CN110414561A - A kind of construction method of the natural scene data set suitable for machine vision - Google Patents
A kind of construction method of the natural scene data set suitable for machine vision Download PDFInfo
- Publication number
- CN110414561A CN110414561A CN201910561320.7A CN201910561320A CN110414561A CN 110414561 A CN110414561 A CN 110414561A CN 201910561320 A CN201910561320 A CN 201910561320A CN 110414561 A CN110414561 A CN 110414561A
- Authority
- CN
- China
- Prior art keywords
- data set
- classification
- picture
- ecosystem
- subclass
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/55—Clustering; Classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/951—Indexing; Web crawling techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a kind of construction methods of natural scene data set suitable for machine vision, include the following steps: the class categories that data set is determined based on ecosystem-type；Original image is downloaded from internet using keyword；Preliminary matches judgement is carried out to the image of downloading, subseries again is carried out to fuzzy category picture, forms image data set；Data set is verified using convolutional neural networks, ambiguity classification is merged further according to confusion matrix.It solves the problems, such as that natural scene classification is difficult to carry out accurate objective definition as object classification, provides reliable data supporting for training depth convolutional neural networks, can be used for natural scene identifying system.
Description
Technical field
The present invention relates to technical field of image processing, and in particular to a kind of natural scene data set suitable for machine vision
Construction method.
Background technique
In technical field of image processing, image recognition was receive more and more attention in recent years, and image recognition technology is
The basis of the practical techniques such as motion analysis, stereoscopic vision, data fusion, wherein target identification is due to convolutional neural networks and super large
The application of scale data collection has very big promotion in terms of the accuracy rate of identification, and utilizes convolutional neural networks to image
The scene of generation is identified the accuracy for being but difficult to obtain phase same level, among these greatly the reason is that due to scene number
Classify caused by inaccuracy according to itself is collected.
Being identified using machine vision to objects in images can be provided probably just in what happened in image, and
The scene that image occurs is identified, us can be made to obtain more information related with picture, while guessing by scene
Go that may occur with future.With the development of automatic Pilot, unmanned vehicle and unmanned air vehicle technique, to ambient enviroment
Progress accurately is detected as providing guarantee safely, therefore the scene database of a standard is particularly important.
For most of scene databases, often there is following two: 1) level is chaotic between classification and classification,
There is the confusion between " farmland " and " Plain " classification in certain data sets, wherein Plain is a kind of geomorphic type, and farmland
It can also be established on Plain, between the two and dependent；2) meaning repeats to obscure, such as " forest farm " and " forest " are only in spy
It just will do it differentiation in different situation.
Have an objective appraisal different from class label in the building process of object data set, contextual data collection often due to
Subjective assessment is different and has different criterion to same piece image, this also results in poor between image in same category
It is different very big, the promotion of algorithm accuracy is limited, therefore need to establish a unified sorting criterion in scene data set.
For a natural scene, biology and environment help the concrete type we determined that scene, and biology and environment structure
At it is unified whole be referred to as the ecosystem in ecological study, these ecosystems have the most a variety of type and big
It is small, the classification in natural scene data set is corresponded to by the classification of the ecosystem, the classification of data set is made to meet one
A unified decision metrics.
Determine natural scene belonging to an image, needs to combine object and background in image；Determine a certain space institute
The ecosystem of category needs to combine the biology and environment in the space；The two has great similitude in judgement, therefore should
Invention proposes that i.e. classification in the ecosystem corresponds to natural scene data using the ecosystem as the decision metrics of categories of datasets
The classification of concentration, in this way operation are all almost the same either from the subjective feeling of classification or in objective requirement.
Summary of the invention
To solve the above-mentioned problems, the present invention provides a kind of methods for constructing natural scene database, it is intended to improve existing
There is defect in data set classification, so that having a unified sorting criterion when constructing the classification of natural scene data set, and makes
Final data set can preferably be suitable for field of machine vision.
