CN106228185A - A kind of general image classifying and identifying system based on neutral net and method - Google Patents
A kind of general image classifying and identifying system based on neutral net and method Download PDFInfo
- Publication number
- CN106228185A CN106228185A CN201610571220.9A CN201610571220A CN106228185A CN 106228185 A CN106228185 A CN 106228185A CN 201610571220 A CN201610571220 A CN 201610571220A CN 106228185 A CN106228185 A CN 106228185A
- Authority
- CN
- China
- Prior art keywords
- parameter
- neutral net
- feature
- neural network
- image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
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
- 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
Abstract
The invention discloses a kind of general image classifying identification method based on neutral net and system, this system includes parameter collection module, expert system database, algorithm generation module and parameter adjustment module, by based on neutral net design template built-in in expert system database, according to the concrete application of user, automatically generate suitable neural network type and a whole set of corresponding algorithm, again by study and inspection, automatically explore parameter space, and adjust parameter to obtain optimal achievement.The present invention can be substantially reduced the design difficulty of neutral net, and user need not the most complicated specialty background, it is possible to directly uses.
Description
Technical field
The present invention relates to pattern recognition, particularly to a kind of general image classifying and identifying system based on neutral net and side
Method.
Background technology
Images steganalysis (Image Classification) is always a core topic of computer vision field,
Image recognition technology through the development come 60 years more, and method based on neutral net has had evolved to one in field of image recognition
Individual brand-new height.But, training pattern still needs the most artificial intervention, especially as the system of this complexity of CNN, as
What selects suitable feature, how to design convolution kernel, and how design feature is polymerized, and needs how many computation layer in hidden layer on earth,
Many times it is also based on experience.And based on neutral net, especially deep neural network, in design and use the most very
Complicated, need the strongest Professional knowledge, need research worker to have higher professional standards, be unfavorable for that technology is popularized.
Summary of the invention
For problem of the prior art, the present invention propose a kind of general image classifying and identifying system based on neutral net and
Method, based on built-in neutral net design template, according to the concrete application of user, automatically generates suitable neural network type
With a whole set of corresponding algorithm, then by study and inspection, automatically explore parameter space, and it is optimal to obtain to adjust parameter
Achievement.The present invention can be substantially reduced the design difficulty of neutral net, and user need not the most complicated specialty background, it is possible to straight
Connect use.
The present invention solves that above-mentioned technical problem be the technical scheme is that
A kind of general image classifying and identifying system based on neutral net, including parameter collection module, specialist system data
Storehouse, algorithm generation module and parameter adjustment module.
Parameter collection module is for gathering the problem parameter of target problem, and system is by asking that parameter collection module collects
Topic parameter determines problem scale and the complexity of image recognition；The problem parameter gathered includes: the classification number of image recognition
Amount, the quantity of sample image, the specification of sample image, whether it is shooting image and uses supervision law or non-supervisory
Method；The problem scale of image recognition refers mainly to classification and the amount of training data of study of classification；Described complexity refers to carry out
Feature kind required for image recognition.
Expert system database is according to the long-term achievement following the trail of up-to-date image recognition algorithm and to merge in vehicle identification,
Car license recognition, the technology of the field accumulation such as recognition of face, an expert database being specifically designed for neutral net of foundation.Its root
According to some typical image identification problems, store multiple neural network structure, feature extracting method and clustering method etc..Problem
The data of parameter acquisition are input to expert system database, and system can automatically select qualified neural network structure, feature
Extracting method and clustering method.Give a concrete illustration, Handwritten Digit Recognition problem, it is only necessary to identify that 0 to 9 have ten numbers altogether
Word, it is only necessary to 10 input neurons, problem scale is less, and system can select to use back transfer (Backward
The neutral net of single hidden layer Propagation).
Algorithm generation module generates according to neural network structure, feature extracting method and the clustering method of Systematic selection and is suitable for
In the neural network algorithm model of process target problem, determine the algorithm degree of depth according to problem scale simultaneously；If user is to algorithm
Model is unsatisfied with, can adjustment algorithm model voluntarily.After algorithm model determines, it is possible to start learning training data, and count
Calculate the parameter in model.
