CN110309727B - Building identification model establishing method, building identification method and building identification device - Google Patents
Building identification model establishing method, building identification method and building identification device Download PDFInfo
- Publication number
- CN110309727B CN110309727B CN201910500682.5A CN201910500682A CN110309727B CN 110309727 B CN110309727 B CN 110309727B CN 201910500682 A CN201910500682 A CN 201910500682A CN 110309727 B CN110309727 B CN 110309727B
- Authority
- CN
- China
- Prior art keywords
- building
- training
- query
- neural network
- convolutional neural
- 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
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention belongs to the technical field of information processing, relates to a building identification model establishing method, a building identification method and a building identification device, and particularly relates to a building identification method based on a convolutional neural network. The method selects a proper APN training sample to train the convolutional neural network, and converts deep layer or/and shallow layer network characteristics into building characteristics to form a building identification model; and inputting the building query formula into the model by the user, obtaining the building characteristic vector of the building query formula, judging and identifying the distance between the building characteristic vector and the building characteristic vector of the image in the query database, and feeding back the result to the user. Different from the existing picture recognition technology, the invention performs feature extraction in a specific field aiming at the building, reduces the difficulty of relevant parameter adjustment, and forms a perfect network based on convolutional neural training for building recognition. And the HNSW algorithm is adopted to optimize the building characteristic vector, so that the operation complexity and the calculation time are reduced, and the building identification and judgment are simple and quick.
Description
Technical Field
The invention belongs to the technical field of information processing, relates to a building identification model establishing method, a building identification method and a building identification device, and particularly relates to a building identification method based on a convolutional neural network.
Background
The building is one of important artificial features, is related to the life information of the public, and the extraction and identification of the information of the building have great promotion effect on the development of digital maps, building information databases, digital city modeling, virtual cities, tourism and the like.
At present, in a traditional mode of extracting and identifying building information, some similarities which may exist are searched for through comparing distribution conditions of pixel points and through variance approximation of the pixel points, and the method is low in accuracy. The deep learning method is currently applied to the field of data identification, such as face identification and the like, and has a better accuracy rate under the condition of a larger data volume, but has a slower speed compared with the traditional algorithm, and meanwhile, the deep learning method is blank in the field of building identification at present. How to reduce the difficulty of adjusting related parameters, reduce the difficulty of deep learning algorithm, and train a more perfect network for building identification is an urgent problem to be solved.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provide a building identification model establishing method, a building identification method and a building identification device, and aims to provide a building identification technology, reduce the operation complexity and the calculation time, and enable the building identification to be simple and rapid in judgment and reliable in result.
In a first aspect of the present invention, a building identification model establishing method is provided, including:
training a convolutional neural network by using APN data samples;
taking the trained convolutional neural network as a building recognition model:
the APN data sample comprises a training building query formula, a positive sample building picture and other building pictures;
the training of the convolutional neural network by using the APN data sample comprises the steps of respectively receiving information of a training building query type + positive sample building picture, a training building query type + other building pictures by using a Simase network, and performing model training.
Preferably, the training of the convolutional neural network using APN data samples includes:
selecting APN training samples with a set number;
inputting the APN training sample into a convolutional neural network to calculate general features;
converting the general characteristics into architectural characteristics by using full-connection neural network calculation;
and training to obtain the minimized error determination model building characteristics of the APN training sample.
Preferably, the common features include features in a shallow network layer and/or a deep network layer, the features in the shallow network layer include image line trends, image inflection points and image fixed point features, and the features in the deep network layer include image categories and styles.
Preferably, the vector algorithm for determining the model architectural features is L (a, P, N) ═ max (| f (a) -f (P) | ^2- | | f (a) -f (N) | | ^2+ α, 0);
wherein, L (A, P, N) is an error caused by an APN training sample; f (A), f (P), f (N) respectively represent training building query formulas, positive sample building pictures and training building feature vectors obtained after other building pictures are trained; alpha is a positive number and is used for increasing the discrimination;
if the distance between the feature vectors of f (A) and f (P) is less than the distance between the feature vectors of f (A) and f (N), the training is finished, or if the distance between the feature vectors of f (A) and f (P) is greater than or equal to the distance between the feature vectors of f (A) and f (N), the number of sample pictures corresponding to each picture type is increased, so as to perform the training again.