To achieve the above object, technical solution of the present invention specifically includes the following steps:
1. a kind of construction method of the natural scene data set suitable for machine vision, which is characterized in that including following step
It is rapid:
Step 1: primarily determining the classification that contains in data set and corresponding subclass according to ecosystem-type；Wherein, raw
State system type is divided into:
Terrestrial ecosystems: including natural terrestrial ecosystems and artificial terrestrial ecosystems；
Aquatic ecosystem: including inland aquatic ecosystem, marine ecosystems and artificial aquatic ecosystem；
Belowground ecosystem: including the cave ecosystem；
Step 2: subclass name being referred to as keyword and is scanned on the internet, 1000 and the subclass are at least searched for
Not corresponding picture, completes the collection of picture；
Step 3: preliminary artificial screening is carried out to collected picture, to the picture for not meeting corresponding classification carry out into
The screening of one step, obtains the data set containing scene type label；
Step 4: data set is divided into training set and verifying collects, is classified using depth convolutional neural networks to image,
And prediction result is generated into similarity matrix in conjunction with true class label using confusion matrix, merge phase according to similarity matrix
High classification is seemingly spent, final data set is obtained, specifically includes:
Step 4.1, raw data set is divided at random to obtain training set and verifying according to 80%/20% ratio
Collection；
Step 4.2, the picture in training set is converted to the picture of 224 × 224 resolution ratio, selects 13 layers of bn-
Inception network structure is as scene prediction sorter network；Containing there are two the original bn- of convolutional layer and 3 × 3 pond layers
Inception network converts the picture of 224 × 224 resolution ratio to 28 × 28 characteristic pattern, so that at subsequent 10
It can be handled faster in inception layers, wherein two layers of step-length is 2, remaining step-length is 1, therefore 28 × 28 feature
The characteristic pattern that the final variation of figure is 7 × 7；Finally the activation across Spatial Dimension is polymerize using global mean value pond layer；
Step 4.3, classification prediction is carried out to verifying collection using model obtained in step 4.2, obtains classification results, it will
Prediction result is combined to obtain confusion matrix C with true classification, intersects mistake between the confusion matrix Display Category and classification, hidden
The similarity degree between them is indicated containing ground, in form, we define this similitude with the following methods:
S=C+CT
Wherein C ∈ RN×NIt is confusion matrix, N is the number of natural scene classification, CijIt indicates the i-th class being classified as jth class
Probability, the value it is larger indicate two classifications between fuzziness it is higher；Similarity matrix S, the equation are generated according to the formula
It ensure that similitude is symmetrically to measure, give a clue to merge two similar categorizations；
Step 4.4, a suitable threshold tau is set, herein τ=0.5, two classes that similarity is more than the threshold tau is carried out
Merge；It analyzes combined reasonability and updates corresponding similarity matrix if the i-th row and jth row merge into same type, delete
I-th and jth row and column, it willIt is added as new row and column, repeats the step until final data set
No more than threshold value or meet other requirements, the data set obtained in this way is more suitable for the processing in terms of machine vision.
2. a kind of construction method of natural scene data set suitable for machine vision according to claim 1,
It is characterized in that, the ecosystem subclass classification will meet:
Condition one, subclass belong to a certain ecosystem；
It is mutually indepedent between condition two, subclass.
3. a kind of construction method of natural scene data set suitable for machine vision according to claim 1,
It is characterized in that, subclass is divided into:
Natural terrestrial ecosystems: polar region and alpine tundra, semidesert, coniferous forest, deciduous forest, natural grasslands, wasteland, often
Green broad-leaf forest, mediterranean bushes, desert, mountainous region, tropical rain forest；
Artificial terrestrial ecosystems: artificial pasture, farmland, greenhouse, city；
Inland aquatic ecosystem: marsh, wetland, river, lake；
Marine ecosystems: coastal, coral reef, deep-sea；
Artificial aquatic ecosystem: pond；
Include the cave ecosystem: cave.
4. a kind of construction method of natural scene data set suitable for machine vision according to claim 1,
Be characterized in that, item name described in step 2 add weather adjective, weather adjective include: it is sunny, cloudy, rain,
It is snowy.
5. a kind of construction method of natural scene data set suitable for machine vision according to claim 1,
It being characterized in that, the picture that corresponding classification is not met described in step 3, which is chosen, to summarize to unfiled picture, if always collecting picture is M,
Total classification is that N class constructs new classification when the picture that can not classify is more than N/M, and new classification meets the son stated in step 1
Category classification condition.
Therefore, the present invention has the advantage that picture sample in same subclass in the data set that (1) is constructed according to the method
The data set total quality of this diversification, building is higher；(2) it is not easy to obscure between different subclass, can be improved measuring accuracy.
Detailed description of the invention
Fig. 1 is the general frame of the invention.
Fig. 2 is 13 layers of bn-inception network structure selected in example.
Specific embodiment
For the clearer elaboration purpose of the present invention, technical scheme and beneficial effects, with reference to the accompanying drawings and examples
It further illustrates.It should be appreciated that the present invention should not necessarily be limited by embodiment disclosure, the scope of protection of present invention is with right
Subject to the range that claim defines.
As shown in Figure 1, the invention discloses a kind of construction method of natural scene data set suitable for machine vision, tool
The step of when body is implemented, is as follows:
Step 1: the classification contained in data set is primarily determined according to ecosystem-type:
Step 1.1, ecosystem-type can be divided into terrestrial ecosystems, aquatic ecosystem, belowground ecosystem most base
This three classifications；
Step 1.2, each classification can be classified again with further progress, and terrestrial ecosystems can be divided into nature land ecology again
System and artificial terrestrial ecosystems, aquatic ecosystem can be divided into inland aquatic ecosystem, marine ecosystems and artificial
Aquatic ecosystem, belowground ecosystem then include the cave ecosystem；
Step 1.3, above-mentioned classification can further separate subclass again, and wherein the classification of subclass can be according to required classification
Number formulated, classification meets two conditions: (1) subclass belongs between a certain ecosystem (2) subclass mutually solely
It is vertical；
Preferably, the subclass can be divided into: coastal, polar region and alpine tundra, marsh, semidesert, coniferous forest, fallen leaves
Woods, natural grasslands, wasteland, evergreen broadleaf forest, mediterranean bushes, desert, mountainous region, tropical rain forest, wetland, artificial pasture,
Farmland, greenhouse, river, lake, pond, coral reef, deep-sea, city, cave.
Step 2: corresponding picture is collected, the collection of picture is completed:
For one of classification C, search engine is inputted using C as keyword, is downloaded using online photographic search engine
Picture.In order to increase the diversity of picture, plus word related with weather (such as sunny, rain, cloudy) in keyword.
Undersized picture is excluded, the downloading of all categories picture is completed.
Step 3: collected picture is screened:
Step 3.1, artificial screening, the picture that retention class is consistent, the figure that classification is not inconsistent are carried out to the picture after downloading
Piece, which is chosen, to summarize to unfiled picture；
Step 3.2, remaining non-classified picture is classified, if the feelings properly classified can not be found by a large amount of pictures occur
Shape considers that the new classification of building, new classification meet the subclass class condition stated in step 1；
Step 3.3, the image of scene, which is rejected, to be judged to minority, completes the classification confirmation of all pictures.
Step 4: similar type is merged using similarity matrix:
Step 4.1, raw data set is divided at random to obtain training set and verifying according to 80%/20% ratio
Collection；
Step 4.2, the picture in training set is converted to the picture of 224 × 224 resolution ratio, as Fig. 2 selects 13 layers of bn-
Inception network structure is as scene prediction sorter network.Containing there are two the original bn- of convolutional layer and 3 × 3 pond layers
Inception network converts the picture of 224 × 224 resolution ratio to 28 × 28 characteristic pattern, and the characteristic pattern of small size makes
It can be handled faster in subsequent 10 inception layers, wherein two layers step-length is 2, remaining step-length is 1, therefore 28 ×
The characteristic pattern that 28 characteristic pattern finally changes as 7 × 7.Finally the activation across Spatial Dimension is carried out using global mean value pond layer
Polymerization；
Step 4.3, classification prediction is carried out to verifying collection using model obtained in step 4.2, obtains classification results, it will
Prediction result is combined to obtain confusion matrix C with true classification, intersects mistake between the confusion matrix Display Category and classification, hidden
The similarity degree between them is indicated containing ground, in form, we define this similitude with the following methods:
S=C+CT
Wherein C ∈ RN×NIt is confusion matrix, N is the number of natural scene classification, CijIt indicates the i-th class being classified as jth class
Probability, the value it is larger indicate two classifications between fuzziness it is higher.Similarity matrix S, the equation are generated according to the formula
It ensure that similitude is symmetrically to measure, give a clue to merge two similar categorizations；
Step 4.4, a suitable threshold tau is set, herein τ=0.5, two classes that similarity is more than the threshold tau is carried out
Merge.It analyzes combined reasonability and updates corresponding similarity matrix if the i-th row and jth row merge into same type, delete
I-th and jth row and column, it willIt is added as new row and column, repeats the step until final data set
No more than threshold value or meet other requirements, the data set obtained in this way is more suitable for the processing in terms of machine vision.