The character of the problem parameter that parameter adjustment module collects according to parameter collection module is tentatively set up neutral net and is calculated
The hunting zone of the key parameter of method model, and according to systematic training and test result, adjust the conversion of key parameter further
Scope, finds out optimal parameter combination and the scope of correct classification, to obtaining relatively stable and correct classification on new data
Result.Key parameter is exactly those parameters having a direct impact classification results.
The core concept of this module is that parameter is considered as a stochastic variable, and its span is to have certain statistical significance
's.Found out an excursion of all parameters by training data, adjust these further by test data the most again
Parameter area, to obtaining relatively stable and correct classification results.The purpose of do so is the parameter tried out by experience
Scope, by training, the step of test finds out the parameter combination of optimum.This can be greatly improved the speed of optimization neural network training
Degree.
A kind of general image classifying identification method based on neutral net, comprises the following steps:
Step 1, user, according to the target problem of image recognition, inputs concrete problem by parameter collection module to system
Parameter；
Step 2, system chooses corresponding neural network structure, spy according to the problems referred to above parameter from expert system database
Levy extracting method and feature clustering method, the preliminary neural network algorithm model set up for analyzing target problem, and the most true
Determine the algorithm degree of depth；
Step 3, learning training data, it is input to the mapping relations of classification information by image, determines neural network algorithm
Series of algorithms parameter in model；
Step 4, according to the character of the problem parameter of input, sets up the hunting zone of algorithm parameter in algorithm model, including
The degree of depth of neutral net, the size of convolution kernel, the parameter of feature extraction；
Step 5, tests neural network algorithm model by test data, needs to use all key parameters in test
Parameter in excursion, and find out optimal parameter combination range.On the premise of classification results is correct, retention parameter combines
And excursion.
The invention has the beneficial effects as follows: can be similar based on what specialist system was preserved according to the concrete application of user
The neural network model of problem, automatically generates suitable neural network type and a whole set of corresponding algorithm, then by study and
Inspection, explores parameter space automatically, and adjusts parameter to obtain optimal achievement, reduces manual intervention.Native system can be big
The big design difficulty reducing neutral net, user need not the most complicated specialty background, it is possible to directly uses.
Accompanying drawing explanation
A kind of based on neutral net the general image identification system architecture diagram that Fig. 1 provides for the present invention
Fig. 2 is a kind of general image recognition methods flow chart based on neutral net of the present invention
Detailed description of the invention
Below in conjunction with the accompanying drawings and embodiment the invention will be further described.
As it is shown in figure 1, the invention provides a kind of general image classifying and identifying system based on neutral net, including parameter
Acquisition module, expert system database, algorithm generation module and parameter adjustment module.Parameter collection module is used for gathering target and asks
The problem parameter of topic, by the problem parameter that parameter collection module collects, system determines that the problem scale of image recognition is with multiple
Miscellaneous degree；The problem parameter gathered includes: the categorical measure of image recognition, the quantity of sample image, the specification chi of sample image
Very little, whether be shooting image and use supervision law or non-supervisory method；The problem scale of image recognition refers mainly to the class of classification
Not with the amount of training data learnt；Described complexity refers to the feature kind carried out required for image recognition.
Expert system database is according to the long-term achievement following the trail of up-to-date image recognition algorithm and to merge in vehicle identification,
Car license recognition, the technology of the field accumulation such as recognition of face, an expert database being specifically designed for neutral net of foundation.Its root
According to some typical image identification problems, store multiple neural network structure, feature extracting method and clustering method etc..Problem
The data of parameter acquisition are input to expert system database, and system can automatically select qualified neural network structure, feature
Extracting method and clustering method.Give a concrete illustration, Handwritten Digit Recognition problem, it is only necessary to identify that 0 to 9 have ten numbers altogether
Word, it is only necessary to 10 input neurons, problem scale is less, and system can select to use back transfer (Backward
The neutral net of single hidden layer Propagation).
Algorithm generation module generates according to neural network structure, feature extracting method and the clustering method of Systematic selection and is suitable for
In the neural network algorithm model of process target problem, determine the algorithm degree of depth according to problem scale simultaneously；If user is to algorithm
Model is unsatisfied with, can adjustment algorithm model voluntarily.After algorithm model determines, it is possible to start learning training data, and count
Calculate the parameter in model.
