B CN 110378916 B CN110378916 B CN 110378916B CN 201910601332 A CN201910601332 A CN 201910601332A CN 110378916 B CN110378916 B CN 110378916B
- Authority
- CN
- China
- Prior art keywords
- image
- segmentation
- data set
- stone
- neural network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/12—Edge-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/181—Segmentation; Edge detection involving edge growing; involving edge linking
-
- 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
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- 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
- G06T2207/30108—Industrial image inspection
- G06T2207/30132—Masonry; Concrete
Abstract
The invention discloses a TBM image slag segmentation method based on multitask deep learning. Sending the original image data set and the segmentation label data set into a first image segmentation neural network for training; sending the original image data set and the edge label data set into a second image segmentation neural network for training; respectively inputting the TBM image to be processed into the two trained image segmentation neural networks to obtain masks, fusing the masks of the two networks, and then performing deconvolution processing to generate a segmentation result graph; and (4) carrying out post-processing to eliminate fine points on the image, generating a stone particle size distribution diagram, and realizing slag tapping segmentation of the TBM image. According to the invention, stone edge subtask learning contour information is introduced, and finally, the output results of the two subtasks are subjected to mask fusion, so that stones which are mutually contacted are effectively separated, and the number of detected stones is increased. Compared with a single task deep learning segmentation algorithm, the segmentation precision and accuracy of the method are greatly improved.
Description
Technical Field
The invention relates to an image slag segmentation method under a TBM system, in particular to a TBM image slag segmentation method based on multitask deep learning.
Background
A Tunnel Boring Machine (TBM) is a large system, and has many parameters, and some TBM parameters need to be known in advance to adjust the system in order to stabilize the system. In the deslagging task, the size of the produced stone can effectively reflect the current state of the TBM, the too large stone reflects that the TBM is too fast, and the too small stone reflects that the energy in the TBM is excessive. Therefore, by detecting the distribution of the stone particle size, the method can bring significant guiding significance to the adjustment of the TBM system.
In the slag task, the acquired image is as shown in fig. 1, because light is poor, shadows on the image are very much, and because the quantity of stones is huge, all stones cannot be segmented, only remarkable stones in the image can be segmented, gaps among the stones are too small, and the edge information of the stones cannot be well extracted by a traditional segmentation algorithm and a single-task deep learning method. According to the multi-task deep learning method provided by the invention, the stone segmentation task is used for segmenting position information of a remarkable stone, the stone edge task is used for extracting outline information of the stone, masks generated by the two tasks are fused, a more accurate prediction result is obtained, and more accurate stone particle size distribution is obtained.
Disclosure of Invention
In order to solve the problems in the background art, the invention mainly provides a TBM image slag segmentation method based on multitask deep learning.
The method is divided into two tasks, wherein the stone segmentation task is used for segmenting position information of the remarkable stones, the stone edge task is used for extracting outline information of the remarkable stones, and masks of the two tasks are fused to obtain a more accurate prediction result and obtain more accurate stone grain size distribution.
The technical scheme adopted by the invention is as follows:
(1) processing to obtain an original image data set, a segmentation label data set and an edge label data set;
(2) sending the original image data set and the segmentation label data set into a first image segmentation neural network for training;
(3) sending the original image data set and the edge label data set into a second image segmentation neural network for training;
(4) respectively inputting a TBM image to be processed into a trained first image segmentation neural network and a trained second image segmentation neural network, not directly outputting the results of the networks, but obtaining masks obtained in the middle steps of the first image segmentation neural network and the second image segmentation neural network, fusing the masks of the two networks, and then performing deconvolution processing to generate a segmentation result graph;
(5) and (4) carrying out post-processing to eliminate fine points on the image, counting the quantity of the stones on the generated stone segmentation result graph, and then generating a stone particle size distribution graph to realize final slag tapping segmentation of the TBM image.
In the step (1), the original image data set is an image data set consisting of a series of original TBM system muck images shot and collected by a camera; the segmentation label data set is a binaryzation TBM system muck image of an image area marked with a stone target, wherein the image area of the stone target is assigned to be white, and the image area outside the stone image area is assigned to be black; the edge label data set is generated by a segmentation label data set, each image in the segmentation label data set is extracted by adopting a contour extraction algorithm to obtain a contour, and then an edge label image is generated by expansion operation processing, wherein the contour line of the stone block is assigned to be white, and the contour lines except the contour line of the stone block are assigned to be black.