Specific embodiment described herein is only an example for the spirit of the invention.The neck of technology belonging to the present invention
The technical staff in domain can make various modifications or additions to the described embodiments or replace by a similar method
In generation, however, it does not deviate from the spirit of the invention or beyond the scope of the appended claims.
Claims (5)
1. a kind of construction method of the natural scene data set suitable for machine vision, which comprises the following steps:
Step 1: primarily determining the classification that contains in data set and corresponding subclass according to ecosystem-type；Wherein, ecosystem
System type is divided into:
Terrestrial ecosystems: including natural terrestrial ecosystems and artificial terrestrial ecosystems；
Aquatic ecosystem: including inland aquatic ecosystem, marine ecosystems and artificial aquatic ecosystem；
Belowground ecosystem: including the cave ecosystem；
Step 2: subclass name being referred to as keyword and is scanned on the internet, 1000 and the subclass pair are at least searched for
The picture answered completes the collection of picture；
Step 3: carrying out preliminary artificial screening to collected picture, the picture for not meeting corresponding classification is carried out further
Screening, obtains the data set containing scene type label；
Step 4: data set being divided into training set and verifying collects, is classified using depth convolutional neural networks to image, and benefit
With confusion matrix prediction result is generated into similarity matrix in conjunction with true class label, merges similarity according to similarity matrix
Excessively high classification obtains final data set, specifically includes:
Step 4.1, raw data set is divided to obtain training set at random according to 80%/20% ratio and verifying collects；
Step 4.2, the picture in training set is converted to the picture of 224 × 224 resolution ratio, selects 13 layers of bn-inception
Network structure is as scene prediction sorter network；Containing there are two the original bn-inception networks of convolutional layer and 3 × 3 pond layers
The picture of 224 × 224 resolution ratio is converted to 28 × 28 characteristic pattern, so that can be more in subsequent 10 inception layers
Fast handle, wherein two layers of step-length is 2, the spy that the characteristic pattern that remaining step-length is 1, therefore 28 × 28 finally changes as 7 × 7
Sign figure；Finally the activation across Spatial Dimension is polymerize using global mean value pond layer；
Step 4.3, classification prediction is carried out to verifying collection using model obtained in step 4.2, obtains classification results, will predicts
As a result it is combined to obtain confusion matrix C with true classification, intersects mistake between the confusion matrix Display Category and classification, impliedly
The similarity degree between them is indicated, in form, we define this similitude with the following methods:
S=C+CT
Wherein C ∈ RN×NIt is confusion matrix, N is the number of natural scene classification, CijIt indicates the i-th class being classified as the general of jth class
Rate, the value is larger to indicate that the fuzziness between two classifications is higher；Similarity matrix S is generated according to the formula, the equation guarantees
Similitude is symmetrically to measure, and is given a clue to merge two similar categorizations；
Step 4.4, a suitable threshold tau is set, herein τ=0.5, two classes that similarity is more than the threshold tau is merged；
Combined reasonability is analyzed, if the i-th row and jth row merge into same type, corresponding similarity matrix is updated, deletes the i-th He
Jth row and column, willIt is added as new row and column, repeats the step until final data set is no more than
Threshold value meets other requirements, and the data set obtained in this way is more suitable for the processing in terms of machine vision.
2. a kind of construction method of natural scene data set suitable for machine vision according to claim 1, feature
It is, the ecosystem subclass classification will meet:
Condition one, subclass belong to a certain ecosystem；
It is mutually indepedent between condition two, subclass.
3. a kind of construction method of natural scene data set suitable for machine vision according to claim 1, feature
It is, subclass is divided into:
Natural terrestrial ecosystems: polar region and alpine tundra, semidesert, coniferous forest, deciduous forest, natural grasslands, wasteland, evergreen wealthy
Ye Lin, mediterranean bushes, desert, mountainous region, tropical rain forest；
Artificial terrestrial ecosystems: artificial pasture, farmland, greenhouse, city；
Inland aquatic ecosystem: marsh, wetland, river, lake；
Marine ecosystems: coastal, coral reef, deep-sea；
Artificial aquatic ecosystem: pond；
Include the cave ecosystem: cave.