The character of the problem parameter that parameter adjustment module collects according to parameter collection module is tentatively set up neutral net and is calculated
The hunting zone of the key parameter of method model, and according to systematic training and test result, adjust the conversion of key parameter further
Scope, finds out optimal parameter combination and the scope of correct classification, to obtaining relatively stable and correct classification on new data
Result.Key parameter be exactly those to classification results than more sensitive parameter.
The present invention also provides for a kind of general image classifying identification method based on neutral net, as in figure 2 it is shown, include following
Step:
Step 1, user, according to the target problem of image recognition, inputs concrete problem by parameter collection module to system
Parameter；
Step 2, system chooses corresponding neural network structure, spy according to the problems referred to above parameter from expert system database
Levy extracting method and feature clustering method, the preliminary neural network algorithm model set up for analyzing target problem, and the most true
Determine the algorithm degree of depth；
Step 3, learning training data, it is input to the mapping relations of classification information by image, determines neural network algorithm
Series of algorithms parameter in model；
Step 4, according to the character of the problem parameter of input, sets up the hunting zone of algorithm parameter in algorithm model, including
The degree of depth of neutral net, the size of convolution kernel, the parameter of feature extraction；
Step 5, tests neural network algorithm model by test data, needs to use all key parameters in test
Parameter in excursion, and find out optimal parameter combination range.On the premise of classification results is correct, retention parameter combines
And excursion.
With the concrete case of car identification, the present invention will be further described below:
Car in present case is identified by learning one section of image that have taken various car, trains mould by supervision law
Type understands car rule in the picture, then identifies car from new image.
First, the profile to image data acceptance of the bid caravan, then extracts this car image, all of car image all
Being planned for unified size, such as 64 pixels take advantage of 64 pixels.
Owing to the image data amount of study is relatively big, uses the template of convolutional neural networks, set up a hidden layer and comprise 6 layers
CNN.Here selecting 6 layers is based on having included a similar identification application, i.e. row in image in this specialist system
People identifies.These 6 layers is convolutional layer 1 (extraction feature) successively, pond layer 2 (cluster feature), convolutional layer 3, and pond layer 4 connects entirely
Layer 5 and classification layer 6.
It follows that be through learning training data to set up vehicle characteristics and the mapping relations of classification information.Namely
By learning this section of image, allow model learning to this car rule on image.Specifically, car image is input to convolutional layer
1, it is carried out the convolution operation that size is 7X7 pixel convolution kernel and step-length is 1 pixel, uses 16 different convolution kernels (special
Levy), it is thus achieved that the characteristic pattern of 16 58X58 pixel sizes.These characteristic patterns are input to pond layer 2, it is maximum pondization behaviour
Making, using pond core is 2X2 pixel, and step-length is 1 pixel.The step for exactly feature is further polymerized, the most originally
4 potting gum be 1 pixel, and take eigenvalue of maximum pixel, obtain the characteristic pattern of 16 29X29 pixel sizes.Then,
It is input to convolutional layer 3, employing 7X7 pixel convolution kernel and the convolution operation that step-length is 1 pixel, altogether uses 32 convolution kernels, this
Sample has just obtained the characteristic pattern of 32 23X23 pixels.It is input to pond layer 4, uses 2X2 pixel pond core and the pixel of step-length 1,
Obtain the characteristic pattern of 32 11X11 pixels.Being input to full articulamentum, dyad turns to 3872 dimensional feature vectors.By this feature
Vector is input to layer of classifying.Classification layer uses SoftMax classification function, is output as the class probability of this car type.
Under supervision law, as long as one can consider that output probability, more than 50%, the most correctly judges.So we can give
One excursion of parameter, parameter in this example is mainly kind and the size of convolution kernel (feature), the kind of Chi Huahe and
Size.Because the kind of Chi Huahe and size are relatively-stationary, mainly consider kind and the size of convolution kernel.Convolution kernel
Size can consider to use the scope of 5X5-13X13.Learning test data, if the classification accuracy rate identified is more than 50%,
Then can retain.Meanwhile, build the corresponding relation of a classification results and parameter variation range, therefrom find out classification accuracy rate relatively
High parameter combination and scope.By learning training data and test data, we are obtained with one with Parameters variation
The neural network model of scope.New image is input in this neural network model with parameter variation range, except
Obtain beyond the probability of classification, it is also possible to obtain an estimation to this probability accuracy.Since so, just improve classification
Accuracy.The process of whole parameter regulation is all automatically performed.