In the step (2), the first image segmentation neural network is a full convolution network, an original image data set and a segmentation label data set are input, and the trained first image segmentation neural network capable of detecting the complete region of the stone block is obtained by supervising and learning the position of the complete region of the stone block; processing the TBM image input to be processed by the first image segmentation neural network enables to predict the complete region position of the stone in the image.
In the step (3), the second image segmentation neural network is a full convolution network, an original image data set and an edge label data set are input, and the trained second image segmentation neural network capable of detecting the stone contour line is obtained by monitoring and learning the contour line position of the stone; processing the TBM image input to be processed by the second image segmentation neural network enables to predict the contour position of the stone in the image.
The invention provides a subtask of edge processing is added to generate a stone edge mask, and stone edge information is extracted.
The invention adds the edge label data set and the learning edge information thereof in the technical scheme, and aims to solve the problem of contact between adjacent stones caused by simple stone segmentation.
In the step (2) and the step (3), the first image segmentation neural network and the second image segmentation neural network have the same structure, a global convolution network is adopted, and specifically, a 50-layer residual error network is adopted.
And (1) extracting the features of the input image through continuous convolution operation, wherein due to the introduction of global convolution, the generated high-dimensional feature map can better express the region information of the object to be segmented in the input image. 2. An edge fine-tuning layer is introduced, so that the problem of edge blurring caused by convolution operation is solved, and edge information of an object to be segmented in the characteristic diagram obtained in the first step is retained. 3. And performing up-sampling on the characteristic diagram obtained in the second step through continuous deconvolution operation to obtain a segmented image with the resolution consistent with that of the input image.
In the step (4), the masks obtained in the intermediate step between the first image segmentation neural network and the second image segmentation neural network are fused, and the formula is as follows:
if x is object and not contour
wherein, x represents a pixel point in the mask, object represents a complete area of the stone in the mask of the first image segmentation neural network, and contour represents a contour line of the stone in the mask of the second image segmentation neural network; the formula represents: and if the pixel point x is in the complete area range of the stone in the mask of the first image segmentation neural network and is not in the outline range of the stone in the mask of the second image segmentation neural network, assigning the pixel point x to be 1 to represent the stone, and otherwise, assigning the pixel point x to be 0 to represent the background.
And generating a predicted image result of the last deconvolution according to the mask, and obtaining a segmentation result from the predicted image result.
In the step (5), the output image in the step (4) is post-processed, and as the edge label data set generated by the expansion operation is adopted in the step (1), specifically, multiple times of expansion operation are adopted to eliminate fine points, the size and the number of stones are counted on the generated stone segmentation result graph, and a stone particle size distribution graph is generated.
The invention has the beneficial effects that:
the TBM muck segmentation algorithm for the multitask deep learning avoids the difficulty of manual parameter adjustment of the traditional segmentation algorithm, and greatly improves the segmentation precision.
The method uses multi-task learning, solves the problem of low precision caused by using a traditional segmentation algorithm, solves the problems of contacting stone separation failure and unclear stone edge division caused by using a single-task deep learning segmentation algorithm, improves the segmentation precision, obtains more accurate and effective stone segmentation images, can be applied to a TBM system to obtain the current TBM state, and has guiding significance for system regulation.
Drawings
FIG. 1 is an exemplary diagram of an image of an original image dataset according to an embodiment;
FIG. 2 is an exemplary diagram of an image of a segmented tag dataset according to an embodiment;
FIG. 3 is an exemplary diagram of an edge tag dataset image according to an embodiment;
FIG. 4 is an exemplary diagram of a global convolutional network structure of an embodiment;
FIG. 5 is an exemplary diagram of an output image of a first image segmentation neural network according to an embodiment;
FIG. 6 is an exemplary diagram of an output image of a second image segmentation neural network according to the embodiment;
FIG. 7 is an exemplary diagram of a model structure of an embodiment of a multitasking deep learning network;
FIG. 8 is an exemplary diagram of an output image after image fusion according to an embodiment;
FIG. 9 is an exemplary graph of an output image of post-processing of an embodiment;
FIG. 10 is a graph showing a comparison of separation effects in examples;
fig. 11 is a graph of the stone grain size distribution obtained in the example.
Detailed Description
The TBM stone segmentation algorithm based on the multitask deep learning is further explained by combining a specific test verification process.