4. a kind of construction method of natural scene data set suitable for machine vision according to claim 1, feature
Be, item name described in step 2 add weather adjective, weather adjective include: it is sunny, cloudy, rain, snow
's.
5. a kind of construction method of natural scene data set suitable for machine vision according to claim 1, feature
It is, the picture that corresponding classification is not met described in step 3, which is chosen, to summarize to unfiled picture, if always collecting picture is M, total class
Not Wei N class construct new classification when the picture that can not classify is more than N/M, new classification meets the subclass stated in step 1
Class condition.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910561320.7A CN110414561A (en) | 2019-06-26 | 2019-06-26 | A kind of construction method of the natural scene data set suitable for machine vision |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910561320.7A CN110414561A (en) | 2019-06-26 | 2019-06-26 | A kind of construction method of the natural scene data set suitable for machine vision |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110414561A true CN110414561A (en) | 2019-11-05 |
Family
ID=68359871
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910561320.7A Pending CN110414561A (en) | 2019-06-26 | 2019-06-26 | A kind of construction method of the natural scene data set suitable for machine vision |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110414561A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111062441A (en) * | 2019-12-18 | 2020-04-24 | 武汉大学 | Scene classification method and device based on self-supervision mechanism and regional suggestion network |
| CN111104898A (en) * | 2019-12-18 | 2020-05-05 | 武汉大学 | Image scene classification method and device based on target semantics and attention mechanism |
| CN111444819A (en) * | 2020-03-24 | 2020-07-24 | 北京百度网讯科技有限公司 | Cutting frame determining method, network training method, device, equipment and storage medium |
| CN112231535A (en) * | 2020-10-23 | 2021-01-15 | 山东科技大学 | Method for making multi-modal data set in field of agricultural diseases and insect pests, processing device and storage medium |
| CN112766427A (en) * | 2021-04-06 | 2021-05-07 | 浙江口碑网络技术有限公司 | Training method, device and equipment of image recognition model |
| CN113569986A (en) * | 2021-08-18 | 2021-10-29 | 网易（杭州）网络有限公司 | Computer vision data classification method and device, electronic equipment and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110026840A1 (en) * | 2009-07-28 | 2011-02-03 | Samsung Electronics Co., Ltd. | System and method for indoor-outdoor scene classification |
| CN103390046A (en) * | 2013-07-20 | 2013-11-13 | 西安电子科技大学 | Multi-scale dictionary natural scene image classification method based on latent Dirichlet model |
| CN108921190A (en) * | 2018-05-24 | 2018-11-30 | 北京飞搜科技有限公司 | A kind of image classification method, device and electronic equipment |
| CN108921037A (en) * | 2018-06-07 | 2018-11-30 | 四川大学 | A kind of Emotion identification method based on BN-inception binary-flow network |
-
2019
- 2019-06-26 CN CN201910561320.7A patent/CN110414561A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110026840A1 (en) * | 2009-07-28 | 2011-02-03 | Samsung Electronics Co., Ltd. | System and method for indoor-outdoor scene classification |
| CN103390046A (en) * | 2013-07-20 | 2013-11-13 | 西安电子科技大学 | Multi-scale dictionary natural scene image classification method based on latent Dirichlet model |
| CN108921190A (en) * | 2018-05-24 | 2018-11-30 | 北京飞搜科技有限公司 | A kind of image classification method, device and electronic equipment |
| CN108921037A (en) * | 2018-06-07 | 2018-11-30 | 四川大学 | A kind of Emotion identification method based on BN-inception binary-flow network |
Non-Patent Citations (2)
| Title |
|---|
| YANFEI LIU等: "SCENE SEMANTIC CLASSIFICATION BASED ON RANDOMSCALE STRETCHED CONVOLUTIONAL NEURAL NETWORK FOR HIGHSPATIAL RESOLUTION REMOTE SENSING IMAGERY", 《IGARSS 2016》 * |
| 龚希等: "融合全局和局部深度特征的高分辨率遥感影像场景分类方法", 《光学学报》 * |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111062441A (en) * | 2019-12-18 | 2020-04-24 | 武汉大学 | Scene classification method and device based on self-supervision mechanism and regional suggestion network |
| CN111104898A (en) * | 2019-12-18 | 2020-05-05 | 武汉大学 | Image scene classification method and device based on target semantics and attention mechanism |
