CN110163301A - A kind of classification method and device of image - Google Patents
A kind of classification method and device of image Download PDFInfo
- Publication number
- CN110163301A CN110163301A CN201910470839.4A CN201910470839A CN110163301A CN 110163301 A CN110163301 A CN 110163301A CN 201910470839 A CN201910470839 A CN 201910470839A CN 110163301 A CN110163301 A CN 110163301A
- Authority
- CN
- China
- Prior art keywords
- image
- target
- target image
- classification
- vector
- 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
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
Abstract
The embodiment of the present application provides the classification method and device of a kind of image, is related to technical field of image processing, which comprises obtains target image to be sorted;Target image is input to image recognition model trained in advance, the corresponding ProbabilityDistribution Vector of output target image, ProbabilityDistribution Vector includes multiple probability, and each probability is corresponding with a label in preset tag set;The corresponding target labels of target image are determined according to ProbabilityDistribution Vector;The corresponding relationship of label according to the pre-stored data and classification determines the corresponding classification of target labels;According to the corresponding classification of target labels, the classification of target image is determined.The cost that each live streaming platform carries out image recognition can be reduced using the application.
Description
Technical field
This application involves technical field of image processing, more particularly to the classification method and device of a kind of image.
Background technique
With the development of computer technology and network technology, network direct broadcasting has obtained widely universal.In order to guarantee user
The experience for watching live streaming, improves the health of live content, and each platform that is broadcast live needs to supervise live content.
In the related technology, the server that platform is broadcast live can identify live video by image recognition model trained in advance
In whether include illegal image, to be handled to direct broadcasting room belonging to the live video comprising illegal image (such as warning,
Close direct broadcasting room etc. down).Specific treatment process are as follows: technical staff needs to obtain training image, and marks the class of each training image
Not, wherein the category may include legal image and illegal image.Then, by the training image after label, to preset first
Beginning image recognition model is trained, the image recognition model after being trained.Wherein, initial pictures identification model can be point
Class device model, neural network model etc..During live streaming, every frame live video that live video is included by server is input to
In image recognition model after training, the corresponding classification results of the live video (i.e. legal image or illegal image) is obtained.
Since the standard that each live streaming platform defines illegal image is different, therefore, it is necessary to be respectively trained for each live streaming platform
Image recognition model expends great amount of cost to carry out image recognition.
Summary of the invention
The classification method and device for being designed to provide a kind of image of the embodiment of the present application, with reduce each live streaming platform into
The cost of row image recognition.Specific technical solution is as follows:
In a first aspect, providing a kind of classification method of image, which comprises
Obtain target image to be sorted;
The target image is input to image recognition model trained in advance, exports the corresponding probability of the target image
Distribution vector, the ProbabilityDistribution Vector include multiple probability, a mark in each probability and preset tag set
It signs corresponding;
The corresponding target labels of the target image are determined according to the ProbabilityDistribution Vector;
The corresponding relationship of label according to the pre-stored data and classification determines the corresponding classification of the target labels;
According to the corresponding classification of the target labels, the classification of the target image is determined.
It is optionally, described that the corresponding target labels of the target image are determined according to the ProbabilityDistribution Vector, comprising:
Label corresponding to maximum probability value in the ProbabilityDistribution Vector is determined as the corresponding mesh of the target image
Mark label;
It is described according to the corresponding classification of the target labels, determine the classification of the target image, comprising:
Using the classification of the target labels as the classification of the target image.
It is optionally, described that the corresponding target labels of the target image are determined according to the ProbabilityDistribution Vector, comprising:
The probability for being greater than preset probability threshold value is determined in the ProbabilityDistribution Vector, and the probability determined institute is right
The label answered is determined as the corresponding target labels of the target image;
It is described according to the corresponding classification of the target labels, determine the classification of the target image, comprising:
The mesh of highest priority is determined in the corresponding classification of the target labels according to corresponding priority of all categories
Classification is marked, and using the target category as the classification of the target image.
Optionally, the image recognition model trained in advance includes feature extraction layer, global average pond layer and defeated
Layer out;
It is described that the target image is input to image recognition model trained in advance, it is corresponding to export the target image
ProbabilityDistribution Vector, comprising:
The target image is input to the feature extraction layer, exports the corresponding feature vector of the target image;
The corresponding feature vector of the target image is input to the global average pond layer, exports the target image
Corresponding global characteristics vector;
The corresponding global characteristics vector of the target image is input to the output layer, it is corresponding to export the target image
Output vector;
According to the corresponding output vector of the target image and preset probabilistic algorithm, it is corresponding to calculate the target image
ProbabilityDistribution Vector.
Optionally, described that the target image is input to image recognition model trained in advance, export the target figure
Before corresponding ProbabilityDistribution Vector, the method also includes:
According to preset image scaling strategy, processing is zoomed in and out to the target image, the target figure after being scaled
Picture.
Optionally, the method also includes: initial image recognition model is trained, obtain it is described in advance training
Image recognition model;It is described that initial image recognition model is trained, comprising:
Preset training sample set is obtained, the training sample set includes multiple sample images and each sample image pair
The sample label answered;
For each sample image, which is input to initial image recognition model, output and the sample
The corresponding ProbabilityDistribution Vector of image;
In the corresponding ProbabilityDistribution Vector of the sample image, determination is corresponding with the sample label of the sample image
Probability;
According to the corresponding probability of the sample label of the sample image and preset loss algorithm, the sample image is calculated
Corresponding loss late;
According to the corresponding loss late of the sample image and preset weight more new algorithm, the initial image is updated
The parameter value for each parameter that identification model includes, so that the parameter value of each parameter meets the preset condition of convergence.
Second aspect, provides a kind of sorter of image, and described device includes:
Module is obtained, for obtaining target image to be sorted;
Output module exports the target for the target image to be input to image recognition model trained in advance
The corresponding ProbabilityDistribution Vector of image, the ProbabilityDistribution Vector include multiple probability, each probability and preset label
A label in set is corresponding;
First determining module, for determining the corresponding target labels of the target image according to the ProbabilityDistribution Vector;
Second determining module determines the target labels for the corresponding relationship of label according to the pre-stored data and classification
Corresponding classification;
Third determining module, for determining the classification of the target image according to the corresponding classification of the target labels.
Optionally, first determining module, is specifically used for:
Label corresponding to maximum probability value in the ProbabilityDistribution Vector is determined as the corresponding mesh of the target image
Mark label;
The third determining module, is specifically used for:
Using the classification of the target labels as the classification of the target image.
Optionally, first determining module, is specifically used for:
The probability for being greater than preset probability threshold value is determined in the ProbabilityDistribution Vector, and the probability determined institute is right
The label answered is determined as the corresponding target labels of the target image;
The third determining module, is specifically used for:
The mesh of highest priority is determined in the corresponding classification of the target labels according to corresponding priority of all categories
Classification is marked, and using the target category as the classification of the target image.
Optionally, the image recognition model trained in advance includes feature extraction layer, global average pond layer and defeated
Layer out;
The output module, is specifically used for:
The target image is input to the feature extraction layer, exports the corresponding feature vector of the target image;
The corresponding feature vector of the target image is input to the global average pond layer, exports the target image
Corresponding global characteristics vector;
The corresponding global characteristics vector of the target image is input to the output layer, it is corresponding to export the target image
Output vector;
According to the corresponding output vector of the target image and preset probabilistic algorithm, it is corresponding to calculate the target image
ProbabilityDistribution Vector.
Optionally, described device further include:
Zoom module, for zooming in and out processing to the target image, being contracted according to preset image scaling strategy
Target image after putting.
Optionally, described device further includes training module, the training module be used for initial image recognition model into
Row training, obtain the image recognition model trained in advance: the training module is specifically used for:
Preset training sample set is obtained, the training sample set includes multiple sample images and each sample image pair
The sample label answered;
For each sample image, which is input to initial image recognition model, output and the sample
The corresponding ProbabilityDistribution Vector of image;
In the corresponding ProbabilityDistribution Vector of the sample image, determination is corresponding with the sample label of the sample image
Probability;
According to the corresponding probability of the sample label of the sample image and preset loss algorithm, the sample image is calculated
Corresponding loss late;
According to the corresponding loss late of the sample image and preset weight more new algorithm, the initial image is updated
The parameter value for each parameter that identification model includes, so that the parameter value of each parameter meets the preset condition of convergence.
The third aspect provides a kind of electronic equipment, including processor, communication interface, memory and communication bus,
In, processor, communication interface, memory completes mutual communication by communication bus;
Memory, for storing computer program;
Processor when for executing the program stored on memory, realizes method and step described in first aspect.
Fourth aspect provides a kind of computer readable storage medium, is stored in the computer readable storage medium
Computer program realizes method and step described in first aspect when the computer program is executed by processor.
5th aspect, provides a kind of computer program product comprising instruction, when run on a computer, so that
Computer executes method described in above-mentioned first aspect.
The classification method and device of a kind of image provided by the embodiments of the present application.Wherein, server is available to be sorted
Target image, target image is input to in advance trained image recognition model, the corresponding probability distribution of output target image
Vector, ProbabilityDistribution Vector include multiple probability, and each probability is corresponding with a label in preset tag set.So
Afterwards, server can determine the corresponding target labels of target image, and label according to the pre-stored data according to ProbabilityDistribution Vector
With the corresponding relationship of classification, the corresponding classification of target labels is determined.Later, server can be according to the corresponding class of target labels
Not, the classification of target image is determined.In this way, different live streaming platforms only need the standard according to itself, pair of label and classification is set
It should be related to, be not necessarily to re -training image recognition model, to reduce the cost that each live streaming platform carries out image recognition.
Certainly, implement the application any product or method it is not absolutely required to and meanwhile reach all the above excellent
Point.
Detailed description of the invention
In order to illustrate the technical solutions in the embodiments of the present application or in the prior art more clearly, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of application for those of ordinary skill in the art without creative efforts, can be with
It obtains other drawings based on these drawings.
Fig. 1 is a kind of flow chart of the classification method of image provided by the embodiments of the present application;
Fig. 2 is a kind of flow chart of the recognition methods of image provided by the embodiments of the present application;
Fig. 3 is a kind of flow chart of the training method of image recognition model provided by the embodiments of the present application;
Fig. 4 is a kind of structural schematic diagram of the sorter of image provided by the embodiments of the present application;
Fig. 5 is a kind of structural schematic diagram of the sorter of image provided by the embodiments of the present application;
Fig. 6 is a kind of structural schematic diagram of the sorter of image provided by the embodiments of the present application;
Fig. 7 is the structural schematic diagram of a kind of electronic equipment provided by the embodiments of the present application.
Specific embodiment
Below in conjunction with the attached drawing in the embodiment of the present application, technical solutions in the embodiments of the present application carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of embodiments of the present application, instead of all the embodiments.It is based on
Embodiment in the application, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall in the protection scope of this application.
The embodiment of the present application provides a kind of classification method of image, can be applied to live streaming platform.Specifically, this method
It can be applied to the server in live streaming platform, or the equipment for being supervised to live content.The embodiment of the present application is to take
It is introduced for business device, other situations are similar therewith.
It is detailed to a kind of classification method progress of image provided by the embodiments of the present application below in conjunction with specific embodiment
Explanation, as shown in Figure 1, the specific steps are as follows:
Step 101, target image to be sorted is obtained.
In an implementation, when needing to be broadcast live content monitoring to a certain live video, the available live streaming of server
Live video in video, as target image to be sorted.For example, every frame in the available live video of server is straight
Image is broadcast, as target image to be sorted;Alternatively, server can also be according to the preset sampling period, periodically from this
Live video is acquired in live video, as target image to be sorted.
Step 102, target image is input to image recognition model trained in advance, the corresponding probability of output target image
Distribution vector.
Wherein, ProbabilityDistribution Vector includes multiple probability, each probability and a label phase in preset tag set
It is corresponding.
In an implementation, it can store image recognition model trained in advance in server.Server gets to be sorted
Target image after, which can be input in advance trained image recognition model.The image trained in advance
Identification model can then export the corresponding ProbabilityDistribution Vector of the target image.Wherein, ProbabilityDistribution Vector may include multiple
Probability, each probability are corresponding with a label in preset tag set.That is, the probability that ProbabilityDistribution Vector includes
Number is identical as the number for the label that tag set includes.Optionally, the type of label can be by technology people in the tag set
Member is rule of thumb configured.
For example, preset tag set is { label 1, label 2, label 3, label 4, label 5 }, the image trained in advance
The corresponding ProbabilityDistribution Vector of the target image of identification model output can be { 0.01,0,0.1,0.59,0.3 }.
Optionally, target image is input to before image recognition model trained in advance by server, can also be according to pre-
If image scaling strategy, processing is zoomed in and out to target image, the target image after being scaled.
In an implementation, image scaling strategy can be previously stored in server.The image scaling strategy can be by technology
Personnel are rule of thumb configured.Target image is input to before in advance trained image recognition model by server, can be with
According to preset image scaling strategy, processing is zoomed in and out to target image, the target image after being scaled, so that instruction in advance
Experienced image recognition model is convenient for handling target image.
For example, the size for the target image that server is got is 180px*260px, then server can be according to preset
Scaling strategy, the target image by the size scaling of target image to 224px*224px, after being scaled.
Step 103, the corresponding target labels of target image are determined according to ProbabilityDistribution Vector.
In an implementation, server obtains the corresponding ProbabilityDistribution Vector of target image (namely each label pair in tag set
The probability answered) after, the probability that can further include according to ProbabilityDistribution Vector determines the corresponding target labels of target image.
Wherein, server according to ProbabilityDistribution Vector determine the corresponding target labels of target image mode can be it is diversified.
The embodiment of the present application provides two kinds of feasible modes, specific as follows:
Label corresponding to maximum probability value in ProbabilityDistribution Vector is determined as the corresponding target of target image by mode one
Label.
In an implementation, server can determine maximum probability value in ProbabilityDistribution Vector, and then determine the maximum probability
It is worth corresponding label, which is the corresponding target labels of target image.
For example, tag set is { label 1, label 2, label 3, label 4, label 5 }, the probability of each label in tag set
For { 0.01,0,0.1,0.59,0.3 }, then the corresponding target labels of target image are label 4.
Mode two determines the probability for being greater than preset probability threshold value, and the probability that will be determined in ProbabilityDistribution Vector
Corresponding label is determined as the corresponding target labels of target image.
In an implementation, probability threshold value can be previously stored in server.The probability threshold value can by technical staff according to
Experience is configured.For each probability in ProbabilityDistribution Vector, it is preset that server may determine that whether the probability is greater than
Probability threshold value.If the probability is greater than preset probability threshold value, it is determined that the corresponding label of the probability is that target image is corresponding
Target labels.
For example, tag set is { label 1, label 2, label 3, label 4, label 5 }, the probability of each label in tag set
For { 0.01,0,0.1,0.59,0.3 }, probability threshold value 0.25, then the corresponding target labels of target image are label 4 and label
5。
Step 104, the corresponding relationship of label according to the pre-stored data and classification determines the corresponding classification of target labels.
In an implementation, the corresponding relationship of label and classification can be previously stored in server.Pair of the label and classification
Should be related to can be rule of thumb configured by technical staff.It, can be with after server obtains the corresponding target labels of target image
In label and the corresponding relationship of classification, the corresponding classification of inquiry target labels.Table one is the corresponding relationship of label and classification.Such as
Shown in table one, label 1 and the corresponding classification of label 5 are classification 1, and label 2 and the corresponding classification of label 4 are classification 2, and label 3 is right
The classification answered is classification 3.
Table one
Serial number | Label | Classification |
1 | Label 1 | Classification 1 |
2 | Label 2 | Classification 2 |
3 | Label 3 | Classification 3 |
4 | Label 4 | Classification 2 |
5 | Label 5 | Classification 1 |
Step 105, according to the corresponding classification of target labels, the classification of target image is determined.
It in an implementation, can be further corresponding according to target labels after server obtains the corresponding classification of target labels
Classification determines the classification of target image.Wherein, corresponding general according to label each in tag set for server in step 103
Rate determines that the mode of the corresponding target labels of target image is different, and server determines target according to the corresponding classification of target labels
The class of image is also different otherwise, specific as follows:
Mode one, in above-mentioned steps 103, server is by label corresponding to maximum probability value in ProbabilityDistribution Vector
The case where being determined as target image corresponding target labels (i.e. mode one), server can be using the classification of target labels as mesh
The classification of logo image.
It in an implementation, can be directly using the classification of target labels as target after server obtains the classification of target labels
The classification of image.
Mode two is greater than preset probability threshold value in above-mentioned steps 103, server determines in ProbabilityDistribution Vector
Probability, and (i.e. mode the case where label corresponding to the probability determined is determined as target image corresponding target labels
Two), server can determine highest priority in the corresponding classification of target labels according to corresponding priority of all categories
Target category, and using target category as the classification of target image.
In an implementation, after server obtains the classification of target labels, can further according to corresponding priority of all categories,
In the corresponding classification of target labels, the target category of highest priority is determined, and using target category as the class of target image
Not.
For example, target labels 1, target labels 2 and the corresponding classification of target labels 3 are classification 1, classification 2 and classification
3, the value of classification 1, classification 2 and the corresponding priority of classification 3 is 0,1 and 2, wherein the value of priority is bigger, classification
Priority it is higher, then target category be classification 3, correspondingly, the classification of target image be classification 3.
A kind of classification method of image provided by the embodiments of the present application.Wherein, the available target to be sorted of server
Target image is input to image recognition model trained in advance by image, exports the corresponding ProbabilityDistribution Vector of target image, generally
Rate distribution vector includes multiple probability, and each probability is corresponding with a label in preset tag set.Then, server
The corresponding target labels of target image, and pair of label according to the pre-stored data and classification can be determined according to ProbabilityDistribution Vector
It should be related to, determine the corresponding classification of target labels.Later, server can determine target according to the corresponding classification of target labels
The classification of image.In this way, different live streaming platforms only need the standard according to itself, the corresponding relationship of label and classification is set, is not necessarily to
Re -training image recognition model, to reduce the cost that each live streaming platform carries out image recognition.
The embodiment of the present application also provides a kind of recognition methods of image, this method can be applied to image recognition model,
The image recognition model may include feature extraction layer, global average pond layer and output layer.As shown in Fig. 2, specific processing
Process is as follows:
Step 201, target image is input to feature extraction layer, the corresponding feature vector of output target image.
In an implementation, this feature extract layer may include convolution function, batch normalized function and activation primitive, can also wrap
Other functions for characteristic vector pickup are included, the embodiment of the present application is not construed as limiting.Server gets target figure to be sorted
As after, which can be input to the feature extraction layer of image recognition model trained in advance.This feature extract layer is then
The corresponding feature vector of the target image can be exported.Wherein, this feature vector is used to indicate the characteristic information of the target image.
Step 202, the corresponding feature vector of target image is input to global average pond layer, output target image is corresponding
Global characteristics vector.
It in an implementation, can be further by the target image after server obtains the corresponding feature vector of the target image
The overall situation that corresponding feature vector is input in advance trained image recognition model is averaged pond layer.The overall situation is averaged pond layer then
Global average pondization processing can be carried out to the corresponding feature vector of the target image, and exports the corresponding overall situation of the target image
Feature vector.Wherein, which is used to indicate the global characteristics information of the target image.
Step 203, by the corresponding global characteristics vector input to output layer of target image, it is corresponding defeated to export target image
Outgoing vector.
In an implementation, which can be full articulamentum.Server obtain the corresponding global characteristics of the target image to
After amount, the corresponding global characteristics vector of the target image can be further input to the defeated of image recognition model trained in advance
Layer out.The output layer, which can then export the corresponding output vector of the target image, (can be denoted as output vector z).Wherein, this is defeated
Outgoing vector z is k dimensional vector, the number for the label that k includes by tag set.That is, output vector z is included by tag set
The corresponding numerical value of k label constitute.
Step 204, according to the corresponding output vector of target image and preset probabilistic algorithm, it is corresponding to calculate target image
ProbabilityDistribution Vector.
In an implementation, probabilistic algorithm can be previously stored in server.Wherein, which can be softmax
Algorithm, can be also other kinds of probabilistic algorithm, and the embodiment of the present application is not construed as limiting.It is corresponding that server obtains the target image
Output vector after, target figure can be calculated further according to the corresponding output vector of target image and preset probabilistic algorithm
As corresponding ProbabilityDistribution Vector.Wherein, ProbabilityDistribution Vector is made of the corresponding probability of label each in preset tag set,
Shown in the softmax algorithm such as formula (1).
Wherein, piFor the probability of i-th of label in tag set, k is the number of label in tag set, ziFor output to
Measure i-th of value in z, i.e. i-th of label corresponding numerical value in output vector z.
Optionally, server can also be trained initial image recognition model, and the image trained in advance is known
Other model.The embodiment of the present application provides a kind of training method of image recognition model, as shown in figure 3, concrete processing procedure is such as
Under:
Step 301, preset training sample set is obtained.
Wherein, training sample set includes multiple sample images and the corresponding sample label of each sample image.
In an implementation, technical staff can obtain preset number sample image in advance.For each sample image, technology
Personnel can be arranged the corresponding sample label of the sample image, obtain training sample set, and deposit according to the content of the sample image
Storage is in the server.When server needs to be trained image recognition model, the available pre-stored instruction of server
Practice sample set.
Step 302, for each sample image, which is input to initial image recognition model, output with
The corresponding ProbabilityDistribution Vector of the sample image.
In an implementation, after server gets training sample set, for each sample image, server can be by the sample
Image is input in initial image recognition model, and it is corresponding which can then export the sample image
ProbabilityDistribution Vector.The ProbabilityDistribution Vector is made of the corresponding probability of label each in preset tag set.Wherein, initially
Image recognition model exports the processed of the corresponding ProbabilityDistribution Vector of the sample image according to the sample image that server inputs
Journey is similar with the treatment process of step 201 to step 204.
Step 303, in the corresponding ProbabilityDistribution Vector of the sample image, the determining sample label pair with the sample image
The probability answered.
In an implementation, server obtains the corresponding ProbabilityDistribution Vector of the sample image (namely each label in tag set
Corresponding probability) after, the sample of the sample image can be determined further in the corresponding ProbabilityDistribution Vector of the sample image
The corresponding probability of label.
Step 304, according to the corresponding probability of the sample label of the sample image and preset loss algorithm, the sample is calculated
The corresponding loss late of image.
In an implementation, loss algorithm can be previously stored in server.Server obtains the sample mark of the sample image
After signing corresponding probability, can further according to the corresponding probability of the sample label of the sample image and preset loss algorithm,
Calculate the corresponding loss late of the sample image.Wherein, shown in the loss algorithm such as formula (2).
L=-log (py) formula (2)
Wherein, L is the corresponding loss late of sample image, pyFor the corresponding probability of y-th of sample label of the sample image.
Step 305, according to the corresponding loss late of the sample image and preset weight more new algorithm, initial figure is updated
As the parameter value for each parameter that identification model includes, so that the parameter value of each parameter meets the preset condition of convergence.
In an implementation, weight more new algorithm and the condition of convergence can be previously stored in server.The weight more new algorithm
It can be stochastic gradient descent algorithm, or other kinds of weight more new algorithm, the embodiment of the present application are not construed as limiting.Clothes
It, can be further according to the corresponding loss late of the sample image and preset power after business device obtains the corresponding loss late of sample image
Weight more new algorithm, updates the parameter value for each parameter that the initial image recognition model includes, until the parameter value of each parameter is full
The preset condition of convergence of foot, thus the image recognition model after being trained.Wherein, the stochastic gradient descent algorithm such as formula
(3) shown in.
Wherein, W is the parameter value (W can be vector or matrix) for each parameter that initial image recognition model includes, and a is
Preset learning rate.
Based on the same technical idea, the embodiment of the present application also provides a kind of sorters of image, as shown in figure 4,
The device includes:
Module 410 is obtained, for obtaining target image to be sorted;
Output module 420 exports the mesh for the target image to be input to image recognition model trained in advance
The corresponding ProbabilityDistribution Vector of logo image, the ProbabilityDistribution Vector include multiple probability, each probability and preset mark
A label in label set is corresponding;
First determining module 430, for determining the corresponding target mark of the target image according to the ProbabilityDistribution Vector
Label;
Second determining module 440 determines the target mark for the corresponding relationship of label according to the pre-stored data and classification
Sign corresponding classification;
Third determining module 450, for determining the class of the target image according to the corresponding classification of the target labels
Not.
Optionally, first determining module 430, is specifically used for:
Label corresponding to maximum probability value in the ProbabilityDistribution Vector is determined as the corresponding mesh of the target image
Mark label;
The third determining module 450, is specifically used for:
Using the classification of the target labels as the classification of the target image.
Optionally, first determining module 430, is specifically used for:
The probability for being greater than preset probability threshold value is determined in the ProbabilityDistribution Vector, and the probability determined institute is right
The label answered is determined as the corresponding target labels of the target image;
The third determining module 450, is specifically used for:
The mesh of highest priority is determined in the corresponding classification of the target labels according to corresponding priority of all categories
Classification is marked, and using the target category as the classification of the target image.
Optionally, the image recognition model trained in advance includes feature extraction layer, global average pond layer and defeated
Layer out;
The output module 420, is specifically used for:
The target image is input to the feature extraction layer, exports the corresponding feature vector of the target image;
The corresponding feature vector of the target image is input to the global average pond layer, exports the target image
Corresponding global characteristics vector;
The corresponding global characteristics vector of the target image is input to the output layer, it is corresponding to export the target image
Output vector;
According to the corresponding output vector of the target image and preset probabilistic algorithm, it is corresponding to calculate the target image
ProbabilityDistribution Vector.
Optionally, as shown in figure 5, described device further include:
Zoom module 460, for zooming in and out processing to the target image, obtaining according to preset image scaling strategy
Target image after to scaling.
Optionally, as shown in fig. 6, described device further includes training module 470, the training module 470 is used for initial
Image recognition model be trained, obtain the image recognition model trained in advance: the training module 470, it is specific to use
In:
Preset training sample set is obtained, the training sample set includes multiple sample images and each sample image pair
The sample label answered;
For each sample image, which is input to initial image recognition model, output and the sample
The corresponding ProbabilityDistribution Vector of image;
In the corresponding ProbabilityDistribution Vector of the sample image, determination is corresponding with the sample label of the sample image
Probability;
According to the corresponding probability of the sample label of the sample image and preset loss algorithm, the sample image is calculated
Corresponding loss late;
According to the corresponding loss late of the sample image and preset weight more new algorithm, the initial image is updated
The parameter value for each parameter that identification model includes, so that the parameter value of each parameter meets the preset condition of convergence.
A kind of sorter of image provided by the embodiments of the present application.Wherein, the available target to be sorted of server
Target image is input to image recognition model trained in advance by image, exports the corresponding ProbabilityDistribution Vector of target image, generally
Rate distribution vector includes multiple probability, and each probability is corresponding with a label in preset tag set.Then, server
The corresponding target labels of target image, and pair of label according to the pre-stored data and classification can be determined according to ProbabilityDistribution Vector
It should be related to, determine the corresponding classification of target labels.Later, server can determine target according to the corresponding classification of target labels
The classification of image.In this way, different live streaming platforms only need the standard according to itself, the corresponding relationship of label and classification is set, is not necessarily to
Re -training image recognition model, to reduce the cost that each live streaming platform carries out image recognition.
The embodiment of the present application also provides a kind of electronic equipment, as shown in fig. 7, comprises processor 701, communication interface 702,
Memory 703 and communication bus 704, wherein processor 701, communication interface 702, memory 703 are complete by communication bus 704
At mutual communication,
Memory 703, for storing computer program;
Processor 701 when for executing the program stored on memory 703, realizes following steps:
Obtain target image to be sorted;
The target image is input to image recognition model trained in advance, exports the corresponding probability of the target image
Distribution vector, the ProbabilityDistribution Vector include multiple probability, a mark in each probability and preset tag set
It signs corresponding;
The corresponding target labels of the target image are determined according to the ProbabilityDistribution Vector;
The corresponding relationship of label according to the pre-stored data and classification determines the corresponding classification of the target labels;
According to the corresponding classification of the target labels, the classification of the target image is determined.
It is optionally, described that the corresponding target labels of the target image are determined according to the ProbabilityDistribution Vector, comprising:
Label corresponding to maximum probability value in the ProbabilityDistribution Vector is determined as the corresponding mesh of the target image
Mark label;
It is described according to the corresponding classification of the target labels, determine the classification of the target image, comprising:
Using the classification of the target labels as the classification of the target image.
It is optionally, described that the corresponding target labels of the target image are determined according to the ProbabilityDistribution Vector, comprising:
The probability for being greater than preset probability threshold value is determined in the ProbabilityDistribution Vector, and the probability determined institute is right
The label answered is determined as the corresponding target labels of the target image;
It is described according to the corresponding classification of the target labels, determine the classification of the target image, comprising:
The mesh of highest priority is determined in the corresponding classification of the target labels according to corresponding priority of all categories
Classification is marked, and using the target category as the classification of the target image.
Optionally, the image recognition model trained in advance includes feature extraction layer, global average pond layer and defeated
Layer out;
It is described that the target image is input to image recognition model trained in advance, it is corresponding to export the target image
ProbabilityDistribution Vector, comprising:
The target image is input to the feature extraction layer, exports the corresponding feature vector of the target image;
The corresponding feature vector of the target image is input to the global average pond layer, exports the target image
Corresponding global characteristics vector;
The corresponding global characteristics vector of the target image is input to the output layer, it is corresponding to export the target image
Output vector;
According to the corresponding output vector of the target image and preset probabilistic algorithm, it is corresponding to calculate the target image
ProbabilityDistribution Vector.
Optionally, described that the target image is input to image recognition model trained in advance, export the target figure
Before corresponding ProbabilityDistribution Vector, the method also includes:
According to preset image scaling strategy, processing is zoomed in and out to the target image, the target figure after being scaled
Picture.
Optionally, the method also includes: initial image recognition model is trained, obtain it is described in advance training
Image recognition model;It is described that initial image recognition model is trained, comprising:
Preset training sample set is obtained, the training sample set includes multiple sample images and each sample image pair
The sample label answered;
For each sample image, which is input to initial image recognition model, output and the sample
The corresponding ProbabilityDistribution Vector of image;
In the corresponding ProbabilityDistribution Vector of the sample image, determination is corresponding with the sample label of the sample image
Probability;
According to the corresponding probability of the sample label of the sample image and preset loss algorithm, the sample image is calculated
Corresponding loss late;
According to the corresponding loss late of the sample image and preset weight more new algorithm, the initial image is updated
The parameter value for each parameter that identification model includes, so that the parameter value of each parameter meets the preset condition of convergence.
The communication bus that above-mentioned electronic equipment is mentioned can be Peripheral Component Interconnect standard (English: Peripheral
Component Interconnect, referred to as: PCI) bus or expanding the industrial standard structure (English: Extended Industry
Standard Architecture, referred to as: EISA) bus etc..The communication bus can be divided into address bus, data/address bus, control
Bus processed etc..Only to be indicated with a thick line in figure convenient for indicating, it is not intended that an only bus or a type of total
Line.
Communication interface is for the communication between above-mentioned electronic equipment and other equipment.
Memory may include random access memory (English: Random Access Memory, abbreviation: RAM), can also
To include nonvolatile memory (English: Non-Volatile Memory, abbreviation: NVM), for example, at least a disk storage
Device.Optionally, memory can also be that at least one is located remotely from the storage device of aforementioned processor.
Above-mentioned processor can be general processor, including central processing unit (English: Central Processing
Unit, referred to as: CPU), network processing unit (English: Network Processor, referred to as: NP) etc.;It can also be digital signal
Processor (English: Digital Signal Processing, abbreviation: DSP), specific integrated circuit (English: Application
Specific Integrated Circuit, referred to as: ASIC), field programmable gate array (English: Field-
Programmable Gate Array, referred to as: FPGA) either other programmable logic device, discrete gate or transistor logic
Device, discrete hardware components.
Based on the same technical idea, the embodiment of the present application also provides a kind of computer readable storage medium, the meters
Computer program is stored in calculation machine readable storage medium storing program for executing, the computer program realizes above-mentioned image when being executed by processor
Classification method and step.
Based on the same technical idea, the embodiment of the present application also provides a kind of computer program product comprising instruction,
When run on a computer, so that the method that computer executes the classification of above-mentioned image.
In the above-described embodiments, can come wholly or partly by software, hardware, firmware or any combination thereof real
It is existing.When implemented in software, it can entirely or partly realize in the form of a computer program product.The computer program
Product includes one or more computer instructions.When loading on computers and executing the computer program instructions, all or
It partly generates according to process or function described in the embodiment of the present application.The computer can be general purpose computer, dedicated meter
Calculation machine, computer network or other programmable devices.The computer instruction can store in computer readable storage medium
In, or from a computer readable storage medium to the transmission of another computer readable storage medium, for example, the computer
Instruction can pass through wired (such as coaxial cable, optical fiber, number from a web-site, computer, server or data center
User's line (DSL)) or wireless (such as infrared, wireless, microwave etc.) mode to another web-site, computer, server or
Data center is transmitted.The computer readable storage medium can be any usable medium that computer can access or
It is comprising data storage devices such as one or more usable mediums integrated server, data centers.The usable medium can be with
It is magnetic medium, (for example, floppy disk, hard disk, tape), optical medium (for example, DVD) or semiconductor medium (such as solid state hard disk
Solid State Disk (SSD)) etc..
It should be noted that, in this document, relational terms such as first and second and the like are used merely to a reality
Body or operation are distinguished with another entity or operation, are deposited without necessarily requiring or implying between these entities or operation
In any actual relationship or order or sequence.Moreover, the terms "include", "comprise" or its any other variant are intended to
Non-exclusive inclusion, so that the process, method, article or equipment including a series of elements is not only wanted including those
Element, but also including other elements that are not explicitly listed, or further include for this process, method, article or equipment
Intrinsic element.In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that
There is also other identical elements in process, method, article or equipment including the element.
Each embodiment in this specification is all made of relevant mode and describes, same and similar portion between each embodiment
Dividing may refer to each other, and each embodiment focuses on the differences from other embodiments.Especially for device
Speech, since it is substantially similar to the method embodiment, so being described relatively simple, referring to the part of embodiment of the method in place of correlation
Explanation.
The foregoing is merely the preferred embodiments of the application, are not intended to limit the protection scope of the application.It is all
Any modification, equivalent replacement, improvement and so within spirit herein and principle are all contained in the protection scope of the application
It is interior.
Claims (14)
1. a kind of classification method of image, which is characterized in that the described method includes:
Obtain target image to be sorted;
The target image is input to image recognition model trained in advance, exports the corresponding probability distribution of the target image
Vector, the ProbabilityDistribution Vector include multiple probability, each probability and a label phase in preset tag set
It is corresponding;
The corresponding target labels of the target image are determined according to the ProbabilityDistribution Vector;
The corresponding relationship of label according to the pre-stored data and classification determines the corresponding classification of the target labels;
According to the corresponding classification of the target labels, the classification of the target image is determined.
2. the method according to claim 1, wherein described determine the target according to the ProbabilityDistribution Vector
The corresponding target labels of image, comprising:
Label corresponding to maximum probability value in the ProbabilityDistribution Vector is determined as the corresponding target mark of the target image
Label;
It is described according to the corresponding classification of the target labels, determine the classification of the target image, comprising:
Using the classification of the target labels as the classification of the target image.
3. the method according to claim 1, wherein described determine the target according to the ProbabilityDistribution Vector
The corresponding target labels of image, comprising:
The probability for being greater than preset probability threshold value is determined in the ProbabilityDistribution Vector, and will be corresponding to the probability that determined
Label is determined as the corresponding target labels of the target image;
It is described according to the corresponding classification of the target labels, determine the classification of the target image, comprising:
The target class of highest priority is determined in the corresponding classification of the target labels according to corresponding priority of all categories
Not, and using the target category as the classification of the target image.
4. the method according to claim 1, wherein the image recognition model trained in advance includes that feature mentions
Take layer, global average pond layer and output layer;
It is described that the target image is input to image recognition model trained in advance, export the corresponding probability of the target image
Distribution vector, comprising:
The target image is input to the feature extraction layer, exports the corresponding feature vector of the target image;
The corresponding feature vector of the target image is input to the global average pond layer, it is corresponding to export the target image
Global characteristics vector;
The corresponding global characteristics vector of the target image is input to the output layer, it is corresponding defeated to export the target image
Outgoing vector;
According to the corresponding output vector of the target image and preset probabilistic algorithm, the corresponding probability of the target image is calculated
Distribution vector.
5. the method according to claim 1, wherein described be input to the target image in figure trained in advance
As identification model, before exporting the corresponding ProbabilityDistribution Vector of the target image, the method also includes:
According to preset image scaling strategy, processing is zoomed in and out to the target image, the target image after being scaled.
6. the method according to claim 1, wherein the method also includes: to initial image recognition model
It is trained, obtains the image recognition model trained in advance;It is described that initial image recognition model is trained, packet
It includes:
Preset training sample set is obtained, the training sample set includes that multiple sample images and each sample image are corresponding
Sample label;
For each sample image, which is input to initial image recognition model, output and the sample image
Corresponding ProbabilityDistribution Vector;
In the corresponding ProbabilityDistribution Vector of the sample image, determine corresponding with the sample label of the sample image general
Rate;
According to the corresponding probability of the sample label of the sample image and preset loss algorithm, it is corresponding to calculate the sample image
Loss late;
According to the corresponding loss late of the sample image and preset weight more new algorithm, the initial image recognition is updated
The parameter value for each parameter that model includes, so that the parameter value of each parameter meets the preset condition of convergence.
7. a kind of sorter of image, which is characterized in that described device includes:
Module is obtained, for obtaining target image to be sorted;
Output module exports the target image for the target image to be input to image recognition model trained in advance
Corresponding ProbabilityDistribution Vector, the ProbabilityDistribution Vector include multiple probability, each probability and preset tag set
In a label it is corresponding;
First determining module, for determining the corresponding target labels of the target image according to the ProbabilityDistribution Vector;
Second determining module determines that the target labels are corresponding for the corresponding relationship of label according to the pre-stored data and classification
Classification;
Third determining module, for determining the classification of the target image according to the corresponding classification of the target labels.
8. device according to claim 7, which is characterized in that first determining module is specifically used for:
Label corresponding to maximum probability value in the ProbabilityDistribution Vector is determined as the corresponding target mark of the target image
Label;
The third determining module, is specifically used for:
Using the classification of the target labels as the classification of the target image.
9. device according to claim 7, which is characterized in that first determining module is specifically used for:
The probability for being greater than preset probability threshold value is determined in the ProbabilityDistribution Vector, and will be corresponding to the probability that determined
Label is determined as the corresponding target labels of the target image;
The third determining module, is specifically used for:
The target class of highest priority is determined in the corresponding classification of the target labels according to corresponding priority of all categories
Not, and using the target category as the classification of the target image.
10. device according to claim 7, which is characterized in that the image recognition model trained in advance includes feature
Extract layer, global average pond layer and output layer;
The output module, is specifically used for:
The target image is input to the feature extraction layer, exports the corresponding feature vector of the target image;
The corresponding feature vector of the target image is input to the global average pond layer, it is corresponding to export the target image
Global characteristics vector;
The corresponding global characteristics vector of the target image is input to the output layer, it is corresponding defeated to export the target image
Outgoing vector;
According to the corresponding output vector of the target image and preset probabilistic algorithm, the corresponding probability of the target image is calculated
Distribution vector.
11. device according to claim 7, which is characterized in that described device further include:
Zoom module, for processing being zoomed in and out to the target image, after obtaining scaling according to preset image scaling strategy
Target image.
12. device according to claim 7, which is characterized in that described device further includes training module, the training module
For being trained to initial image recognition model, the image recognition model trained in advance: the training module is obtained,
It is specifically used for:
Preset training sample set is obtained, the training sample set includes that multiple sample images and each sample image are corresponding
Sample label;
For each sample image, which is input to initial image recognition model, output and the sample image
Corresponding ProbabilityDistribution Vector;
In the corresponding ProbabilityDistribution Vector of the sample image, determine corresponding with the sample label of the sample image general
Rate;
According to the corresponding probability of the sample label of the sample image and preset loss algorithm, it is corresponding to calculate the sample image
Loss late;
According to the corresponding loss late of the sample image and preset weight more new algorithm, the initial image recognition is updated
The parameter value for each parameter that model includes, so that the parameter value of each parameter meets the preset condition of convergence.
13. a kind of electronic equipment, which is characterized in that including processor, communication interface, memory and communication bus, wherein processing
Device, communication interface, memory complete mutual communication by communication bus;
Memory, for storing computer program;
Processor when for executing the program stored on memory, realizes any method and step of claim 1-6.
14. a kind of computer readable storage medium, which is characterized in that be stored with computer in the computer readable storage medium
Program realizes claim 1-6 any method and step when the computer program is executed by processor.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910470839.4A CN110163301A (en) | 2019-05-31 | 2019-05-31 | A kind of classification method and device of image |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910470839.4A CN110163301A (en) | 2019-05-31 | 2019-05-31 | A kind of classification method and device of image |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110163301A true CN110163301A (en) | 2019-08-23 |
Family
ID=67630917
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910470839.4A Pending CN110163301A (en) | 2019-05-31 | 2019-05-31 | A kind of classification method and device of image |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110163301A (en) |
Cited By (18)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110633746A (en) * | 2019-09-12 | 2019-12-31 | 苏州大学 | Image classification method, device, equipment and computer readable storage medium |
CN110674854A (en) * | 2019-09-09 | 2020-01-10 | 东软集团股份有限公司 | Image classification model training method, image classification method, device and equipment |
CN110751675A (en) * | 2019-09-03 | 2020-02-04 | 平安科技(深圳)有限公司 | Urban pet activity track monitoring method based on image recognition and related equipment |
CN110795703A (en) * | 2019-09-20 | 2020-02-14 | 华为技术有限公司 | Data anti-theft method and related product |
CN110830807A (en) * | 2019-11-04 | 2020-02-21 | 腾讯科技(深圳)有限公司 | Image compression method, device and storage medium |
CN111127179A (en) * | 2019-12-12 | 2020-05-08 | 恩亿科(北京)数据科技有限公司 | Information pushing method and device, computer equipment and storage medium |
CN111191706A (en) * | 2019-12-25 | 2020-05-22 | 深圳市赛维网络科技有限公司 | Picture identification method, device, equipment and storage medium |
CN111242230A (en) * | 2020-01-17 | 2020-06-05 | 腾讯科技(深圳)有限公司 | Image processing method and image classification model training method based on artificial intelligence |
CN111274735A (en) * | 2020-01-19 | 2020-06-12 | 南方科技大学 | Thermal diffusion coefficient testing method and device, computer equipment and storage medium |
CN111353542A (en) * | 2020-03-03 | 2020-06-30 | 腾讯科技(深圳)有限公司 | Training method and device of image classification model, computer equipment and storage medium |
CN111368789A (en) * | 2020-03-18 | 2020-07-03 | 腾讯科技(深圳)有限公司 | Image recognition method and device, computer equipment and storage medium |
CN111444813A (en) * | 2020-03-23 | 2020-07-24 | 北京字节跳动网络技术有限公司 | Method, device, equipment and storage medium for identifying attribute classification of target object |
CN111753618A (en) * | 2019-12-20 | 2020-10-09 | 北京沃东天骏信息技术有限公司 | Image recognition method and device, computer equipment and computer readable storage medium |
CN112686159A (en) * | 2020-12-30 | 2021-04-20 | 北京墨迹风云科技股份有限公司 | Allergic plant distribution statistical method and device |
CN112926608A (en) * | 2019-12-05 | 2021-06-08 | 北京金山云网络技术有限公司 | Image classification method and device, electronic equipment and storage medium |
CN113127667A (en) * | 2019-12-30 | 2021-07-16 | 阿里巴巴集团控股有限公司 | Image processing method and device, and image classification method and device |
CN113240032A (en) * | 2021-05-25 | 2021-08-10 | 北京有竹居网络技术有限公司 | Image classification method, device, equipment and storage medium |
CN115630289A (en) * | 2022-12-21 | 2023-01-20 | 白杨时代(北京)科技有限公司 | Target identification method and device based on evidence theory |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103092911A (en) * | 2012-11-20 | 2013-05-08 | 北京航空航天大学 | K-neighbor-based collaborative filtering recommendation system for combining social label similarity |
CN105159881A (en) * | 2015-08-28 | 2015-12-16 | 北京奇艺世纪科技有限公司 | Method and device for polymerizing data module in page |
CN105824930A (en) * | 2016-03-17 | 2016-08-03 | 深圳市金立通信设备有限公司 | Voice message processing method and terminal |
CN107145904A (en) * | 2017-04-28 | 2017-09-08 | 北京小米移动软件有限公司 | Determination method, device and the storage medium of image category |
CN107609503A (en) * | 2017-09-05 | 2018-01-19 | 刘宇红 | Intelligent cancerous tumor cell identifying system and method, cloud platform, server, computer |
US20180089543A1 (en) * | 2016-09-23 | 2018-03-29 | International Business Machines Corporation | Image classification utilizing semantic relationships in a classification hierarchy |
CN108509660A (en) * | 2018-05-29 | 2018-09-07 | 维沃移动通信有限公司 | A kind of broadcasting object recommendation method and terminal device |
CN109245996A (en) * | 2018-09-18 | 2019-01-18 | 平安科技(深圳)有限公司 | Mail push method, device, computer equipment and storage medium |
-
2019
- 2019-05-31 CN CN201910470839.4A patent/CN110163301A/en active Pending
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103092911A (en) * | 2012-11-20 | 2013-05-08 | 北京航空航天大学 | K-neighbor-based collaborative filtering recommendation system for combining social label similarity |
CN105159881A (en) * | 2015-08-28 | 2015-12-16 | 北京奇艺世纪科技有限公司 | Method and device for polymerizing data module in page |
CN105824930A (en) * | 2016-03-17 | 2016-08-03 | 深圳市金立通信设备有限公司 | Voice message processing method and terminal |
US20180089543A1 (en) * | 2016-09-23 | 2018-03-29 | International Business Machines Corporation | Image classification utilizing semantic relationships in a classification hierarchy |
CN107145904A (en) * | 2017-04-28 | 2017-09-08 | 北京小米移动软件有限公司 | Determination method, device and the storage medium of image category |
CN107609503A (en) * | 2017-09-05 | 2018-01-19 | 刘宇红 | Intelligent cancerous tumor cell identifying system and method, cloud platform, server, computer |
CN108509660A (en) * | 2018-05-29 | 2018-09-07 | 维沃移动通信有限公司 | A kind of broadcasting object recommendation method and terminal device |
CN109245996A (en) * | 2018-09-18 | 2019-01-18 | 平安科技(深圳)有限公司 | Mail push method, device, computer equipment and storage medium |
Cited By (27)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110751675A (en) * | 2019-09-03 | 2020-02-04 | 平安科技(深圳)有限公司 | Urban pet activity track monitoring method based on image recognition and related equipment |
CN110751675B (en) * | 2019-09-03 | 2023-08-11 | 平安科技(深圳)有限公司 | Urban pet activity track monitoring method based on image recognition and related equipment |
CN110674854A (en) * | 2019-09-09 | 2020-01-10 | 东软集团股份有限公司 | Image classification model training method, image classification method, device and equipment |
CN110674854B (en) * | 2019-09-09 | 2022-05-17 | 东软集团股份有限公司 | Image classification model training method, image classification method, device and equipment |
CN110633746A (en) * | 2019-09-12 | 2019-12-31 | 苏州大学 | Image classification method, device, equipment and computer readable storage medium |
CN110795703A (en) * | 2019-09-20 | 2020-02-14 | 华为技术有限公司 | Data anti-theft method and related product |
CN110795703B (en) * | 2019-09-20 | 2024-04-16 | 华为技术有限公司 | Data theft prevention method and related product |
CN110830807A (en) * | 2019-11-04 | 2020-02-21 | 腾讯科技(深圳)有限公司 | Image compression method, device and storage medium |
CN110830807B (en) * | 2019-11-04 | 2022-08-23 | 腾讯科技(深圳)有限公司 | Image compression method, device and storage medium |
CN112926608A (en) * | 2019-12-05 | 2021-06-08 | 北京金山云网络技术有限公司 | Image classification method and device, electronic equipment and storage medium |
CN111127179A (en) * | 2019-12-12 | 2020-05-08 | 恩亿科(北京)数据科技有限公司 | Information pushing method and device, computer equipment and storage medium |
CN111127179B (en) * | 2019-12-12 | 2023-08-29 | 恩亿科(北京)数据科技有限公司 | Information pushing method, device, computer equipment and storage medium |
CN111753618A (en) * | 2019-12-20 | 2020-10-09 | 北京沃东天骏信息技术有限公司 | Image recognition method and device, computer equipment and computer readable storage medium |
CN111191706A (en) * | 2019-12-25 | 2020-05-22 | 深圳市赛维网络科技有限公司 | Picture identification method, device, equipment and storage medium |
CN113127667A (en) * | 2019-12-30 | 2021-07-16 | 阿里巴巴集团控股有限公司 | Image processing method and device, and image classification method and device |
CN111242230A (en) * | 2020-01-17 | 2020-06-05 | 腾讯科技(深圳)有限公司 | Image processing method and image classification model training method based on artificial intelligence |
CN111274735A (en) * | 2020-01-19 | 2020-06-12 | 南方科技大学 | Thermal diffusion coefficient testing method and device, computer equipment and storage medium |
CN111353542B (en) * | 2020-03-03 | 2023-09-19 | 腾讯科技(深圳)有限公司 | Training method and device for image classification model, computer equipment and storage medium |
CN111353542A (en) * | 2020-03-03 | 2020-06-30 | 腾讯科技(深圳)有限公司 | Training method and device of image classification model, computer equipment and storage medium |
CN111368789A (en) * | 2020-03-18 | 2020-07-03 | 腾讯科技(深圳)有限公司 | Image recognition method and device, computer equipment and storage medium |
CN111368789B (en) * | 2020-03-18 | 2023-05-26 | 腾讯科技(深圳)有限公司 | Image recognition method, device, computer equipment and storage medium |
CN111444813A (en) * | 2020-03-23 | 2020-07-24 | 北京字节跳动网络技术有限公司 | Method, device, equipment and storage medium for identifying attribute classification of target object |
CN112686159A (en) * | 2020-12-30 | 2021-04-20 | 北京墨迹风云科技股份有限公司 | Allergic plant distribution statistical method and device |
CN113240032A (en) * | 2021-05-25 | 2021-08-10 | 北京有竹居网络技术有限公司 | Image classification method, device, equipment and storage medium |
CN113240032B (en) * | 2021-05-25 | 2024-01-30 | 北京有竹居网络技术有限公司 | Image classification method, device, equipment and storage medium |
CN115630289B (en) * | 2022-12-21 | 2023-09-26 | 白杨时代(北京)科技有限公司 | Target identification method and device based on evidence theory |
CN115630289A (en) * | 2022-12-21 | 2023-01-20 | 白杨时代(北京)科技有限公司 | Target identification method and device based on evidence theory |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110163301A (en) | A kind of classification method and device of image | |
WO2020239015A1 (en) | Image recognition method and apparatus, image classification method and apparatus, electronic device, and storage medium | |
CN109165645B (en) | Image processing method and device and related equipment | |
CN107678845B (en) | Application program control method and device, storage medium and electronic equipment | |
CN107341716A (en) | A kind of method, apparatus and electronic equipment of the identification of malice order | |
CN110738235B (en) | Pulmonary tuberculosis judging method, device, computer equipment and storage medium | |
CN108961183A (en) | Image processing method, terminal device and computer readable storage medium | |
WO2020253508A1 (en) | Abnormal cell detection method and apparatus, and computer readable storage medium | |
CN108197250A (en) | Picture retrieval method, electronic equipment and storage medium | |
US10769784B2 (en) | Image analyzing method and electrical device | |
CN111160411B (en) | Classification model training method, image processing method, device, medium and equipment | |
CN109101931A (en) | A kind of scene recognition method, scene Recognition device and terminal device | |
CN108319888A (en) | The recognition methods of video type and device, terminal | |
CN111008561A (en) | Livestock quantity determination method, terminal and computer storage medium | |
CN109165309A (en) | Negative training sample acquisition method, device and model training method, device | |
CN109508405B (en) | Method and device for determining recommended video, electronic equipment and storage medium | |
CN111159481B (en) | Edge prediction method and device for graph data and terminal equipment | |
CN113869098A (en) | Plant disease identification method and device, electronic equipment and storage medium | |
CN108647986A (en) | A kind of target user determines method, apparatus and electronic equipment | |
CN110738077B (en) | Foreign matter detection method and device | |
CN109361929A (en) | A kind of method and relevant device of determining direct broadcasting room label | |
CN112016679B (en) | Test sample category determining method and device for twin network and terminal equipment | |
CN111414966B (en) | Classification method, classification device, electronic equipment and computer storage medium | |
CN107330710A (en) | A kind of insertion method, device and the electronic equipment of primary advertisement | |
CN110969602B (en) | Image definition detection method and device |
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 |