The specific embodiment of the invention is as follows:
step 1: the method comprises the steps of manufacturing a data set, manufacturing a muck data set applied by a TBM system, selecting a remarkable stone by manual labeling when manufacturing the data set because too many stones are input into an image, generating a segmentation label data set of the remarkable stone, wherein a stone segmentation label image is shown in figure 2, and separating the outline of the stone segmentation label data set by an outline extraction algorithm because of a stone edge subtask, and generating a stone edge label data set by image expansion operation, wherein the stone edge label image is shown in figure 3.
Step 2: the method comprises the steps of constructing a deep learning image segmentation model, extracting semantic features of an image by adopting a full convolution network through a continuous convolution filter, and then gradually up-sampling to recover resolution. The concrete structure is as follows:
the neural network adopts a GCN network, semantic features are extracted through continuous convolution operation, and compared with the traditional segmentation network, the global convolutional layer is introduced and has a larger receptive field, so that better semantic information can be obtained. And because the global convolution network introduces the boundary fine tuning layer, the output prediction image has more accurate object contour and meets the requirement of slag tapping segmentation items. And finally, outputting a prediction result graph with the same resolution as that of the original graph through deconvolution operation, wherein the network structure schematic diagram is shown in FIG. 4.
And step 3: and (3) a stone segmentation subtask, namely segmenting the neural network by using the first image written in the step (2), taking the original image data set and the segmentation label data set as input, training a model by adjusting parameters, generating a stone segmentation image, and predicting the complete region position of the stone in the image. The training process is as follows:
step 3.1: because few images are collected in an entity scene and cannot meet the requirements of deep learning training, 10 images of 512 images are randomly cut out, randomly rotated and horizontally turned over for each 1600-1200 shot image, and a data set is expanded.
Step 3.2: the basic framework used by the global convolutional network is a residual network, so a pre-trained residual network model is used, a learning rate is set to be 0.0001, a weight attenuation rate is set to be 0.0005, a supervised learning mode is adopted, the input data set and the segmentation tag data set are input into the network together, the result is gradually coincided with the segmentation tag data set by adjusting network parameters, and the block segmentation subtask output schematic diagram is shown in fig. 5.
And 4, step 4: stone block edge subtasks. In the segmentation result output in the step 3, the stones which are contacted with each other are all segmented together by mistake, so the invention proposes to add an edge subtask, segment the neural network by using the second image written in the step 2, train the original image data set and the edge label data set as input, generate a stone edge mask, extract stone edge information, and the specific training steps are as follows:
similar to step 3.2, after a pre-trained residual network model is used, a learning rate is set to be 0.0001, and a weight attenuation rate is set to be 0.0005, since image segmentation operation is to perform classification operation on each pixel point of an image, and the sub-task needs to distinguish whether the pixel point is a contour or a background, since the number of the contour pixel points is far smaller than that of the background pixel points, a number ratio of 1.4:6.4 is adopted as a new sub-task training standard, and a predicted image is output at the edge of a stone as shown in fig. 6.
And 5: and (4) performing sub-task mask fusion, namely performing sub-mask fusion by using the output results of the step (3) and the step (4), wherein the overall multi-task deep learning segmentation network model is shown in a figure 7.
Mask fusion was performed using the following equation 1.
if x is object and not contour
Wherein, x represents a pixel point in the image, object represents a stone complete region in the output image of the first image segmentation neural network in the step 3, and contour represents a stone contour line in the output image of the second image segmentation neural network in the step 4, if the pixel point x is in the stone complete region range in the output image of the first image segmentation neural network and is not in the stone contour line range in the output image of the second image segmentation neural network, the pixel point x is assigned to 1 to represent the stone, otherwise, the pixel point x is assigned to 0 to represent the background, thus obtaining the fused image.
Step 6: the post-processing eliminates the tiny dots, the segmented image output in the step 5 has a plurality of tiny dots which are not beneficial to counting the stone grain size distribution, therefore, the post-processing operation of corrosion expansion on the image is adopted to eliminate the tiny dots, the morphological operation used in the step 4 is the expansion operation, the invention adopts multiple expansion operations to eliminate the tiny dots, and fig. 9 is an output image of the post-processing.
And generating a final prediction image, and counting a stone particle size distribution diagram, wherein a separation effect comparison diagram is shown in fig. 10, and an obtained stone particle size distribution diagram is shown in fig. 11.