| CN111104898B (en) * | 2019-12-18 | 2022-03-25 | 武汉大学 | Image scene classification method and device based on target semantics and attention mechanism |
| CN111444819A (en) * | 2020-03-24 | 2020-07-24 | 北京百度网讯科技有限公司 | Cutting frame determining method, network training method, device, equipment and storage medium |
| CN111444819B (en) * | 2020-03-24 | 2024-01-23 | 北京百度网讯科技有限公司 | Cut frame determining method, network training method, device, equipment and storage medium |
| CN112231535A (en) * | 2020-10-23 | 2021-01-15 | 山东科技大学 | Method for making multi-modal data set in field of agricultural diseases and insect pests, processing device and storage medium |
| CN112231535B (en) * | 2020-10-23 | 2022-11-15 | 山东科技大学 | Method for making multi-modal data set in field of agricultural diseases and insect pests, processing device and storage medium |
| CN112766427A (en) * | 2021-04-06 | 2021-05-07 | 浙江口碑网络技术有限公司 | Training method, device and equipment of image recognition model |
| CN113569986A (en) * | 2021-08-18 | 2021-10-29 | 网易（杭州）网络有限公司 | Computer vision data classification method and device, electronic equipment and storage medium |
| CN113569986B (en) * | 2021-08-18 | 2023-06-30 | 网易（杭州）网络有限公司 | Computer vision data classification method, device, electronic equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110414561A (en) | A kind of construction method of the natural scene data set suitable for machine vision | |
| CN109948425B (en) | Pedestrian searching method and device for structure-aware self-attention and online instance aggregation matching | |
| CN106909924B (en) | Remote sensing image rapid retrieval method based on depth significance | |
| CN109993072B (en) | Low-resolution pedestrian re-identification system and method based on super-resolution image generation | |
| CN106682233A (en) | Method for Hash image retrieval based on deep learning and local feature fusion | |
| CN109102014A (en) | The image classification method of class imbalance based on depth convolutional neural networks | |
| CN110533631A (en) | SAR image change detection based on the twin network of pyramid pondization | |
| CN110246141B (en) | Vehicle image segmentation method based on joint corner pooling under complex traffic scene | |
| CN111310756B (en) | Damaged corn particle detection and classification method based on deep learning | |
| CN106096661B (en) | The zero sample image classification method based on relative priority random forest | |
| CN103324677B (en) | Hierarchical fast image global positioning system (GPS) position estimation method | |
| CN110751209B (en) | Intelligent typhoon intensity determination method integrating depth image classification and retrieval | |
| CN104268552B (en) | One kind is based on the polygonal fine classification sorting technique of part | |
| CN114241273A (en) | Multi-modal image processing method and system based on Transformer network and hypersphere space learning | |
| CN115995018A (en) | Long tail distribution visual classification method based on sample perception distillation | |
| CN110688888B (en) | Pedestrian attribute identification method and system based on deep learning | |
| CN107958067A (en) | It is a kind of based on without mark Automatic Feature Extraction extensive electric business picture retrieval system | |
| CN110222767A (en) | Three-dimensional point cloud classification method based on nested neural and grating map | |
| CN111652835A (en) | Method for detecting insulator loss of power transmission line based on deep learning and clustering | |
| CN112528934A (en) | Improved YOLOv3 traffic sign detection method based on multi-scale feature layer | |
| CN116091946A (en) | Yolov 5-based unmanned aerial vehicle aerial image target detection method | |
| CN115690549A (en) | Target detection method for realizing multi-dimensional feature fusion based on parallel interaction architecture model | |
| CN107704509A (en) | A kind of method for reordering for combining stability region and deep learning | |
| CN113220915B (en) | Remote sensing image retrieval method and device based on residual attention | |
| CN110188811A (en) | Underwater target detection method based on normed Gradient Features and convolutional neural networks |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191105 |
|
| RJ01 | Rejection of invention patent application after publication |