The part not illustrated in description is prior art or common knowledge.The present embodiment is merely to illustrate this invention,
Rather than restriction the scope of the present invention, the equivalent replacement that those skilled in the art are made for the present invention etc. is revised and is all considered
Fall in this invention claims institute protection domain.
Claims (7)
1. a general image classifying and identifying system based on neutral net, it is characterised in that: include parameter collection module, expert
System database, algorithm generation module and parameter adjustment module；
Described parameter collection module is for gathering the problem parameter of target problem, and system is by asking that parameter collection module collects
Topic parameter determines problem scale and the complexity of image recognition；
Described expert system database is used for storing multiple neural network structure, feature extracting method and clustering method, system
Qualified neural network structure, spy is selected from expert system database according to the problem parameter that parameter collection module collects
Levy extracting method and clustering method；
Described algorithm generation module generates according to neural network structure, feature extracting method and the clustering method of Systematic selection and is suitable for
In the neural network algorithm model of process target problem, determine the algorithm degree of depth according to problem scale simultaneously；
The character of the problem parameter that described parameter adjustment module collects according to parameter collection module primarily determines that neutral net is calculated
The series of algorithms parameter of method model, and key parameter is set excursion, and according to systematic training and test result, enter one
The transformation range of successive step key parameter, finds out optimal parameter combination and the scope of correct classification, to obtaining on new data
Relatively stable and correct classification results；Described key parameter refers to the parameter having a direct impact the result of classification.
A kind of general image classifying and identifying system based on neutral net the most according to claim 1, it is characterised in that: institute
The problem parameter stating parameter collection module collection includes: the categorical measure of image recognition, the quantity of sample image, sample image
Specification, whether it is shooting image and uses supervision law or non-supervisory method.
A kind of general image classifying and identifying system based on neutral net the most according to claim 1, it is characterised in that: institute
The problem scale stating image recognition refers mainly to classification and the amount of training data of study of classification；Described complexity refers to carry out figure
As the feature kind required for identification.
A kind of general image classifying and identifying system based on neutral net the most according to claim 1, it is characterised in that: institute
State the neural network structure of storage in expert system database to include:
(1) neutral net based on BP, for small-scale problem of image recognition；
(2) neutral net based on DBN, for medium scale problem of image recognition；
(3) neutral net based on CNN, for the problem of image recognition that problem scale is big；
(4) neutral net based on RNN, the problem of image recognition in time series.
A kind of general image classifying and identifying system based on neutral net the most according to claim 3, it is characterised in that: institute
The method of the feature extraction stating in expert system database storage includes: local feature region description vectors, local line's feature, locally
One or more in region feature, contour feature, region color feature histogram and frequency domain feature.
A kind of general image classifying and identifying system based on neutral net the most according to claim 4, it is characterised in that: institute
The method of the feature clustering of storage in expert system database of stating includes: the region histogram of characteristic vector and the system of characteristic vector
Evaluation.