In fig. 11, the abscissa represents the area of the stone, the ordinate represents the number of the stone, the square line is the labeled stone grain size distribution broken line of the manual statistics under the real condition, the circular line is the stone grain size distribution broken line obtained by the single stone dividing subtask, and the star line is the stone grain size distribution broken line obtained by the multitask learning after the edge subtask is addedAnd (4) distributing folding lines, namely, the particle size distribution folding lines obtained by the division method of the multitask deep learning are closer to the particle size distribution folding lines of the label stone blocks in a small area interval. The area of the stone block is 0.2 x 104For example, the number of the stones on the label broken line in the image is 33, the method provided by the invention counts that the number of the stones is 36, which is far superior to 16 counted by a single stone segmentation subtask, therefore, the method adds stone edge subtask learning contour information, and finally performs image fusion on the results of the two, thereby effectively separating the stones which are contacted with each other, and increasing the number of the detected stones. Compared with a single task deep learning segmentation algorithm, the segmentation precision and accuracy of the method are greatly improved.
Claims (4)
1. A TBM image slag tapping segmentation method based on multitask deep learning is characterized by comprising the following steps: the method comprises the following steps:
(1) processing to obtain an original image data set, a segmentation label data set and an edge label data set;
the edge label data set is generated by a segmentation label data set, each image in the segmentation label data set is extracted by adopting a contour extraction algorithm to obtain a contour, and then an edge label image is generated by expansion operation processing, wherein the contour line of the stone block is assigned to be white, and the contour lines except the contour line of the stone block are assigned to be black;
(2) sending the original image data set and the segmentation label data set into a first image segmentation neural network for training;
in the step (2), the first image segmentation neural network is a full convolution network, an original image data set and a segmentation label data set are input, and the trained first image segmentation neural network capable of detecting the complete region of the stone block is obtained by supervising and learning the position of the complete region of the stone block;
(3) sending the original image data set and the edge label data set into a second image segmentation neural network for training;
in the step (3), the second image segmentation neural network is a full convolution network, an original image data set and an edge label data set are input, and the trained second image segmentation neural network capable of detecting the stone contour line is obtained by monitoring and learning the contour line position of the stone;
in the step (2) and the step (3), the first image segmentation neural network and the second image segmentation neural network have the same structure, and a global convolution network is adopted; the global convolution network adopts a residual error network of a GCN network;
(4) respectively inputting a TBM image to be processed into a trained first image segmentation neural network and a trained second image segmentation neural network, not directly outputting the results of the networks, but obtaining masks obtained in the middle steps of the first image segmentation neural network and the second image segmentation neural network, fusing the masks of the two networks, and then performing deconvolution processing to generate a segmentation result graph;
(5) and (4) carrying out post-processing to eliminate fine points on the image, counting the quantity of the stones on the generated stone segmentation result graph, and then generating a stone particle size distribution graph to realize final slag tapping segmentation of the TBM image.
2. The TBM image slag segmentation method based on multitask deep learning according to claim 1, which is characterized in that: in the step (1), the original image data set is an image data set consisting of a series of original TBM system muck images shot and collected by a camera; the segmentation label data set is a binaryzation TBM system muck image marked with an image area of the stone target, wherein the image area of the stone target is assigned to be white, and the image area outside the stone image area is assigned to be black.
3. The TBM image slag segmentation method based on multitask deep learning according to claim 1, which is characterized in that: in the step (4), the masks obtained in the intermediate step between the first image segmentation neural network and the second image segmentation neural network are fused, and the formula is as follows:
if x is object and not contour
wherein, x represents the pixel in the mask, object represents the complete region of the stone in the mask of the first image segmentation neural network, contour represents the contour line of the stone in the mask of the second image segmentation neural network.
