CN112990237B - Subway tunnel image leakage detection method based on deep learning - Google Patents
Subway tunnel image leakage detection method based on deep learning Download PDFInfo
- Publication number
- CN112990237B CN112990237B CN201911210533.1A CN201911210533A CN112990237B CN 112990237 B CN112990237 B CN 112990237B CN 201911210533 A CN201911210533 A CN 201911210533A CN 112990237 B CN112990237 B CN 112990237B
- Authority
- CN
- China
- Prior art keywords
- image
- cnn
- alexnet
- scale
- leakage
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Image Analysis (AREA)
Abstract
A subway tunnel image leakage detection method based on deep learning includes utilizing data enhancement technology to increase training sample number, respectively building corresponding tunnel leakage point image training data sets according to two neural network models based on AlexNet and based on Faster R-CNN, generating tunnel inner wall leakage detection model by utilizing two neural network models based on AlexNet and based on Faster R-CNN and respective corresponding training data sets, and finally utilizing mixed neural network based on AlexNet and Faster R-CNN to realize detection and marking of tunnel inner wall leakage points.
Description
Technical Field
The invention relates to a technology in the field of neural network application, in particular to a subway tunnel image leakage detection method based on deep learning.
Background
Subways are important transportation means in modern cities. In 2018, the average daily passenger capacity of the Shanghai subway line network breaks through 1000 ten thousand times, and the Shanghai subway line network becomes an important mode for people to go out in Shanghai cities. The importance of subways is such that even small-scale subway failures can affect the daily lives of numerous people.
Leakage in tunnels is one of the most common safety hazards for subways. But because modern cities are large in population, the passenger carrying task of subways is difficult. The daily operation time of the subway is saturated, and the working window reserved for maintenance personnel is very limited.
The traditional subway tunnel leakage point detection depends on manual patrol of maintenance personnel in non-operation time periods. This approach is time and labor consuming, costly, and also frosts snow, which is an otherwise burdensome maintenance task. At the same time, the work content is tedious and requires the worker to focus on a single repetitive task. This is wasteful of labor costs and increases the likelihood of inadvertent work due to mental breakdown.
Deep learning techniques have received a great deal of attention in recent years. The performance of the method in the fields of image recognition, target detection and the like exceeds that of the traditional mode recognition method, and is even better than that of human beings in a few fields. The task of identifying the leakage points on the inner wall of the subway tunnel conforms to the field of target detection in which the current deep technology is widely applied. The potential savings in time and human resources are enormous if the leak point identification work can be simplified or even automated using correlation techniques.
However, since the popularization time of the deep learning technology is short, the application attempts in the related art are all at an early stage, and thus no relatively mature commercial system for detecting image leakage of the inner wall of the subway tunnel is currently applied on a large scale. The identification of the leakage points of the subway tunnel still mainly depends on the traditional manual method.
Meanwhile, the leakage detection task of the inner wall of the subway tunnel belongs to a narrow and specific field. The number of data set samples available for model training is small, and how to properly expand the data set is also a problem to be solved. The method combines a plurality of data enhancement technologies suitable for the tunnel inner wall image leakage point detection task scene, reasonably expands a small amount of manually labeled data sets, and avoids the model overfitting phenomenon of the neural network caused by too few samples in the training set.
On the other hand, the scale invariance of the deep neural network is poor, and for a fully trained model, if the scale of an input image changes, the accuracy of the model is also obviously reduced. The mixed neural network model provided by the invention judges the scale of the image firstly, and then adjusts the scale of the input image to make the mixed neural network model suitable for the identification network for target detection, thereby obviously improving the accuracy of the model under the condition of uncertain input scale and improving the generalization capability of the whole identification system.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a subway tunnel image leakage detection method based on deep learning, which is based on a mixed neural network of AlexNet and Faster R-CNN, has high identification precision, can effectively identify one or more leakage targets in the same image, and has higher detection accuracy rate under the condition that the dimension of an input image is uncertain.
The invention is realized by the following technical scheme:
the invention relates to a tunnel inner wall leakage detection method based on a hybrid neural network of AlexNet and Faster R-CNN, which comprises the steps of increasing the number of training samples by using a data enhancement technology, respectively establishing corresponding tunnel leakage point image training data sets according to two neural network models based on AlexNet and Faster R-CNN, generating a tunnel inner wall leakage detection model by using the two neural network models based on AlexNet and Faster R-CNN and the corresponding training data sets, and finally detecting and marking tunnel inner wall leakage points by using the hybrid neural network based on AlexNet and Faster R-CNN.
The data enhancement technique includes at least one of:
1. adjusting brightness and contrast: randomly adjusting the brightness and the contrast of the image;
2. blurring: applying Gaussian blur of different scales to the image;
3. zooming: only aiming at an AlexNet-based image scale correction model, scaling an input image in the same or different scales in the horizontal and vertical directions;
4. implanting noise: the method comprises the steps of implanting Gaussian noise, selecting the size of an implantation area, and randomly positioning a black or white rectangular area;
5. random erasing: randomly selecting one or a plurality of areas of the image, erasing image information, and covering the area by using gray;
the hybrid neural network based on AlexNet and Faster R-CNN comprises a model based on AlexNet image scale correction and a model based on fast R-CNN image leakage point identification, wherein: the method comprises the steps of firstly judging the image scale of an input image through an AlexNet-based image scale correction model, then carrying out proper amplification or reduction on an original image according to the difference between the original image scale and the image scale suitable for a Faster R-CNN-based identification network model, and then transmitting the image with the adjusted scale to a Faster R-CNN-based image leakage point identification model for leakage point detection.
Technical effects
Compared with the traditional target detection model based on deep learning, the method has the advantages that the deep learning technology is utilized to identify the tunnel inner wall image collected by special equipment, the position and the size of the leakage area are automatically marked, and higher accuracy can be maintained when the dimension of the input image is not in line with expectation; the effects thus produced include: and training to obtain the tunnel leakage detection model with high recognition rate and wide universality under the condition of low labor cost. Once model training is completed, the identification and positioning of the leakage points can be automatically carried out on the images acquired later, the time cost and the labor cost for examining the leakage points are obviously reduced, and:
1. because the image data acquisition of the inner wall of the tunnel is difficult, the data used for training needs to be manually marked, and therefore the sample size of the training set is small. The invention provides a reasonable data enhancement method, which expands the size of a small amount of manually marked training sets to a certain extent and effectively avoids the risk of model overfitting.
2. The two deep neural network models work independently, so that aiming at each network, purposeful data set expansion is carried out through different data enhancement methods.
3. In order to improve the precision of the detection of the leakage points, a hybrid neural network based on AlexNet and Faster R-CNN is utilized. Firstly, scaling the image scale to a scale suitable for a target detection model by using an AlexNet-based image scale correction network model, and then carrying out leakage point detection on the scale-adjusted image by using a fast R-CNN-based target detection network model. The hybrid neural network model based on AlexNet and Faster R-CNN effectively improves the detection precision when the scale of the input image is not in accordance with expectation, and improves the generalization capability of the recognition model.
4. The neural network is optimized aiming at the image leakage detection of the inner wall of the tunnel, and the optimal neural network structure and parameters suitable for the task are finally determined through a comparison experiment.
Drawings
FIG. 1 is a flow chart of the method of the present invention.
Detailed Description
In the embodiment, the model using AlexNet-based image scale correction and the model using fast R-CNN-based image leakage point identification comprises the following steps:
step 1, manually marking the acquired subway tunnel inner wall image to obtain the image and a target leakage point information file corresponding to the image.
The corresponding target leakage point information file comprises information such as the size of each image and the coordinates of the upper left corner and the lower right corner of each leakage point in the image, and a pair of the coordinates of the upper left corner and the lower right corner can correspond to a rectangular area where the leakage point is located. The rectangular area is generated by manual labeling.
Due to the tedious manual labeling process, the number of the image samples obtained in step 1 is usually only hundreds, although each image may contain a plurality of leakage points, the total number of the samples is still too small to be suitable for training the deep neural network, so that it is necessary to increase the number of the samples by adopting a suitable data enhancement technology and prevent the phenomenon of overfitting of the training network caused by too single data enhancement means.
Step 2 extends the data set composed of the images acquired in step 1 using data enhancement techniques.
Step 2.1 a sample image data set S is prepared, the sample data set being obtained by step 1.
Step 2.2 image generation: for each original image in S, respectively making several times of different random brightness and contrast adjustments to obtain brightness and contrast-adjusted image and original image, adding them into expansion set S (1) 。
Step 2.3 for S (1) Each image of (1) is processed byAdding the image with Gaussian blur and the original image without Gaussian blur into an expansion set S (2) 。
Step 2.4 for S (2) Respectively implanting Gaussian noise for a plurality of times, randomly selecting the distribution and the quantity of the Gaussian noise within a set range, and adding the images after and before the noise implantation into a new set S (3.1) Then to S (3.1) The black or white rectangular area is implanted into each image, the black or white rectangular area covers the original pixel color of the area, the size and the number of the rectangular areas are randomly selected in a set range, the generation position of the rectangle is any random position of the whole image, and finally the image implanted into the black and white rectangular area and the image not implanted into the black and white rectangular area are added into a set S (3.2) 。
Step 2.5 for set S (3.2) All the images in (2) are randomly erased, the random erasing process is similar to the implantation of the black and white rectangular area in step 2.4, but in principle, the erased area covers the color of the area itself with the gray scale of 127, the size of the erased rectangular area is relatively larger, and the number of the erased areas in the same image is relatively smaller. In principle, it is avoided that the erase region completely covers the marked bleed region. Adding the randomly erased image and the non-randomly erased image to the expanded data set S (4) 。
And 2.6, supplementing the leakage point annotation file for each image which does not exist in the original image set S. Since the size of the image itself and the position of the leakage point in the image are not changed by all the data enhancement methods in step 2, the specific method for supplementing the annotation file is as follows: the corresponding label file of the original image in the set S is copied and renamed to the name consistent with the corresponding generated image, and all label files corresponding to all images are added into the set S (4) And obtaining a final expansion data set S'. The data enhancement process for the model based on fast R-CNN image leakage point identification is now complete.
In specific implementation, for example, the minimum and maximum values of the sizes of the black and white rectangular color blocks, and the minimum and maximum values of the number of the rectangular color blocks implanted in the same image need to be set in advance, and these super parameters can be set according to specific needs.
Step 3 performs an additional expansion of the expanded data set S' generated in step 2 for the AlexNet based image scale modified model.
Step 3.1 copies the expanded data set S 'resulting from step 2 and sets the result of the copying as a set T'.
Step 3.2 generates a label 1.0 for each image in the set T'. This value of 1.0 represents the original scale of the image, i.e., the scale size best suited for the fast R-CNN based target detection network model.
Step 3.3 of several random degree enlargements or reductions of each image in the set T', defining the zoom ratio
The scaling ratio of the unscaled image was 1.0. And then, randomly cutting the zoomed image, wherein the minimum width and the minimum height after cutting are both set in advance, generally are 10% of the width and the height of the original image, the zoom ratio r is used as a label corresponding to the generated image, and the zoomed image and the label corresponding to the zoomed image are added into a set T'. Up to this point, the additional expansion process of the data set for the AlexNet based image scale corrected model is completed.
Step 4 trains a target detection network model based on fast R-CNN using the set S'.
Step 5 uses the set T' to train an AlexNet based image scale modified model.
Step 6, utilizing the trained model to carry out the alignment on the new image P to be detected 0 And (5) detecting a leakage point.
Step 6.1 New image P to be detected 0 The recognition model is input.
Step 6.2 generating the scaling ratio r between the image and the optimal image scale from the AlexNet based image scale modification model 0 。
Step 6.3 baseAt r is 0 To P 0 Zooming is carried out to P 0 New image P 'adjusted to scale size suitable for fast R-CNN based target detection network model' 0 。
Step 6.4P' 0 And inputting a target detection network model based on fast R-CNN to obtain the position and size information of the leakage point. And ending the process of identifying and detecting the leakage point.
The method has expandability, can train more detection models of leakage types according to the deep learning technology, and increases the recognition capability of the models.
The foregoing embodiments may be modified in many different ways by those skilled in the art without departing from the spirit and scope of the invention, which is defined by the appended claims and all changes that come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein.
Claims (1)
1. A method for detecting leakage of inner walls of tunnels of hybrid neural networks based on AlexNet and Faster R-CNN is characterized in that the number of training samples is increased by using a data enhancement technology, corresponding tunnel leakage point image training data sets are respectively established according to two neural network models based on AlexNet and Faster R-CNN, then two neural network models based on AlexNet and Faster R-CNN and respective corresponding training data sets are used for generating tunnel inner wall leakage detection models, and finally detection and marking of the tunnel inner wall leakage points are realized by using the hybrid neural networks based on AlexNet and Faster R-CNN;
the data enhancement technology comprises the following steps:
a. adjusting brightness and contrast: randomly adjusting the brightness and the contrast of the image;
b. blurring: applying Gaussian blur of different scales to the image;
c. zooming: only aiming at an AlexNet-based image scale correction model, carrying out scaling on an input image in the same or different scales in the horizontal and vertical directions;
d. implanting noise: the method comprises the steps of implanting Gaussian noise, selecting the size of an implantation area, and randomly positioning a black or white rectangular area;
e. random erasing: randomly selecting one or a plurality of areas of the image, erasing image information, and covering the areas by gray;
the hybrid neural network based on AlexNet and Faster R-CNN comprises a model based on AlexNet image scale correction and a model based on fast R-CNN image leakage point identification, wherein: the method comprises the steps that firstly, an input image is subjected to AlexNet-based image scale correction model to judge the image scale, then the original image is properly amplified or reduced according to the difference between the original image scale and the image scale suitable for the fast R-CNN-based recognition network model, and then the image with the adjusted scale is transmitted to the fast R-CNN-based image leakage point recognition model to detect leakage points;
the tunnel inner wall leakage detection specifically comprises:
step 1, manually marking the acquired image of the inner wall of the subway tunnel to obtain an image and a target leakage point information file corresponding to the image, wherein the corresponding target leakage point information file comprises the size information of each image and the upper left corner coordinate and the lower right corner coordinate of each leakage point in the image, a pair of upper left corner coordinate and lower right corner coordinate corresponds to a rectangular area where the leakage point is located, and the rectangular area is generated by manual marking;
step 2, expanding the data set formed by the images acquired in the step 1 by using a data enhancement technology, and specifically comprising the following steps:
step 2.1 preparing a sample image dataset S, the sample dataset being obtained in step 1;
step 2.2 image generation: for each original image in S, respectively making several times of different random brightness and contrast adjustments to obtain brightness and contrast-adjusted image and original image, adding them into expansion set S (1) ；
Step 2.3 for S (1) Each image in the image processing system is subjected to Gaussian blur of different degrees for a plurality of times, and the image subjected to Gaussian blur and the original image not subjected to Gaussian blur are added into an expansion set S (2) ；
Step 2.4 for S （2) Respectively implanting Gaussian noise for a plurality of times, randomly selecting the distribution and the quantity of the Gaussian noise within a set range, and adding the images after and before the noise implantation into a new set S (3.1) Then to S (3.1) The black or white rectangular area is implanted into each image, the black or white rectangular area covers the original pixel color of the area, the size and the number of the rectangular areas are randomly selected in a set range, the generation position of the rectangle is any random position of the whole image, and finally the image implanted with the black and white rectangular area and the image not implanted with the black and white rectangular area are added into a set S (3.2) ；
Step 2.5 for set S (3.2) The erasing area covers the color of the area by using the gray with the gray level of 127, the erasing area is prevented from completely covering the marked leakage area, and the randomly erased image are added into the expanded data set S (4) ；
Step 2.6 is to supplement the leakage point annotation file for each image which does not exist in the original image set S, and since all the data enhancement methods in step 2 do not change the size of the image itself and the position of the leakage point in the image, the specific way of supplementing the annotation file is as follows: the corresponding label file of the original image in the set S is copied and renamed to the name consistent with the corresponding generated image, and all label files corresponding to all images are added into the set S (4) Obtaining a final expansion data set S', and finishing the data enhancement process aiming at the model identified by the image leakage points based on the Faster R-CNN;
step 3, performing additional expansion on the extended data set S' generated in step 2 for the model based on AlexNet image scale modification, specifically including:
step 3.1 copy the augmented data set S 'generated by step 2 and set the result of the copy as the set T';
step 3.2, generating a label 1.0 for each image in the set T', wherein the value 1.0 represents the original scale of the image, namely the scale size which is most suitable for a target detection network model based on Faster R-CNN;
step 3.3 random degree enlargement or reduction of each image in the set T' several times, defining the zoom ratio
The scaling ratio of an image which is not scaled is 1.0, then the image which is scaled is randomly cut, the minimum width and the minimum height after cutting are both set in advance and are 10% of the width and the height of the original image, the scaling ratio r is used as a label corresponding to a generated image, and the scaled image and the label corresponding to the scaled image are added into a set T', so that the process of additionally expanding the data set of the AlexNet-based image scale correction model is completed;
step 4, training a target detection network model based on Faster R-CNN by using the set S';
step 5, training the AlexNet-based image scale correction model by using the set T';
step 6, the trained model is utilized to carry out the detection on the new image P to be detected 0 And (3) detecting a leakage point, and specifically comprising:
step 6.1 New image P to be detected 0 Inputting a recognition model;
step 6.2 generating the scaling ratio r between the image and the optimal image scale from the AlexNet based image scale modification model 0 ；
Step 6.3 is based on r 0 To P 0 Zooming is carried out to P 0 New image P 'adjusted to scale size suitable for fast R-CNN based target detection network model' 0 ；
Step 6.4P' 0 Inputting a target detection network model based on fast R-CNN to obtain the position and size information of the leakage point, and ending the identification and detection process of the leakage point.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911210533.1A CN112990237B (en) | 2019-12-02 | 2019-12-02 | Subway tunnel image leakage detection method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911210533.1A CN112990237B (en) | 2019-12-02 | 2019-12-02 | Subway tunnel image leakage detection method based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112990237A CN112990237A (en) | 2021-06-18 |
| CN112990237B true CN112990237B (en) | 2023-04-07 |
Family
ID=76330955
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911210533.1A Active CN112990237B (en) | 2019-12-02 | 2019-12-02 | Subway tunnel image leakage detection method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112990237B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113674216A (en) * | 2021-07-27 | 2021-11-19 | 南京航空航天大学 | Subway tunnel disease detection method based on deep learning |
| CN115424129B (en) * | 2022-10-13 | 2023-08-11 | 哈尔滨市科佳通用机电股份有限公司 | Abnormality detection method and abnormality detection system for wallboard damage |
| CN115546614B (en) * | 2022-12-02 | 2023-04-18 | 天津城建大学 | Safety helmet wearing detection method based on improved YOLOV5 model |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2017107334A1 (en) * | 2015-12-25 | 2017-06-29 | 同济大学 | Subway tunnel structure cross section deformation rapid detection device |
| CN108230292A (en) * | 2017-04-11 | 2018-06-29 | 北京市商汤科技开发有限公司 | The training method of object detecting method and neural network, device and electronic equipment |
| CN109190680A (en) * | 2018-08-11 | 2019-01-11 | 复旦大学 | The detection and classification method of Medicines image based on deep learning |
| CN109344883A (en) * | 2018-09-13 | 2019-02-15 | 西京学院 | Fruit tree diseases and pests recognition methods under a kind of complex background based on empty convolution |
| CN109359336A (en) * | 2018-09-14 | 2019-02-19 | 上海交通大学 | A kind of similar distortion model construction method of lashing bridge based on multiple-objection optimization |
| CN109615653A (en) * | 2018-12-19 | 2019-04-12 | 上海大学 | Percolating water area detecting and recognition methods based on deep learning and visual field projection model |
| CN110378862A (en) * | 2019-07-19 | 2019-10-25 | 国网河北省电力有限公司电力科学研究院 | A kind of raising transmission line of electricity breaks the data enhancement methods of target identification accuracy outside |
-
2019
- 2019-12-02 CN CN201911210533.1A patent/CN112990237B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2017107334A1 (en) * | 2015-12-25 | 2017-06-29 | 同济大学 | Subway tunnel structure cross section deformation rapid detection device |
| CN108230292A (en) * | 2017-04-11 | 2018-06-29 | 北京市商汤科技开发有限公司 | The training method of object detecting method and neural network, device and electronic equipment |
| CN109190680A (en) * | 2018-08-11 | 2019-01-11 | 复旦大学 | The detection and classification method of Medicines image based on deep learning |
| CN109344883A (en) * | 2018-09-13 | 2019-02-15 | 西京学院 | Fruit tree diseases and pests recognition methods under a kind of complex background based on empty convolution |
| CN109359336A (en) * | 2018-09-14 | 2019-02-19 | 上海交通大学 | A kind of similar distortion model construction method of lashing bridge based on multiple-objection optimization |
| CN109615653A (en) * | 2018-12-19 | 2019-04-12 | 上海大学 | Percolating water area detecting and recognition methods based on deep learning and visual field projection model |
| CN110378862A (en) * | 2019-07-19 | 2019-10-25 | 国网河北省电力有限公司电力科学研究院 | A kind of raising transmission line of electricity breaks the data enhancement methods of target identification accuracy outside |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112990237A (en) | 2021-06-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112990237B (en) | Subway tunnel image leakage detection method based on deep learning | |
| CN108230339B (en) | Stomach cancer pathological section labeling completion method based on pseudo label iterative labeling | |
| US11854244B2 (en) | Labeling techniques for a modified panoptic labeling neural network | |
| CN108776772B (en) | Cross-time building change detection modeling method, detection device, method and storage medium | |
| CN105608455B (en) | A kind of license plate sloped correcting method and device | |
| CN107563379B (en) | Method for positioning text in natural scene image | |
| CN105373794A (en) | Vehicle license plate recognition method | |
| CN101710425B (en) | Self-adaptive pre-segmentation method based on gray scale and gradient of image and gray scale statistic histogram | |
| CN111967313B (en) | Unmanned aerial vehicle image annotation method assisted by deep learning target detection algorithm | |
| Chen et al. | Shadow-based Building Detection and Segmentation in High-resolution Remote Sensing Image. | |
| CN111582339B (en) | Vehicle detection and recognition method based on deep learning | |
| CN113689445B (en) | High-resolution remote sensing building extraction method combining semantic segmentation and edge detection | |
| CN106570874A (en) | Image marking method combining local image constraint and overall target constraint | |
| CN108509950B (en) | Railway contact net support number plate detection and identification method based on probability feature weighted fusion | |
| CN102542244A (en) | Face detection method and system and computer program product | |
| CN113870236B (en) | Composite material defect nondestructive inspection method based on deep learning algorithm | |
| CN110599453A (en) | Panel defect detection method and device based on image fusion and equipment terminal | |
| CN104700415A (en) | Method of selecting matching template in image matching tracking | |
| CN112733711A (en) | Remote sensing image damaged building extraction method based on multi-scale scene change detection | |
| CN110059704B (en) | Intelligent extraction method of remote sensing information of rare earth mining area driven by visual attention model | |
| CN111768405A (en) | Method, device, equipment and storage medium for processing annotated image | |
| CN112819834B (en) | Method and device for classifying stomach pathological images based on artificial intelligence | |
| CN103093241B (en) | Based on the remote sensing image nonuniformity cloud layer method of discrimination of homogeneity process | |
| CN110348339B (en) | Method for extracting handwritten document text lines based on case segmentation | |
| CN115578725B (en) | Landslide boundary and crack marking method and system based on man-machine cooperation |
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 |