7. a general image classifying identification method based on neutral net, it is characterised in that: comprise the following steps:
Step 1, user, according to the target problem of image recognition, is joined to the problem that system input is concrete by parameter collection module
Number；
Step 2, system chooses corresponding neural network structure according to the problems referred to above parameter from expert system database, feature carries
Access method and feature clustering method, the preliminary neural network algorithm model set up for analyzing target problem, and automatically determine calculation
The method degree of depth；
Step 3, learning training data, it is input to the mapping relations of classification information by image, determines neural network algorithm model
In series of algorithms parameter；
Step 4, according to the character of the problem parameter of input, sets up the hunting zone of algorithm parameter in algorithm model, including nerve
The degree of depth of network, the size of convolution kernel, the parameter of feature extraction；
Step 5, tests neural network algorithm model by test data, needs to use the change of all key parameters in test
In the range of parameter, and find out optimal parameter combination range.On the premise of classification results is correct, retention parameter combination and
Excursion.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610571220.9A CN106228185B (en) | 2016-07-20 | 2016-07-20 | A kind of general image classifying and identifying system neural network based and method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610571220.9A CN106228185B (en) | 2016-07-20 | 2016-07-20 | A kind of general image classifying and identifying system neural network based and method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106228185A true CN106228185A (en) | 2016-12-14 |
| CN106228185B CN106228185B (en) | 2019-10-15 |
Family
ID=57532063
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610571220.9A Expired - Fee Related CN106228185B (en) | 2016-07-20 | 2016-07-20 | A kind of general image classifying and identifying system neural network based and method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106228185B (en) |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108189043A (en) * | 2018-01-10 | 2018-06-22 | 北京飞鸿云际科技有限公司 | A kind of method for inspecting and crusing robot system applied to high ferro computer room |
| CN108256633A (en) * | 2018-02-06 | 2018-07-06 | 苏州体素信息科技有限公司 | A kind of method of test depth Stability of Neural Networks |
| CN108345661A (en) * | 2018-01-31 | 2018-07-31 | 华南理工大学 | A kind of Wi-Fi clustering methods and system based on extensive Embedding technologies |
| CN108734184A (en) * | 2017-04-17 | 2018-11-02 | 苏宁云商集团股份有限公司 | A kind of method and device that sensitive image is analyzed |
| WO2019011093A1 (en) * | 2017-07-12 | 2019-01-17 | 腾讯科技（深圳）有限公司 | Machine learning model training method and apparatus, and facial expression image classification method and apparatus |
| WO2019062534A1 (en) * | 2017-09-27 | 2019-04-04 | 深圳市商汤科技有限公司 | Image retrieval method, apparatus, device and readable storage medium |
| CN109658706A (en) * | 2018-12-18 | 2019-04-19 | 深圳市捷顺科技实业股份有限公司 | A kind of vehicle data processing method and relevant device |
| CN109710819A (en) * | 2018-12-29 | 2019-05-03 | 北京航天数据股份有限公司 | A kind of model display method, apparatus, equipment and medium |
| CN109726818A (en) * | 2018-12-29 | 2019-05-07 | 北京航天数据股份有限公司 | A kind of model editing method, apparatus, equipment and medium |
| CN109886206A (en) * | 2019-02-21 | 2019-06-14 | 电子科技大学中山学院 | Three-dimensional object identification method and equipment |
| CN110163231A (en) * | 2018-02-14 | 2019-08-23 | 倍加科技股份有限公司 | Establish the method, computer system and recording medium of data identification model |
| CN110348428A (en) * | 2017-11-01 | 2019-10-18 | 腾讯科技（深圳）有限公司 | Eye fundus image classification method, device and computer readable storage medium |
| CN110456955A (en) * | 2019-08-01 | 2019-11-15 | 腾讯科技（深圳）有限公司 | Exposure dress ornament detection method, device, system, equipment and storage medium |
| CN110770759A (en) * | 2017-05-19 | 2020-02-07 | 渊慧科技有限公司 | Neural network system |
| CN110892409A (en) * | 2017-06-05 | 2020-03-17 | 西门子股份公司 | Method and apparatus for analyzing images |
| CN111295676A (en) * | 2017-10-26 | 2020-06-16 | 罗伯特·博世有限公司 | Method and apparatus for automatically generating artificial neural network |
| CN111368909A (en) * | 2020-03-03 | 2020-07-03 | 温州大学 | Vehicle logo identification method based on convolutional neural network depth features |