4. The TBM image slag segmentation method based on multitask deep learning according to claim 1, which is characterized in that: and (5) performing post-processing on the output image in the step (4), specifically, eliminating fine points by adopting multiple times of expansion operation, and performing statistics on the size and the number of the stones on the generated stone segmentation result graph to generate a stone particle size distribution graph.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910601332.8A CN110378916B (en) | 2019-07-03 | 2019-07-03 | TBM image slag segmentation method based on multitask deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910601332.8A CN110378916B (en) | 2019-07-03 | 2019-07-03 | TBM image slag segmentation method based on multitask deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110378916A CN110378916A (en) | 2019-10-25 |
| CN110378916B true CN110378916B (en) | 2021-11-09 |
Family
ID=68252054
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910601332.8A Active CN110378916B (en) | 2019-07-03 | 2019-07-03 | TBM image slag segmentation method based on multitask deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110378916B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20210087792A (en) * | 2020-01-03 | 2021-07-13 | 엘지전자 주식회사 | User verification |
| CN111881811B (en) * | 2020-07-24 | 2022-04-26 | 浙江大学 | Self-accretion attention pooling mechanism-based rock classification method for tunneling equipment |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103871064B (en) * | 2014-03-25 | 2017-02-08 | 中国石油大学(华东) | Preprocessing and segmentation threshold value determining method of volcanic CT images |
| CN106504262A (en) * | 2016-10-21 | 2017-03-15 | 泉州装备制造研究所 | A kind of small tiles intelligent locating method of multiple features fusion |
| CN108182440A (en) * | 2018-01-09 | 2018-06-19 | 中铁工程装备集团有限公司 | It is a kind of that the method for obtaining surrounding rock category is identified based on slag picture |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| MX349448B (en) * | 2012-08-10 | 2017-07-28 | Ingrain Inc | Method for improving the accuracy of rock property values derived from digital images. |
| CN109408649A (en) * | 2018-09-27 | 2019-03-01 | 中铁工程装备集团有限公司 | Slag piece intelligent analysis method and system based on textural characteristics and machine learning |
| CN109461162B (en) * | 2018-12-03 | 2020-05-12 | 中国科学技术大学 | Method for segmenting target in image |
-
2019
- 2019-07-03 CN CN201910601332.8A patent/CN110378916B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103871064B (en) * | 2014-03-25 | 2017-02-08 | 中国石油大学(华东) | Preprocessing and segmentation threshold value determining method of volcanic CT images |
| CN106504262A (en) * | 2016-10-21 | 2017-03-15 | 泉州装备制造研究所 | A kind of small tiles intelligent locating method of multiple features fusion |
| CN108182440A (en) * | 2018-01-09 | 2018-06-19 | 中铁工程装备集团有限公司 | It is a kind of that the method for obtaining surrounding rock category is identified based on slag picture |
Non-Patent Citations (2)
| Title |
|---|
| A unified two-parallel-branch deep neural network for joint gland contour and segmentation learning;Linbo Wang et.al;《Future Generation Computer Systems》;20190517;第317-323页 * |
| Application of local region-growing method based on normalized cross-Correlation to TBM rock slag identification and measurement;Xiuzhen Hu et.al;《2017 Chinese Automation Congress (CAC)》;20180101;第3375-3378页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110378916A (en) | 2019-10-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106960195B (en) | Crowd counting method and device based on deep learning | |
| CN108562589B (en) | Method for detecting surface defects of magnetic circuit material | |
| CN104063883B (en) | A kind of monitor video abstraction generating method being combined based on object and key frame | |
| CN109272509B (en) | Target detection method, device and equipment for continuous images and storage medium | |
| CN111179249A (en) | Power equipment detection method and device based on deep convolutional neural network | |
| CN112232349A (en) | Model training method, image segmentation method and device | |
| CN108305253B (en) | Pathological image classification method based on multiple-time rate deep learning | |
| CN111967313B (en) | Unmanned aerial vehicle image annotation method assisted by deep learning target detection algorithm | |
| CN110414367B (en) | Time sequence behavior detection method based on GAN and SSN | |
| CN110176024B (en) | Method, device, equipment and storage medium for detecting target in video | |
| CN108734109B (en) | Visual target tracking method and system for image sequence | |
| CN110378916B (en) | TBM image slag segmentation method based on multitask deep learning | |
| CN109886159B (en) | Face detection method under non-limited condition | |
| CN113111916A (en) | Medical image semantic segmentation method and system based on weak supervision | |
| CN115482387A (en) | Weak supervision image semantic segmentation method and system based on multi-scale class prototype | |
| CN116205876A (en) | Unsupervised notebook appearance defect detection method based on multi-scale standardized flow | |
| CN103945227A (en) | Video semantic block partition method based on light stream clustering | |
| CN110363198B (en) | Neural network weight matrix splitting and combining method | |
| CN114612450B (en) | Image detection segmentation method and system based on data augmentation machine vision and electronic equipment | |
| CN112084855A (en) | Outlier elimination method for video stream based on improved RANSAC method | |
| CN116095363A (en) | Mobile terminal short video highlight moment editing method based on key behavior recognition | |
| CN114897214A (en) | Metal additive manufacturing time prediction system and method based on graphical processing | |
| CN114299012A (en) | Object surface defect detection method and system based on convolutional neural network | |
| CN112528058B (en) | Fine-grained image classification method based on image attribute active learning | |
| CN114494272A (en) | Metal part fast segmentation method based on deep learning |
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 |