In a second aspect of the present invention, there is provided a building identification method, including:
acquiring a building query formula input by a user;
inputting the images in the building query formula and the query database into the building identification model, and converting the images into respective building characteristic vectors;
and (4) building identification judgment, wherein the same building is determined when the distance between the building characteristic vector of the building query type and the building characteristic vector of the image in the query database is smaller than a threshold value, and otherwise, different buildings are determined.
Preferably, the building identification judgment optimizes the building feature vector by adopting an HNSW algorithm.
In a third aspect of the present invention, there is provided a training device for building identification models established by the building identification model establishing method, including:
the APN data sample comprises a training building query formula, a positive sample building picture and other building pictures;
a training unit for training a convolutional neural network using the sample training library;
and the model generation unit is used for generating a building identification model by the trained convolutional neural network.
Preferably, the APN data sample is obtained by a crawler program or a manual marking or a crawler program assisted with a manual marking.
Preferably, the training unit comprises:
the classification extraction unit is used for extracting general features by using the building identification model and extracting network shallow features;
a computing unit, configured to convert the vector of the general feature into a building feature vector, where an algorithm of the building feature vector is L (a, P, N) max (| f (a) -f (P) | | ^2- | | f (a) -f (N) | | | ^2+ α,0),
wherein, L (A, P, N) is an error caused by an APN training sample; f (A), f (P), f (N) respectively represent training building query formulas, positive sample building pictures and training building feature vectors obtained after other building pictures are trained; α is a positive number for increasing the discrimination, i.e. if α > is 0, max (α,0) is α; if α <0, max (α,0) ═ 0;
and the judging unit is used for judging whether the convolutional neural network meets the set requirement.
In a fourth aspect of the present invention, there is provided a building identification apparatus comprising:
the building query type acquisition unit is used for acquiring a building query type input by a user;
the computing unit is used for inputting the images in the building query formula and the query database into a building recognition model established by the training device and converting the images into respective building characteristic vectors through computation;
the query database is used for storing national well-known building information and images thereof;
the judging unit is used for identifying and judging the buildings according to the building characteristic vectors obtained through calculation, and the same buildings are determined when the distance between the building characteristic vectors of the building query type and the building characteristic vectors of the images in the query database is smaller than a threshold value, otherwise, different buildings are determined;
a result output unit for providing a result of the identification judgment to the user.
Firstly, selecting a proper APN training sample to train a convolutional neural network, converting shallow network characteristics into building characteristics, and forming a building identification model; and inputting the building query expression input by the user and the image in the query database into the model, obtaining the building characteristic vector of the building query expression, judging and identifying the distance between the building characteristic vector and the building characteristic vector of the image in the query database, and feeding back the result to the user. Different from the existing picture recognition technology, the invention performs feature extraction in a specific field aiming at the building, reduces the difficulty of relevant parameter adjustment, and forms a perfect network based on convolutional neural training for building recognition. And the HNSW algorithm is adopted to optimize the building characteristic vector, so that the operation complexity and the calculation time are reduced, and the building identification and judgment are simple and quick.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the principles of the invention and not to limit the invention. In the drawings:
FIG. 1 is a flow chart illustrating an exemplary training process of a convolutional neural network in an embodiment of the present invention;
FIG. 2 is an exemplary process of building identification via a building query in an embodiment of the present invention;
FIG. 3 is a training apparatus for applying a building identification model according to an embodiment of the present invention;
fig. 4 is a diagram illustrating a building identification apparatus according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Before discussing exemplary embodiments in greater detail, it should be noted that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although a flowchart may describe the operations or steps as a sequential process, many of the operations can be performed in parallel, concurrently, or simultaneously. In addition, the order of the operations may be re-arranged. An associated process or step may be terminated when its operations are completed, but may have additional steps not included in the figure. The processes or steps may correspond to methods, functions, procedures, subroutines, and so on.
The embodiment provides a building recognition model convolutional neural network model establishment method, a convolutional neural network training method and an object recognition method based on the convolutional neural network, wherein the building recognition model convolutional neural network model establishment method comprises the following steps:
example 1
A building identification model establishing method comprises the following steps:
training a convolutional neural network by using APN data samples;
taking the trained convolutional neural network as a building recognition model:
figure 1 illustrates an example flow of a training process of APN data samples to a convolutional neural network. In various implementations of the example flow, steps may be deleted, combined, or divided into sub-steps. The example process may include a preparation phase and a training phase.
In the preparation stage, training sample data is required to be prepared, wherein the training sample data includes a large amount of multi-channel data, such as multi-channel image samples of thousands of orders of magnitude, and correct recognition results corresponding to each sample are marked. In this embodiment, the APN data sample includes a training building query formula, a positive sample building picture, and other building pictures;
in the training phase, known convolutional neural networks are used for training through samples in the input APN data samples. The training of the convolutional neural network by using the APN data sample comprises the steps of respectively receiving information of a training building query type + positive sample building picture, a training building query type + other building pictures by using a Simase network, and performing model training.
Specifically, the training of the convolutional neural network by using the APN data samples comprises the following steps 110-150:
in step 110, a set number of APN training samples are selected;
it should be noted that a deep learning calculation is needed to train a model for analyzing whether buildings in two different pictures are the same building, a large number of pictures are needed to train the model, and which pictures belong to the same building are marked, and the larger the data size is, the more accurate the data marking is, the higher the accuracy of the trained model is. The crawler program can be used together with manual work to arrange and process high-quality standard data to form an APN database. APN data samples in the APN database comprise training building query formulas, positive sample building pictures and other building pictures.
In step 120, inputting the APN training sample to a convolutional neural network to calculate a general feature; specifically, the common features include features in a shallow network layer, the features in the shallow network layer include image line trends, image inflection points and image fixed point features, and the features in the deep network layer include image categories and styles.
In step 130, the generic features are converted to architectural features using fully-connected neural network computations;
in step 140, the minimization error determination model building features of the APN training samples are trained.
Specifically, the vector algorithm for determining the model architectural features is L (a, P, N) ═ max (| f (a) -f (P) | ^2- | | f (a) -f (N) | | ^2+ α, 0);
wherein, L (A, P, N) is an error caused by an APN training sample; f (A), f (P), f (N) respectively represent training building query formulas, positive sample building pictures and training building feature vectors obtained after other building pictures are trained; alpha is a positive number and is used for increasing the discrimination;
if the distance between the feature vectors of f (A) and f (P) is less than the distance between the feature vectors of f (A) and f (N), the training is finished, or if the distance between the feature vectors of f (A) and f (P) is greater than or equal to the distance between the feature vectors of f (A) and f (N), the number of sample pictures corresponding to each picture type is increased, so as to perform the training again.
Finally, in step 150, when the calculated loss is small enough, or the loss does not change for a long time, the convolutional neural network training process ends. Typically, the entire data set has been read hundreds, thousands, or more times at the end of the training.
It should be noted that the generic feature vector is converted into the building feature vector when passing through the fully-connected neural network, which is determined by the loss function. The universal characteristic vector can be converted into an arbitrary vector through the fully-connected neural network, but the vector distance between the same building pictures is smaller due to the loss function, which means that the fully-connected neural network must extract some vectors capable of representing the building from the universal characteristic, so that similar vectors can be obtained when the same building universal vector passes through the fully-connected neural network. The vector is a set of fixed-length values, the error, i.e., the loss function, determines the fully-connected network, and this step is performed in a training phase, and the parameters in the fully-connected neural network are continuously updated until the parameters minimize the loss function. Wherein, the parameters in the fully connected neural network determine how to calculate the building feature vector from the general feature vector.
Example 2
A building identification method, comprising:
acquiring a building query formula input by a user;
obtaining a building feature vector;
and (5) identifying and judging the building.
FIG. 2 illustrates an example flow of building identification through a trained convolutional neural network by inputting a building query. In various implementations of the example flow, steps may be deleted, combined, or divided into sub-steps.
Specifically, the building identification method comprises the following steps 210-250:
in step 210, obtaining a building query formula input by a user; specifically, the building query formula is a building picture, and may also be a description of the picture or a part of the whole building picture.
In the step 220-240, the building query is input into a training model, and a building feature vector of the building query is extracted and converted into through a building recognition model;
in step 250, the same building is found when the distance between the building feature vector of the building query formula and the building feature vector of the image in the query database is smaller than the threshold value, otherwise, the same building is found. Specifically, the building identification and judgment adopts an HNSW algorithm to optimize the building feature vector.
The threshold may be set randomly, and the calculation and the judgment are performed by calculation to obtain the optimal threshold. For example, different thresholds are randomly selected, the accuracy of the threshold in the test set is calculated, and the threshold which can lead the accuracy of the test set to be the highest is selected. Wherein, the test set can be an APN data set, and the precision thereof is as follows: correctly distinguish the number/total number of the same building.
Example 3
As shown in fig. 3, a training apparatus using a building recognition model includes:
the APN data sample comprises a training building query formula, a positive sample building picture and other building pictures;
a training unit for training a convolutional neural network using the sample training library;
and the model generation unit is used for generating a building identification model by the trained convolutional neural network.
Specifically, the APN data sample is obtained by a crawler program or a manual marking or a crawler program assisted by a manual marking.
Specifically, the training unit includes:
the classification extraction unit is used for extracting general features by using the building identification model and extracting network shallow features;
a computing unit, configured to convert the vector of the general feature into a building feature vector, where an algorithm of the building feature vector is L (a, P, N) max (| f (a) -f (P) | | ^2- | | f (a) -f (N) | | | ^2+ α,0),
wherein, L (A, P, N) is an error caused by an APN training sample; f (A), f (P), f (N) respectively represent training building query formulas, positive sample building pictures and training building feature vectors obtained after other building pictures are trained; alpha is a positive number and is used for increasing the discrimination;
and the judging unit is used for judging whether the convolutional neural network meets the set requirement.
Example 4
As shown in fig. 4, a building recognition apparatus includes:
the building query type acquisition unit is used for acquiring a building query type input by a user;
the computing unit is used for inputting the images in the building query formula and the query database into a building recognition model established by a training device and converting the images into respective building characteristic vectors through computation;
the query database is used for storing national well-known building information and images thereof;
the judging unit is used for identifying and judging the buildings according to the building characteristic vectors obtained through calculation, and the same buildings are determined when the distance between the building characteristic vectors of the building query type and the building characteristic vectors of the images in the query database is smaller than a threshold value, otherwise, different buildings are determined;
a result output unit for providing a result of the identification judgment to the user.
Finally, it should be noted that: although the present invention has been described in detail with reference to the foregoing embodiments, it will be apparent to those skilled in the art that changes may be made in the embodiments and/or equivalents thereof without departing from the spirit and scope of the invention. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (7)
1. A building identification model establishing method is characterized by comprising the following steps:
training a convolutional neural network by using APN data samples;
taking the trained convolutional neural network as a building recognition model:
the APN data sample comprises a training building query formula, a positive sample building picture and other building pictures;
the training of the convolutional neural network by using the APN data sample comprises the steps of respectively receiving information of a training building query formula and a positive sample building picture, and information of the training building query formula and other building pictures by using a Simase network, and performing model training;
the training of the convolutional neural network using APN data samples comprises:
selecting APN training samples with a set number;
inputting the APN training sample into the convolutional neural network to calculate general features;
converting the general characteristics into architectural characteristics by using full-connection neural network calculation;
training to obtain a minimized error determination model building characteristic of an APN training sample;
the general features comprise features in a network shallow layer and/or a network deep layer, the features in the network shallow layer comprise image line trends, image inflection points and image fixed point features, and the features in the network deep layer comprise image categories and styles;
the vector algorithm for determining the model building features is L (A, P, N) ═ max (| | f (A) -f (P) | | ^2- | | | f (A) -f (N) | | | ^2+ alpha, 0);
wherein, L (A, P, N) is an error caused by an APN training sample; f (A), f (P), f (N) respectively represent training building query formulas, positive sample building pictures and training building feature vectors obtained after other building pictures are trained; alpha is a positive number and is used for increasing the discrimination;
if the distance between the feature vectors of f (A) and f (P) is less than the distance between the feature vectors of f (A) and f (N), the training is finished, or if the distance between the feature vectors of f (A) and f (P) is greater than or equal to the distance between the feature vectors of f (A) and f (N), the number of sample pictures corresponding to each picture type is increased, so as to perform the training again.
2. A building identification method, comprising:
acquiring a building query formula input by a user;
inputting the images in the building query formula and query database into a building identification model established by the method of claim 1, and converting the images into respective building feature vectors;
and (4) building identification judgment, wherein the same building is determined when the distance between the building characteristic vector of the building query type and the building characteristic vector of the image in the query database is smaller than a threshold value, and otherwise, different buildings are determined.
3. The method of claim 2, wherein the building identification decision optimizes a building feature vector using a HNSW algorithm.
4. A training device for building recognition models created by applying the method of claim 1, comprising:
the APN data sample comprises a training building query formula, a positive sample building picture and other building pictures;
a training unit for training a convolutional neural network using the sample training library;
and the model generation unit is used for generating a building identification model by the trained convolutional neural network.
5. Training device according to claim 4, wherein the APN data samples are obtained by means of a crawler program or a manual tagging or a crawler program assisted by a manual tagging.
6. Training device according to claim 4, wherein the training unit comprises:
the classification extraction unit is used for extracting general features by using the building identification model, and the general features are network shallow features;
a computing unit, configured to convert the vector of the general feature into a building feature vector, where an algorithm of the building feature vector is L (a, P, N) max (| f (a) -f (P) | | ^2- | | f (a) -f (N) | | | ^2+ α,0),
wherein, L (A, P, N) is an error caused by an APN training sample; f (A), f (P), f (N) respectively represent training building query formulas, positive sample building pictures and training building feature vectors obtained after other building pictures are trained; alpha is a positive number and is used for increasing the discrimination;
the judging unit is used for judging whether the convolutional neural network meets the setting requirement, wherein the setting requirement means that the distance between the characteristic vectors of f (A) and f (P) is required to be smaller than the distance between the characteristic vectors of f (A) and f (N).
7. A building identification device, comprising:
the building query type acquisition unit is used for acquiring a building query type input by a user;
a calculation unit, configured to input the building query expression and the image in the query database into the building recognition model established by the training apparatus of claim 5 or 6, and convert the building recognition model into respective building feature vectors through calculation;
the query database is used for storing national well-known building information and images thereof;
the judging unit is used for identifying and judging the buildings according to the building characteristic vectors obtained through calculation, and the same buildings are determined when the distance between the building characteristic vectors of the building query type and the building characteristic vectors of the images in the query database is smaller than a threshold value, otherwise, different buildings are determined;
a result output unit for providing a result of the identification judgment to the user.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910500682.5A CN110309727B (en) | 2019-06-11 | 2019-06-11 | Building identification model establishing method, building identification method and building identification device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910500682.5A CN110309727B (en) | 2019-06-11 | 2019-06-11 | Building identification model establishing method, building identification method and building identification device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110309727A CN110309727A (en) | 2019-10-08 |
| CN110309727B true CN110309727B (en) | 2022-02-18 |
Family
ID=68077097
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910500682.5A Active CN110309727B (en) | 2019-06-11 | 2019-06-11 | Building identification model establishing method, building identification method and building identification device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110309727B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113269223B (en) * | 2021-03-16 | 2022-04-22 | 重庆市地理信息和遥感应用中心 | City style classification method based on spatial culture modular factorial analysis |
| CN115394036A (en) * | 2022-08-24 | 2022-11-25 | 爱瑞克(大连)安全技术集团有限公司 | Monitoring and early warning method and system for building fire |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104899562A (en) * | 2015-05-29 | 2015-09-09 | 河南理工大学 | Texture segmentation and fusion based radar remote-sensing image artificial building recognition algorithm |
| CN109002784A (en) * | 2018-06-29 | 2018-12-14 | 国信优易数据有限公司 | The training method and system of streetscape identification model, streetscape recognition methods and system |
| CN109753928A (en) * | 2019-01-03 | 2019-05-14 | 北京百度网讯科技有限公司 | The recognition methods of architecture against regulations object and device |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9965717B2 (en) * | 2015-11-13 | 2018-05-08 | Adobe Systems Incorporated | Learning image representation by distilling from multi-task networks |
| US10380459B2 (en) * | 2016-08-30 | 2019-08-13 | Imagry (Israel) Ltd. | System and method for image classification |
-
2019
- 2019-06-11 CN CN201910500682.5A patent/CN110309727B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104899562A (en) * | 2015-05-29 | 2015-09-09 | 河南理工大学 | Texture segmentation and fusion based radar remote-sensing image artificial building recognition algorithm |
| CN109002784A (en) * | 2018-06-29 | 2018-12-14 | 国信优易数据有限公司 | The training method and system of streetscape identification model, streetscape recognition methods and system |
| CN109753928A (en) * | 2019-01-03 | 2019-05-14 | 北京百度网讯科技有限公司 | The recognition methods of architecture against regulations object and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110309727A (en) | 2019-10-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108898137B (en) | Natural image character recognition method and system based on deep neural network | |
| CN108399428B (en) | Triple loss function design method based on trace ratio criterion | |
| CN113240691B (en) | Medical image segmentation method based on U-shaped network | |
| WO2020253127A1 (en) | Facial feature extraction model training method and apparatus, facial feature extraction method and apparatus, device, and storage medium | |
| EP3772036A1 (en) | Detection of near-duplicate image | |
| CN111950453A (en) | Optional-shape text recognition method based on selective attention mechanism | |
| CN110781829A (en) | Light-weight deep learning intelligent business hall face recognition method | |
| CN112949408B (en) | Real-time identification method and system for target fish passing through fish channel | |
| US20230041943A1 (en) | Method for automatically producing map data, and related apparatus | |
| CN110309727B (en) | Building identification model establishing method, building identification method and building identification device | |
| CN114329034A (en) | Image text matching discrimination method and system based on fine-grained semantic feature difference | |
| CN111079826B (en) | Construction progress real-time identification method integrating SLAM and image processing | |
| CN111144462A (en) | Unknown individual identification method and device for radar signals | |
| CN116468935A (en) | Multi-core convolutional network-based stepwise classification and identification method for traffic signs | |
| CN112037180A (en) | Chromosome segmentation method and device | |
| CN108428234B (en) | Interactive segmentation performance optimization method based on image segmentation result evaluation | |
| CN112560925A (en) | Complex scene target detection data set construction method and system | |
| CN109255794B (en) | Standard part depth full convolution characteristic edge detection method | |
| CN112446345B (en) | Low-quality three-dimensional face recognition method, system, equipment and storage medium | |
| CN115661904A (en) | Data labeling and domain adaptation model training method, device, equipment and medium | |
| CN113744205B (en) | End-to-end road crack detection system | |
| CN114943741A (en) | Visual SLAM method based on target detection and geometric probability in dynamic scene | |
| CN113450321A (en) | Single-stage target detection method based on edge detection | |
| CN112907541A (en) | Palm image quality evaluation model construction method and device | |
| CN110796150A (en) | Image emotion recognition method based on emotion significant region detection |
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 |