| US11049017B2 (en) * | 2016-12-01 | 2021-06-29 | Fujitsu Limited | Device and method for determining convolutional neural network model |
| CN116151352A (en) * | 2023-04-13 | 2023-05-23 | 中浙信科技咨询有限公司 | Convolutional neural network diagnosis method based on brain information path integration mechanism |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP0646878A1 (en) * | 1993-09-30 | 1995-04-05 | Motorola, Inc. | Method for converting an existing expert system into one utilizing one or more neural networks |
| CN101976373A (en) * | 2010-11-02 | 2011-02-16 | 上海电机学院 | Neural network structural design method based on high-dimensional space classifier |
| CN102332162A (en) * | 2011-09-19 | 2012-01-25 | 西安百利信息科技有限公司 | Method for automatic recognition and stage compression of medical image regions of interest based on artificial neural network |
| CN102393908A (en) * | 2011-06-29 | 2012-03-28 | 湖南大学 | Method for identifying three bottles on mixed production line based on machine vision detection |
| CN103559542A (en) * | 2013-10-31 | 2014-02-05 | 华北水利水电大学 | Extension neural network pattern recognition method based on priori knowledge |
-
2016
- 2016-07-20 CN CN201610571220.9A patent/CN106228185B/en not_active Expired - Fee Related
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP0646878A1 (en) * | 1993-09-30 | 1995-04-05 | Motorola, Inc. | Method for converting an existing expert system into one utilizing one or more neural networks |
| CN101976373A (en) * | 2010-11-02 | 2011-02-16 | 上海电机学院 | Neural network structural design method based on high-dimensional space classifier |
| CN102393908A (en) * | 2011-06-29 | 2012-03-28 | 湖南大学 | Method for identifying three bottles on mixed production line based on machine vision detection |
| CN102332162A (en) * | 2011-09-19 | 2012-01-25 | 西安百利信息科技有限公司 | Method for automatic recognition and stage compression of medical image regions of interest based on artificial neural network |
| CN103559542A (en) * | 2013-10-31 | 2014-02-05 | 华北水利水电大学 | Extension neural network pattern recognition method based on priori knowledge |
Cited By (28)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11049017B2 (en) * | 2016-12-01 | 2021-06-29 | Fujitsu Limited | Device and method for determining convolutional neural network model |
| CN108734184B (en) * | 2017-04-17 | 2022-06-07 | 苏宁易购集团股份有限公司 | Method and device for analyzing sensitive image |
| CN108734184A (en) * | 2017-04-17 | 2018-11-02 | 苏宁云商集团股份有限公司 | A kind of method and device that sensitive image is analyzed |
| CN110770759B (en) * | 2017-05-19 | 2023-12-29 | 渊慧科技有限公司 | Neural network system |
| CN110770759A (en) * | 2017-05-19 | 2020-02-07 | 渊慧科技有限公司 | Neural network system |
| CN110892409B (en) * | 2017-06-05 | 2023-09-22 | 西门子股份公司 | Method and device for analyzing images |
| CN110892409A (en) * | 2017-06-05 | 2020-03-17 | 西门子股份公司 | Method and apparatus for analyzing images |
| US11537884B2 (en) | 2017-07-12 | 2022-12-27 | Tencent Technology (Shenzhen) Company Limited | Machine learning model training method and device, and expression image classification method and device |
| WO2019011093A1 (en) * | 2017-07-12 | 2019-01-17 | 腾讯科技（深圳）有限公司 | Machine learning model training method and apparatus, and facial expression image classification method and apparatus |
| WO2019062534A1 (en) * | 2017-09-27 | 2019-04-04 | 深圳市商汤科技有限公司 | Image retrieval method, apparatus, device and readable storage medium |
| US11256737B2 (en) | 2017-09-27 | 2022-02-22 | Shenzhen Sensetime Technology Co., Ltd. | Image retrieval methods and apparatuses, devices, and readable storage media |
| CN111295676A (en) * | 2017-10-26 | 2020-06-16 | 罗伯特·博世有限公司 | Method and apparatus for automatically generating artificial neural network |
| CN110348428A (en) * | 2017-11-01 | 2019-10-18 | 腾讯科技（深圳）有限公司 | Eye fundus image classification method, device and computer readable storage medium |
| CN110348428B (en) * | 2017-11-01 | 2023-03-24 | 腾讯科技（深圳）有限公司 | Fundus image classification method and device and computer-readable storage medium |
| CN108189043A (en) * | 2018-01-10 | 2018-06-22 | 北京飞鸿云际科技有限公司 | A kind of method for inspecting and crusing robot system applied to high ferro computer room |
| CN108345661B (en) * | 2018-01-31 | 2020-04-28 | 华南理工大学 | Wi-Fi clustering method and system based on large-scale Embedding technology |
| CN108345661A (en) * | 2018-01-31 | 2018-07-31 | 华南理工大学 | A kind of Wi-Fi clustering methods and system based on extensive Embedding technologies |
| CN108256633A (en) * | 2018-02-06 | 2018-07-06 | 苏州体素信息科技有限公司 | A kind of method of test depth Stability of Neural Networks |
| CN110163231A (en) * | 2018-02-14 | 2019-08-23 | 倍加科技股份有限公司 | Establish the method, computer system and recording medium of data identification model |
| CN109658706A (en) * | 2018-12-18 | 2019-04-19 | 深圳市捷顺科技实业股份有限公司 | A kind of vehicle data processing method and relevant device |
| CN109710819A (en) * | 2018-12-29 | 2019-05-03 | 北京航天数据股份有限公司 | A kind of model display method, apparatus, equipment and medium |
| CN109726818A (en) * | 2018-12-29 | 2019-05-07 | 北京航天数据股份有限公司 | A kind of model editing method, apparatus, equipment and medium |
| CN109886206B (en) * | 2019-02-21 | 2021-09-14 | 电子科技大学中山学院 | Three-dimensional object identification method and equipment |
| CN109886206A (en) * | 2019-02-21 | 2019-06-14 | 电子科技大学中山学院 | Three-dimensional object identification method and equipment |
| CN110456955B (en) * | 2019-08-01 | 2022-03-29 | 腾讯科技（深圳）有限公司 | Exposed clothing detection method, device, system, equipment and storage medium |
| CN110456955A (en) * | 2019-08-01 | 2019-11-15 | 腾讯科技（深圳）有限公司 | Exposure dress ornament detection method, device, system, equipment and storage medium |
| CN111368909A (en) * | 2020-03-03 | 2020-07-03 | 温州大学 | Vehicle logo identification method based on convolutional neural network depth features |
| CN116151352A (en) * | 2023-04-13 | 2023-05-23 | 中浙信科技咨询有限公司 | Convolutional neural network diagnosis method based on brain information path integration mechanism |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106228185B (en) | 2019-10-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106228185A (en) | A kind of general image classifying and identifying system based on neutral net and method | |
| CN108681752B (en) | Image scene labeling method based on deep learning | |
| CN108021947B (en) | A kind of layering extreme learning machine target identification method of view-based access control model | |
| CN111696101A (en) | Light-weight solanaceae disease identification method based on SE-Inception | |
| CN112347970B (en) | Remote sensing image ground object identification method based on graph convolution neural network | |
| CN112115967B (en) | Image increment learning method based on data protection | |
| CN111861906A (en) | Pavement crack image virtual augmentation model establishment and image virtual augmentation method | |
| CN109871892A (en) | A kind of robot vision cognitive system based on small sample metric learning | |
| CN111582395B (en) | Product quality classification system based on convolutional neural network | |
| CN114693064B (en) | Building group scheme generation performance evaluation method | |
| WO2021051987A1 (en) | Method and apparatus for training neural network model | |
| CN113705641B (en) | Hyperspectral image classification method based on rich context network | |
| CN112733602B (en) | Relation-guided pedestrian attribute identification method | |
| CN113344077A (en) | Anti-noise solanaceae disease identification method based on convolution capsule network structure | |
| CN113205103A (en) | Lightweight tattoo detection method | |
| CN116543282A (en) | Crop disease segmentation method, system, equipment and medium based on multi-scale fusion and CBAM-ResNet50 | |
| CN112967296B (en) | Point cloud dynamic region graph convolution method, classification method and segmentation method | |
| CN113420173A (en) | Minority dress image retrieval method based on quadruple deep learning | |
| CN112527547A (en) | Mechanical intelligent fault prediction method based on automatic convolution neural network | |
| CN115719453B (en) | Deep learning-based remote sensing extraction method for rice planting structure | |
| CN115330759B (en) | Method and device for calculating distance loss based on Hausdorff distance | |
| CN111340133A (en) | Image classification processing method based on deep convolutional neural network | |
| CN116258899A (en) | Corn ear classification method based on custom light convolutional neural network | |
| CN112818982B (en) | Agricultural pest image detection method based on depth feature autocorrelation activation | |
| CN110750672B (en) | Image retrieval method based on deep measurement learning and structure distribution learning loss |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | 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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20191015 Termination date: 20210720 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |