CN110378420A - A kind of image detecting method, device and computer readable storage medium - Google Patents
A kind of image detecting method, device and computer readable storage medium Download PDFInfo
- Publication number
- CN110378420A CN110378420A CN201910657541.4A CN201910657541A CN110378420A CN 110378420 A CN110378420 A CN 110378420A CN 201910657541 A CN201910657541 A CN 201910657541A CN 110378420 A CN110378420 A CN 110378420A
- Authority
- CN
- China
- Prior art keywords
- image
- classification
- target
- test object
- detected
- 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
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The embodiment of the present application discloses a kind of image detecting method, device and computer readable storage medium, mark training sample concentrates position and the classification of at least two test objects in each sample image in advance, obtains the labeling position and mark classification of test object；Target detection network is trained using the training sample set and the labeling position of the test object, obtains target detection model；Target classification network is trained using the training sample set and the labeling position of the test object and mark classification, obtains target classification network model；To obtain the target detection model of test object position for identification, and the target category model of test object classification for identification；Using the position and classification of target detection model and target category model difference recognition detection object, the accuracy of test object identification can be improved.
Description
Technical field
This application involves image technique more particularly to a kind of image detecting methods, device and computer readable storage medium.
Background technique
Existing image detecting technique typically directly uses a target detection net for extensive more object detections in real time
Network carrys out frame and selects multiple target objects, and the mark of classification setting rectangle frame is carried out to the target object for including in each rectangle frame
Label.Because in the case where accomplishing real-time, the network model of lightweight can be used, such as mobilenet-ssd, yolo series etc.,
For the detection for realizing more objects, the training set of network model at least needs hundreds and thousands of tired life common items.But it uses existing
The label for having technology to obtain rectangle frame is easy to appear mistake, for example, " desk " is misidentified as " automobile ", " display " is accidentally known
Not Wei " window " etc., the accuracy of object identification is lower.
Summary of the invention
In order to solve the above technical problems, the embodiment of the present application is intended to provide a kind of image detecting method, device and computer
Storage medium.
The technical solution of the application is achieved in that
In a first aspect, a kind of image detecting method is provided, this method comprises:
Obtain the training sample set comprising at least two sample images；It wherein, include at least two in each sample image
A test object；
The position of at least two test objects and classification in mark sample image in advance, obtain the labeling position of test object
With mark classification；
Target detection network is trained using training sample set and the labeling position of test object, obtains target
Detection model；Wherein, target detection model is used to detect the position of at least two test objects in an image；
Target classification network is instructed using training sample set and the labeling position of test object and mark classification
Practice, obtains target classification network model；Wherein, object-class model is for detecting at least two test objects in an image
Classification；
Image to be detected is identified using the target detection model and the object-class model, obtain it is described to
The location information and classification information of at least two test objects in detection image.
Second aspect provides a kind of image detection device, which includes:
Acquiring unit, for obtaining the training sample set comprising at least two sample images；Wherein, each sample image
In contain at least two test object；
Mark unit is detected for marking the position of at least two test objects and classification in sample image in advance
The labeling position and mark classification of object；
Processing unit carries out target detection network for the labeling position using training sample set and test object
Training, obtains target detection model；Wherein, target detection model is used to detect at least two test objects in an image
Position；
Processing unit for the labeling position using training sample set and test object and marks classification to target point
Class network is trained, and obtains target classification network model；Wherein, object-class model is for detecting at least two in an image
The classification of a test object；
Processing unit, for being known using the target detection model and the object-class model to image to be detected
Not, the location information and classification information of at least two test objects in described image to be detected are obtained.
The third aspect provides a kind of image detection device, comprising: processor and be configured to storage can be on a processor
The memory of the computer program of operation, wherein when the processor is configured to run the computer program, execute aforementioned side
The step of method.
Fourth aspect provides a kind of computer readable storage medium, is stored thereon with computer program, wherein the meter
The step of preceding method is realized when calculation machine program is executed by processor.
By adopting the above technical scheme, mark training sample concentrates at least two detections pair in each sample image in advance
The position of elephant and classification obtain the labeling position and mark classification of test object；Utilize the training sample set and the inspection
The labeling position for surveying object is trained target detection network, obtains target detection model；Using the training sample set, with
And the labeling position of the test object and mark classification are trained target classification network, obtain target classification network mould
Type；To obtain the target detection model of test object position for identification, and the target of test object classification for identification
Class models.In this way, utilizing the position and classification of a model while recognition detection object, the application in compared to the prior art
Using the position and classification of target detection model and target category model difference recognition detection object, test object knowledge can be improved
Other accuracy.
Detailed description of the invention
Fig. 1 is the first pass schematic diagram of image detecting method in the embodiment of the present application；
Fig. 2 is the second procedure schematic diagram of image detecting method in the embodiment of the present application；
Fig. 3 is the realization schematic diagram of a scenario of image detecting method in device embodiment itself；
Fig. 4 is the first composed structure schematic diagram of image detection device in the embodiment of the present application；
Fig. 5 is the second composed structure schematic diagram of image detection device in the embodiment of the present application.
Specific embodiment
The characteristics of in order to more fully hereinafter understand the embodiment of the present application and technology contents, with reference to the accompanying drawing to this Shen
Please the realization of embodiment be described in detail, appended attached drawing purposes of discussion only for reference is not used to limit the embodiment of the present application.
Embodiment one
The embodiment of the present application provides a kind of image detecting method, as shown in Figure 1, this method can specifically include:
Step 101: obtaining the training sample set comprising at least two sample images；Wherein, it is wrapped in each sample image
Containing at least two test objects；
Step 102: marking the position of at least two test objects and classification in the sample image in advance, obtain detection pair
The labeling position and mark classification of elephant；
Step 103: using the training sample set and the test object labeling position to target detection network into
Row training, obtains target detection model；Wherein, the target detection model is used to detect at least two detections in an image
The position of object；
Step 104: using the labeling position and mark classification of the training sample set and the test object to target
Sorter network is trained, and obtains target classification network model；Wherein, the object-class model is for detecting in an image
The classification of at least two test objects；
Step 105: image to be detected being identified using the target detection model and the object-class model, is obtained
The location information and classification information of at least two test objects into described image to be detected.
Here, the executing subject of step 101 to step 105 can be the processor of image detection device.
In some embodiments, the position using test object in the training sample set and the sample image
Confidence breath is trained target detection network, comprising: the sample image that the training sample is concentrated is input to the target
It detects in network, exports the predicted position of test object；The mesh is adjusted using the predicted position and labeling position of test object
Mark detection network, obtains target detection model.
Here, standard of the labeling position as the training effect of target detection model, between predicted position and labeling position
Error be unsatisfactory for default error criterion, show that current detection model does not reach desired effects, determine failure to train, need after
Continuous to adjust model parameter, the error between the predicted position and normal place made meets default error criterion, shows current
Detection model reaches desired effects, using current detection model as target detection model.
Here, lightweight detection network can be used in target detection network, such as: mobilenet-ssd, faster-
Rcnn, yolov3 etc..
In some embodiments, the labeling position and mark for utilizing the training sample set and the test object
Note classification is trained target classification network, comprising: using the labeling position of the test object, extracts the sample image
In include at least two test object at least two target image blocks；By corresponding at least two mesh of the sample image
Logo image block is input in the target classification network, exports the type of prediction of test object；Utilize the pre- of the test object
It surveys type and marking types adjusts the target classification network, obtain object-class model.
Here, standard of the mark classification as the training effect of object-class model is predicted between classification and mark classification
Error be unsatisfactory for default error criterion, show that current class model does not reach desired effects, determine failure to train, need after
Continuous adjustment model parameter, the prediction classification made is identical with standard category, shows that current class model reaches desired effects, will
Current detection model is as target detection model.
In some embodiments, the labeling position using the test object, extracts in the sample image and includes
At least two target image blocks of at least two test object, comprising: using the labeling position of the test object, extract
It include at least two subimage blocks of at least two test object in the sample image；To at least two subgraph
Block carries out size adjusting, obtains at least two target image blocks of pre-set dimension.
That is, since test object region area shared in sample image is different, in order to improve sorter network
Training effectiveness, handle the image that various sizes of image block is adjusted to fixed size by zooming in and out to image block
Block.
Here, lightweight sorter network can be used in target classification network, such as: mobilenet_v2, resnet50,
Shufflennetv2 etc..
In some embodiments, described to obtain the training sample set comprising at least two sample images, comprising: to obtain at least
One sample image；Data enhancing is carried out to an at least sample image, obtains at least two sample images；Using described
At least two sample images establish the training sample set.
Here, in order to increase data volume, alleviate network over-fitting, enhancing processing is carried out to sample image, can be obtained more
More training samples, multiple available images, amplified sample quantity can be improved model training on the basis of an image
Efficiency.Enhancing processing includes: to carry out random cropping, flip horizontal, size scaling, hue adjustment, brightness to given sample image
The pretreatments such as adjustment, saturation degree adjustment.
Further, step 105 can specifically include: obtain image to be detected；Wherein, include in described image to be detected
At least two test objects；Described image to be detected is input in the target detection model, described image to be detected is obtained
In at least two target detection objects location information；Utilize the position of at least two target detection objects in described image to be detected
Confidence breath, extracts at least two target image blocks comprising at least two target detections object；By at least two mesh
Logo image block is input in the object-class model, exports the classification letter of at least two test objects in described image to be detected
Breath.
In practical application, after the target detection model and object-class model that training is completed do model quantization, it can dispose
In the terminal with image detection function, when the function is called, target detection network frame first selects current image frame
In multiple target detection objects, then the picture that frame is selected is cut, target classification network model is given and is handled,
Thus arrive correct classification information.Here, terminal can be smart phone, PC (such as tablet computer, desktop
Brain, notebook, net book, palm PC), mobile phone, E-book reader, portable media player, audio/video
Player, video camera, virtual reality device and wearable device etc..
By adopting the above technical scheme, mark training sample concentrates at least two detections pair in each sample image in advance
The position of elephant and classification obtain the labeling position and mark classification of test object；Utilize the training sample set and the inspection
The labeling position for surveying object is trained target detection network, obtains target detection model；Using the training sample set, with
And the labeling position of the test object and mark classification are trained target classification network, obtain target classification network mould
Type；To obtain the target detection model of test object position for identification, and the target of test object classification for identification
Class models.In this way, utilizing the position and classification of a model while recognition detection object, the application in compared to the prior art
Using the position and classification of target detection model and target category model difference recognition detection object, test object knowledge can be improved
Other accuracy.
Embodiment two
The embodiment of the present application provides a kind of image detecting method, as shown in Fig. 2, this method can specifically include:
Step 201: obtaining the training sample set comprising at least two sample images；Wherein, it is wrapped in each sample image
Containing at least two test objects；
Here it is possible to establish training sample set according to test object, test object can be common object in daily life
Body, for example, desk, chair, automobile, animal, plant, people etc., test object may be the subdivision type of a type objects, such as
Different types of animal, plant etc..
In some embodiments, the method for obtaining training sample set can specifically include: obtain an at least sample image；
Data enhancing is carried out to an at least sample image, obtains at least two sample images；Utilize at least two sample
Image establishes the training sample set.
Here, in order to increase data volume, alleviate network over-fitting, enhancing processing is carried out to sample image, can be obtained more
More training samples, multiple available images, amplified sample quantity can be improved model training on the basis of an image
Efficiency.Enhancing processing includes: to carry out random cropping, flip horizontal, size scaling, hue adjustment, brightness to given sample image
The pretreatments such as adjustment, saturation degree adjustment.
Step 202: marking the position of at least two test objects and classification in the sample image in advance, obtain detection pair
The labeling position and mark classification of elephant；
It here, can be by the method that manually marks in each sample image after getting training sample set
The position of test object and classification are labeled, and obtained labeling position and mark classification are used as and judge target detection network and mesh
Mark the foundation that sorter network training is completed.Specifically, the position of test object can select the seat of the rectangle frame of test object for frame
Cursor position, the classification of test object can be the level-one of test object, second level, three-level or further segment classification.
Step 203: using the training sample set and the test object labeling position to target detection network into
Row training, obtains target detection model；Wherein, the target detection model is used to detect at least two detections in an image
The position of object；
Specifically, the sample image that the training sample is concentrated is input in the target detection network, output detection
The predicted position of object；The target detection network is adjusted using the predicted position and labeling position of test object, obtains target
Detection model.
Here, standard of the labeling position as the training effect of target detection model, between predicted position and labeling position
Error be unsatisfactory for default error criterion, show that current detection model does not reach desired effects, determine failure to train, need after
Continuous to adjust model parameter, the error between the predicted position and normal place made meets default error criterion, shows current
Detection model reaches desired effects, using current detection model as target detection model.
Here, lightweight detection network can be used in target detection network, such as: mobilenet-ssd, faster-
Rcnn, yolov3 etc..
Step 204: using the labeling position and mark classification of the training sample set and the test object to target
Sorter network is trained, and obtains target classification network model；Wherein, the object-class model is for detecting in an image
The classification of at least two test objects；
Specifically, being extracted in the sample image using the labeling position of the test object comprising described at least two
At least two target image blocks of test object；Corresponding at least two target image block of the sample image is input to described
In target classification network, the type of prediction of test object is exported；Utilize the type of prediction and marking types tune of the test object
The whole target classification network, obtains object-class model.
Here, standard of the mark classification as the training effect of object-class model is predicted between classification and mark classification
Error be unsatisfactory for default error criterion, show that current class model does not reach desired effects, determine failure to train, need after
Continuous adjustment model parameter, the prediction classification made is identical with standard category, shows that current class model reaches desired effects, will
Current detection model is as target detection model.
In some embodiments, the labeling position using the test object, extracts in the sample image and includes
At least two target image blocks of at least two test object, comprising: using the labeling position of the test object, extract
It include at least two subimage blocks of at least two test object in the sample image；To at least two subgraph
Block carries out size adjusting, obtains at least two target image blocks of pre-set dimension.
That is, since test object region area shared in sample image is different, in order to improve sorter network
Training effectiveness, handle the image that various sizes of image block is adjusted to fixed size by zooming in and out to image block
Block.
Here, lightweight sorter network can be used in target classification network, such as: mobilenet_v2, resnet50,
Shufflennetv2 etc..
Step 205: obtaining image to be detected；Described image to be detected is input in the target detection model, is obtained
The location information of at least two target detection objects in described image to be detected；
Image-recognizing method provided by the embodiments of the present application is mainly used in the real-time detection of more objects, i.e. image to be detected
In contain at least two test object, can more accurately detect the classification information of different objects compared to the prior art.
It should be noted that image-recognizing method provided by the present application also can be applied to the real-time detection of single object, i.e.,
It include a test object in image to be detected.
Here, when containing at least two test object in image to be detected, due to the image clearly of different test objects
Degree may be different, and clearly test object can identify that its classification information, the test object of Relative Fuzzy can not identify relatively
Its classification information out, therefore be determined to carry out at least two targets inspection of classification detection from least two test object
Survey object.
Specifically, described be input to described image to be detected in the target detection model, the mapping to be checked is obtained
The location information of at least two target detection objects as in, comprising: described image to be detected is input to the target detection mould
In type, the confidence of the location information and each location information of at least two test objects in described image to be detected is exported
Degree；Choose the location information that confidence level is greater than at least two target detection objects of confidence threshold value.Here, confidence level is used for table
The credibility of rectangle frame positional accuracy is levied, confidence level is higher, and the position for showing rectangle frame is more credible, and confidence level is lower to be shown
The position of rectangle frame is more insincere.
Step 206: using the location information of at least two target detection objects in described image to be detected, extracting includes institute
State at least two target image blocks of at least two target detection objects；
Step 207: at least two target image block being input in the object-class model, is exported described to be checked
The classification information of at least two test objects in altimetric image.
In some embodiments, the method also includes: based at least two test objects in described image to be detected
Classification information obtains the corresponding related information of classification information；The corresponding association of Display Category information in described image to be detected
Information.
That is, after the location information and classification information of test object has been determined, according to correct classification information
Provide the related information of test object, related information be specifically as follows being discussed in detail of test object, commodity purchasing address or
Other peripheral informations.Specifically, related information can be website links, user, which clicks classification information, can enter the network address chain
It connects.
For example, test object is plant, object-class model can be the subdivision class model of plant, obtain the specific of plant
Classification simultaneously provides related introduction information.Test object is cat, and object-class model can be the subdivision class model of cat, obtains cat
Specific category simultaneously provides related introduction information.
A kind of image-detection process is shown in Fig. 3, includes three test objects, i.e. refrigerator, air-conditioning in image to be detected
And people, image to be detected is input in target detection model first, determination block selects the position of the rectangle frame of test object, and
The confidence level of each rectangle frame determines that confidence level is greater than the target rectangle frame of confidence threshold value；Target rectangle circle is selected
Image is given in object-class model after being cut, and obtains the classification of each test object, finally image after sensing
The middle rectangle frame and classification for showing each test object.
By adopting the above technical scheme, mark training sample concentrates at least two detections pair in each sample image in advance
The position of elephant and classification obtain the labeling position and mark classification of test object；Utilize the training sample set and the inspection
The labeling position for surveying object is trained target detection network, obtains target detection model；Using the training sample set, with
And the labeling position of the test object and mark classification are trained target classification network, obtain target classification network mould
Type；To obtain the target detection model of test object position for identification, and the target of test object classification for identification
Class models.In this way, utilizing the position and classification of a model while recognition detection object, the application in compared to the prior art
Using the position and classification of target detection model and target category model difference recognition detection object, test object knowledge can be improved
Other accuracy.
Embodiment three
A kind of image detection device is additionally provided in the embodiment of the present application, as shown in figure 4, the device includes:
Acquiring unit 401, for obtaining the training sample set comprising at least two sample images；Wherein, each sample
Test object is contained at least two in image；
Mark unit 402 is obtained for marking the position of at least two test objects and classification in the sample image in advance
To the labeling position and mark classification of test object；
Processing unit 403 examines target for the labeling position using the training sample set and the test object
Survey grid network is trained, and obtains target detection model；Wherein, the target detection model is for detecting in an image at least
The position of two test objects；
Processing unit 403 is also used to labeling position and mark using the training sample set and the test object
Classification is trained target classification network, obtains target classification network model；Wherein, the object-class model is for detecting
The classification of at least two test objects in one image；
Processing unit 403 is also used to using the target detection model and the object-class model to image to be detected
It is identified, obtains the location information and classification information of at least two test objects in described image to be detected.
In some embodiments, the processing unit is inputted specifically for the sample image for concentrating the training sample
Into the target detection network, the predicted position of test object is exported；Utilize the predicted position and labeling position of test object
The target detection network is adjusted, target detection model is obtained.
In some embodiments, the processing unit extracts institute specifically for the labeling position using the test object
State at least two target image blocks in sample image comprising at least two test object；The sample image is corresponding
At least two target image blocks are input in the target classification network, export the type of prediction of test object；Utilize the inspection
The type of prediction and marking types for surveying object adjust the target classification network, obtain object-class model.
In some embodiments, the processing unit extracts institute specifically for the labeling position using the test object
State at least two subimage blocks in sample image comprising at least two test object；To at least two subimage block
Size adjusting is carried out, at least two target image blocks of pre-set dimension are obtained.
In some embodiments, the acquiring unit is specifically used for obtaining an at least sample image；To described at least one
It opens sample image and carries out data enhancing, obtain at least two sample images；Using described at least two sample image foundation
Training sample set.
In some embodiments, the acquiring unit is also used to obtain image to be detected；Wherein, described image to be detected
In contain at least two test object；
The processing unit is also used to for described image to be detected being input in the target detection model, obtains described
The location information of at least two target detection objects in image to be detected；It is examined using at least two targets in described image to be detected
The location information of object is surveyed, at least two target image blocks comprising at least two target detections object are extracted；
The processing unit is also used at least two target image block being input in the object-class model,
Export the classification information of at least two test objects in described image to be detected.
In some embodiments, the processing unit is examined specifically for described image to be detected is input to the target
It surveys in model, exports setting for the location information and each location information of at least two test objects in described image to be detected
Reliability；Choose the location information that confidence level is greater than at least two target detection objects of confidence threshold value.
Based on the hardware realization of each unit in above-mentioned image detection device, the embodiment of the present application also provides another images
Detection device, as shown in figure 5, the device includes: processor 501 and is configured to store the computer that can be run on a processor
The memory 502 of program；
Wherein, when processor 501 is configured to operation computer program, the method and step in previous embodiment is executed.
Certainly, when practical application, as shown in figure 5, the various components in the device are coupled in one by bus system 503
It rises.It is understood that bus system 503 is for realizing the connection communication between these components.It includes data/address bus that bus system 503, which is removed,
Except, it further include power bus, control bus and status signal bus in addition.It, will be various in Fig. 5 but for the sake of clear explanation
Bus is all designated as bus system 503.
In practical applications, above-mentioned processor can be application-specific IC (ASIC, Application
Specific Integrated Circuit), digital signal processing device (DSPD, Digital Signal Processing
Device), programmable logic device (PLD, Programmable Logic Device), field programmable gate array
At least one of (Field-Programmable Gate Array, FPGA), controller, microcontroller, microprocessor.It can
To understand ground, for different equipment, the electronic device for realizing above-mentioned processor function can also be other, the application reality
Example is applied to be not especially limited.
Above-mentioned memory can be volatile memory (volatile memory), such as random access memory (RAM,
Random-Access Memory)；Or nonvolatile memory (non-volatile memory), such as read-only memory
(ROM, Read-Only Memory), flash memory (flash memory), hard disk (HDD, Hard Disk Drive) or solid
State hard disk (SSD, Solid-State Drive)；Or the combination of the memory of mentioned kind, and to processor provide instruction and
Data.
The embodiment of the present application also provides a kind of computer readable storage mediums, for storing computer program.
Optionally, which can be applied to the dress of any one image detection in the embodiment of the present application
In setting, and the computer program executes computer in each method of the embodiment of the present application by the corresponding of processor realization
Process, for sake of simplicity, details are not described herein.
In several embodiments provided herein, it should be understood that disclosed device and method can pass through it
Its mode is realized.Apparatus embodiments described above are merely indicative, for example, the division of the unit, only
A kind of logical function partition, there may be another division manner in actual implementation, such as: multiple units or components can combine, or
It is desirably integrated into another system, or some features can be ignored or not executed.In addition, shown or discussed each composition portion
Mutual coupling or direct-coupling or communication connection is divided to can be through some interfaces, the INDIRECT COUPLING of equipment or unit
Or communication connection, it can be electrical, mechanical or other forms.
Above-mentioned unit as illustrated by the separation member, which can be or may not be, to be physically separated, aobvious as unit
The component shown can be or may not be physical unit, it can and it is in one place, it may be distributed over multiple network lists
In member；Some or all of units can be selected to achieve the purpose of the solution of this embodiment according to the actual needs.
In addition, each functional unit in various embodiments of the present invention can be fully integrated into a processing module, it can also
To be each unit individually as a unit, can also be integrated in one unit with two or more units；It is above-mentioned
Integrated unit both can take the form of hardware realization, can also realize in the form of hardware adds SFU software functional unit.This
Field those of ordinary skill, which is understood that, realizes that all or part of the steps of above method embodiment can be by program instruction phase
The hardware of pass is completed, and program above-mentioned can be stored in a computer readable storage medium, which when being executed, holds
Row step including the steps of the foregoing method embodiments；And storage medium above-mentioned include: movable storage device, read-only memory (ROM,
Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic or disk etc. it is various
It can store the medium of program code.
Disclosed method in several embodiments of the method provided herein, in the absence of conflict can be any group
It closes, obtains new embodiment of the method.
Disclosed feature in several product embodiments provided herein, in the absence of conflict can be any group
It closes, obtains new product embodiments.
Disclosed feature in several methods provided herein or apparatus embodiments, in the absence of conflict can be with
Any combination obtains new embodiment of the method or apparatus embodiments.
The above description is merely a specific embodiment, but scope of protection of the present invention is not limited thereto, any
Those familiar with the art in the technical scope disclosed by the present invention, can easily think of the change or the replacement, and should all contain
Lid is within protection scope of the present invention.Therefore, protection scope of the present invention should be based on the protection scope of the described claims.
Claims (10)
1. a kind of image detecting method, which is characterized in that the described method includes:
Obtain the training sample set comprising at least two sample images；Wherein, inspection is contained at least two in each sample image
Survey object；
The position of at least two test objects and classification in the sample image are marked in advance, obtain the labeling position of test object
With mark classification；
Target detection network is trained using the training sample set and the labeling position of the test object, is obtained
Target detection model；Wherein, the target detection model is used to detect the position of at least two test objects in an image；
Target classification network is carried out using the training sample set and the labeling position of the test object and mark classification
Training, obtains target classification network model；Wherein, the object-class model is for detecting at least two detection in an image
The classification of object；
Image to be detected is identified using the target detection model and the object-class model, is obtained described to be detected
The location information and classification information of at least two test objects in image.
2. the method according to claim 1, wherein described utilize the training sample set and the sample
The location information of test object is trained target detection network in image, comprising:
The sample image that the training sample is concentrated is input in the target detection network, the prediction bits of test object are exported
It sets；
The target detection network is adjusted using the predicted position and labeling position of test object, obtains target detection model.
3. the method according to claim 1, wherein described utilize the training sample set and the detection
The labeling position and mark classification of object are trained target classification network, comprising:
Using the labeling position of the test object, extract in the sample image comprising at least two test object extremely
Few two target image blocks；
Corresponding at least two target image block of the sample image is input in the target classification network, output detection pair
The type of prediction of elephant；
The target classification network is adjusted using the type of prediction and marking types of the test object, obtains target classification mould
Type.
4. according to the method described in claim 3, it is characterized in that, the labeling position using the test object, is extracted
It include at least two target image blocks of at least two test object in the sample image, comprising:
Using the labeling position of the test object, extract in the sample image comprising at least two test object extremely
Few two subimage blocks；
Size adjusting is carried out at least two subimage block, obtains at least two target image blocks of pre-set dimension.
5. the method according to claim 1, wherein described obtain the training sample comprising at least two sample images
This collection, comprising:
Obtain an at least sample image；
Data enhancing is carried out to an at least sample image, obtains at least two sample images；
The training sample set is established using at least two sample image.
6. the method according to claim 1, wherein described utilize the target detection model and the target point
Class model identifies image to be detected, obtains the location information and class of at least two test objects in described image to be detected
Other information, comprising:
Obtain image to be detected；Wherein, test object is contained at least two in described image to be detected；
Described image to be detected is input in the target detection model, at least two targets in described image to be detected are obtained
The location information of test object；
Using the location information of at least two target detection objects in described image to be detected, extracting includes at least two mesh
Mark at least two target image blocks of test object；
At least two target image block is input in the object-class model, is exported in described image to be detected at least
The classification information of two test objects.
7. according to the method described in claim 6, it is characterized in that, described be input to the target inspection for described image to be detected
It surveys in model, obtains the location information of at least two target detection objects in described image to be detected, comprising:
Described image to be detected is input in the target detection model, at least two detection in described image to be detected is exported
The confidence level of the location information of object and each location information；
Choose the location information that confidence level is greater than at least two target detection objects of confidence threshold value.
8. a kind of image detection device, which is characterized in that described device includes:
Acquiring unit, for obtaining the training sample set comprising at least two sample images；Wherein, it is wrapped in each sample image
Containing at least two test objects；
Mark unit is detected for marking the position of at least two test objects and classification in the sample image in advance
The labeling position and mark classification of object；
Processing unit, for the labeling position using the training sample set and the test object to target detection network
It is trained, obtains target detection model；Wherein, the target detection model is used to detect at least two inspections in an image
Survey the position of object；
Processing unit for the labeling position using the training sample set and the test object and marks classification to mesh
Mark sorter network is trained, and obtains target classification network model；Wherein, the object-class model is for detecting an image
In at least two test objects classification；
Processing unit, for being identified using the target detection model and the object-class model to image to be detected,
Obtain the location information and classification information of at least two test objects in described image to be detected.
9. a kind of image detection device, described device includes: processor and is configured to store the meter that can be run on a processor
The memory of calculation machine program,
Wherein, when the processor is configured to run the computer program, perform claim requires any one of 1 to 7 the method
The step of.
10. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the computer program quilt
The step of claim 1 to 7 described in any item methods are realized when processor executes.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910657541.4A CN110378420A (en) | 2019-07-19 | 2019-07-19 | A kind of image detecting method, device and computer readable storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910657541.4A CN110378420A (en) | 2019-07-19 | 2019-07-19 | A kind of image detecting method, device and computer readable storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110378420A true CN110378420A (en) | 2019-10-25 |
Family
ID=68254337
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910657541.4A Pending CN110378420A (en) | 2019-07-19 | 2019-07-19 | A kind of image detecting method, device and computer readable storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110378420A (en) |
Cited By (31)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110675399A (en) * | 2019-10-28 | 2020-01-10 | 上海悦易网络信息技术有限公司 | Screen appearance flaw detection method and equipment |
| CN110796107A (en) * | 2019-11-04 | 2020-02-14 | 南京北旨智能科技有限公司 | Power inspection image defect identification method and system and power inspection unmanned aerial vehicle |
| CN110796647A (en) * | 2019-10-28 | 2020-02-14 | 上海悦易网络信息技术有限公司 | Method and device for detecting defects of screen area of electronic device |
| CN110827246A (en) * | 2019-10-28 | 2020-02-21 | 上海悦易网络信息技术有限公司 | Electronic equipment frame appearance flaw detection method and equipment |
| CN110827249A (en) * | 2019-10-28 | 2020-02-21 | 上海悦易网络信息技术有限公司 | Electronic equipment backboard appearance flaw detection method and equipment |
| CN110827244A (en) * | 2019-10-28 | 2020-02-21 | 上海悦易网络信息技术有限公司 | Method and equipment for detecting appearance flaws of electronic equipment |
| CN111061898A (en) * | 2019-12-13 | 2020-04-24 | Oppo（重庆）智能科技有限公司 | Image processing method, image processing device, computer equipment and storage medium |
| CN111078908A (en) * | 2019-11-28 | 2020-04-28 | 北京云聚智慧科技有限公司 | Data annotation detection method and device |
| CN111124862A (en) * | 2019-12-24 | 2020-05-08 | 北京安兔兔科技有限公司 | Intelligent equipment performance testing method and device and intelligent equipment |
| CN111291799A (en) * | 2020-01-21 | 2020-06-16 | 青梧桐有限责任公司 | Room window classification model construction method, room window classification method and room window classification system |
| CN111400533A (en) * | 2020-03-02 | 2020-07-10 | 北京三快在线科技有限公司 | Image screening method and device, electronic equipment and storage medium |
| CN111444807A (en) * | 2020-03-19 | 2020-07-24 | 北京迈格威科技有限公司 | Target detection method, device, electronic equipment and computer readable medium |
| CN111640103A (en) * | 2020-05-29 | 2020-09-08 | 北京百度网讯科技有限公司 | Image detection method, device, equipment and storage medium |
| CN111738183A (en) * | 2020-06-28 | 2020-10-02 | 达闼机器人有限公司 | Data labeling method and device |
| CN111783635A (en) * | 2020-06-30 | 2020-10-16 | 北京百度网讯科技有限公司 | Image annotation method, device, equipment and storage medium |
| CN111797934A (en) * | 2020-07-10 | 2020-10-20 | 北京嘉楠捷思信息技术有限公司 | Road sign identification method and device |
| CN111881944A (en) * | 2020-07-08 | 2020-11-03 | 贵州无忧天空科技有限公司 | Method, electronic device and computer readable medium for image authentication |
| CN111950391A (en) * | 2020-07-23 | 2020-11-17 | 中国农业科学院农业信息研究所 | Fruit tree bud recognition method and device |
| CN112132220A (en) * | 2020-09-24 | 2020-12-25 | 杭州海康威视数字技术股份有限公司 | Self-training method, system, device, electronic equipment and storage medium |
| CN112308054A (en) * | 2020-12-29 | 2021-02-02 | 广东科凯达智能机器人有限公司 | Automatic reading method of multifunctional digital meter based on target detection algorithm |
| CN112418287A (en) * | 2020-11-16 | 2021-02-26 | 深圳市优必选科技股份有限公司 | Image pre-labeling method and device, electronic equipment and medium |
| CN112699842A (en) * | 2021-01-13 | 2021-04-23 | 上海镜河科技有限公司 | Pet identification method, device, equipment and computer readable storage medium |
| CN112861948A (en) * | 2021-01-29 | 2021-05-28 | 南京中兴力维软件有限公司 | Method, device and equipment for identifying working state of siphon type sludge discharge equipment |
| CN112850436A (en) * | 2019-11-28 | 2021-05-28 | 宁波微科光电股份有限公司 | Pedestrian trend detection method and system of elevator intelligent light curtain |
| CN113034449A (en) * | 2021-03-11 | 2021-06-25 | 深圳市优必选科技股份有限公司 | Target detection model training method and device and communication equipment |
| CN113076860A (en) * | 2021-03-30 | 2021-07-06 | 南京大学环境规划设计研究院集团股份公司 | Bird detection system under field scene |
| CN113095434A (en) * | 2021-04-27 | 2021-07-09 | 深圳市商汤科技有限公司 | Target detection method and device, electronic equipment and storage medium |
| CN113158969A (en) * | 2021-05-10 | 2021-07-23 | 上海畅选科技合伙企业(有限合伙) | Apple appearance defect identification system and method |
| CN113591901A (en) * | 2021-06-10 | 2021-11-02 | 中国航天时代电子有限公司 | Target detection method based on anchor frame |
| CN113762292A (en) * | 2020-06-03 | 2021-12-07 | 杭州海康威视数字技术股份有限公司 | Training data acquisition method and device and model training method and device |
| CN113780480A (en) * | 2021-11-11 | 2021-12-10 | 深圳佑驾创新科技有限公司 | Method for constructing multi-target detection and category identification model based on YOLOv5 |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108961267A (en) * | 2018-06-19 | 2018-12-07 | Oppo广东移动通信有限公司 | Image processing method, picture processing unit and terminal device |
| CN108985214A (en) * | 2018-07-09 | 2018-12-11 | 上海斐讯数据通信技术有限公司 | The mask method and device of image data |
| CN109063740A (en) * | 2018-07-05 | 2018-12-21 | 高镜尧 | The detection model of ultrasonic image common-denominator target constructs and detection method, device |
| CN109801284A (en) * | 2019-01-25 | 2019-05-24 | 华中科技大学 | A kind of high iron catenary insulator breakdown detection method based on deep learning |
| CN109977872A (en) * | 2019-03-27 | 2019-07-05 | 北京迈格威科技有限公司 | Motion detection method, device, electronic equipment and computer readable storage medium |
-
2019
- 2019-07-19 CN CN201910657541.4A patent/CN110378420A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108961267A (en) * | 2018-06-19 | 2018-12-07 | Oppo广东移动通信有限公司 | Image processing method, picture processing unit and terminal device |
| CN109063740A (en) * | 2018-07-05 | 2018-12-21 | 高镜尧 | The detection model of ultrasonic image common-denominator target constructs and detection method, device |
| CN108985214A (en) * | 2018-07-09 | 2018-12-11 | 上海斐讯数据通信技术有限公司 | The mask method and device of image data |
| CN109801284A (en) * | 2019-01-25 | 2019-05-24 | 华中科技大学 | A kind of high iron catenary insulator breakdown detection method based on deep learning |
| CN109977872A (en) * | 2019-03-27 | 2019-07-05 | 北京迈格威科技有限公司 | Motion detection method, device, electronic equipment and computer readable storage medium |
Cited By (42)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110796647A (en) * | 2019-10-28 | 2020-02-14 | 上海悦易网络信息技术有限公司 | Method and device for detecting defects of screen area of electronic device |
| CN110827246A (en) * | 2019-10-28 | 2020-02-21 | 上海悦易网络信息技术有限公司 | Electronic equipment frame appearance flaw detection method and equipment |
| CN110827249A (en) * | 2019-10-28 | 2020-02-21 | 上海悦易网络信息技术有限公司 | Electronic equipment backboard appearance flaw detection method and equipment |
| CN110827244A (en) * | 2019-10-28 | 2020-02-21 | 上海悦易网络信息技术有限公司 | Method and equipment for detecting appearance flaws of electronic equipment |
| CN110675399A (en) * | 2019-10-28 | 2020-01-10 | 上海悦易网络信息技术有限公司 | Screen appearance flaw detection method and equipment |
| CN110796107A (en) * | 2019-11-04 | 2020-02-14 | 南京北旨智能科技有限公司 | Power inspection image defect identification method and system and power inspection unmanned aerial vehicle |
| CN111078908B (en) * | 2019-11-28 | 2023-06-09 | 北京云聚智慧科技有限公司 | Method and device for detecting data annotation |
| CN111078908A (en) * | 2019-11-28 | 2020-04-28 | 北京云聚智慧科技有限公司 | Data annotation detection method and device |
| CN112850436A (en) * | 2019-11-28 | 2021-05-28 | 宁波微科光电股份有限公司 | Pedestrian trend detection method and system of elevator intelligent light curtain |
| CN111061898A (en) * | 2019-12-13 | 2020-04-24 | Oppo（重庆）智能科技有限公司 | Image processing method, image processing device, computer equipment and storage medium |
| CN111124862A (en) * | 2019-12-24 | 2020-05-08 | 北京安兔兔科技有限公司 | Intelligent equipment performance testing method and device and intelligent equipment |
| CN111124862B (en) * | 2019-12-24 | 2024-01-30 | 北京安兔兔科技有限公司 | Intelligent device performance testing method and device and intelligent device |
| CN111291799A (en) * | 2020-01-21 | 2020-06-16 | 青梧桐有限责任公司 | Room window classification model construction method, room window classification method and room window classification system |
| CN111400533B (en) * | 2020-03-02 | 2023-10-17 | 北京三快在线科技有限公司 | Image screening method, device, electronic equipment and storage medium |
| CN111400533A (en) * | 2020-03-02 | 2020-07-10 | 北京三快在线科技有限公司 | Image screening method and device, electronic equipment and storage medium |
| CN111444807A (en) * | 2020-03-19 | 2020-07-24 | 北京迈格威科技有限公司 | Target detection method, device, electronic equipment and computer readable medium |
| CN111444807B (en) * | 2020-03-19 | 2023-09-22 | 北京迈格威科技有限公司 | Target detection method, device, electronic equipment and computer readable medium |
| CN111640103A (en) * | 2020-05-29 | 2020-09-08 | 北京百度网讯科技有限公司 | Image detection method, device, equipment and storage medium |
| CN113762292A (en) * | 2020-06-03 | 2021-12-07 | 杭州海康威视数字技术股份有限公司 | Training data acquisition method and device and model training method and device |
| CN113762292B (en) * | 2020-06-03 | 2024-02-02 | 杭州海康威视数字技术股份有限公司 | Training data acquisition method and device and model training method and device |
| CN111738183A (en) * | 2020-06-28 | 2020-10-02 | 达闼机器人有限公司 | Data labeling method and device |
| CN111783635A (en) * | 2020-06-30 | 2020-10-16 | 北京百度网讯科技有限公司 | Image annotation method, device, equipment and storage medium |
| CN111881944A (en) * | 2020-07-08 | 2020-11-03 | 贵州无忧天空科技有限公司 | Method, electronic device and computer readable medium for image authentication |
| CN111797934A (en) * | 2020-07-10 | 2020-10-20 | 北京嘉楠捷思信息技术有限公司 | Road sign identification method and device |
| CN111950391A (en) * | 2020-07-23 | 2020-11-17 | 中国农业科学院农业信息研究所 | Fruit tree bud recognition method and device |
| CN112132220A (en) * | 2020-09-24 | 2020-12-25 | 杭州海康威视数字技术股份有限公司 | Self-training method, system, device, electronic equipment and storage medium |
| CN112418287B (en) * | 2020-11-16 | 2023-10-13 | 深圳市优必选科技股份有限公司 | Image pre-labeling method, device, electronic equipment and medium |
| CN112418287A (en) * | 2020-11-16 | 2021-02-26 | 深圳市优必选科技股份有限公司 | Image pre-labeling method and device, electronic equipment and medium |
| CN112308054A (en) * | 2020-12-29 | 2021-02-02 | 广东科凯达智能机器人有限公司 | Automatic reading method of multifunctional digital meter based on target detection algorithm |
| CN112308054B (en) * | 2020-12-29 | 2021-07-20 | 广东科凯达智能机器人有限公司 | Automatic reading method of multifunctional digital meter based on target detection algorithm |
| CN112699842A (en) * | 2021-01-13 | 2021-04-23 | 上海镜河科技有限公司 | Pet identification method, device, equipment and computer readable storage medium |
| CN112861948A (en) * | 2021-01-29 | 2021-05-28 | 南京中兴力维软件有限公司 | Method, device and equipment for identifying working state of siphon type sludge discharge equipment |
| CN112861948B (en) * | 2021-01-29 | 2024-04-09 | 南京中兴力维软件有限公司 | Method, device and equipment for identifying working state of siphon type sludge discharge equipment |
| CN113034449A (en) * | 2021-03-11 | 2021-06-25 | 深圳市优必选科技股份有限公司 | Target detection model training method and device and communication equipment |
| CN113034449B (en) * | 2021-03-11 | 2023-12-15 | 深圳市优必选科技股份有限公司 | Target detection model training method and device and communication equipment |
| CN113076860B (en) * | 2021-03-30 | 2022-02-25 | 南京大学环境规划设计研究院集团股份公司 | Bird detection system under field scene |
| CN113076860A (en) * | 2021-03-30 | 2021-07-06 | 南京大学环境规划设计研究院集团股份公司 | Bird detection system under field scene |
| CN113095434A (en) * | 2021-04-27 | 2021-07-09 | 深圳市商汤科技有限公司 | Target detection method and device, electronic equipment and storage medium |
| CN113158969A (en) * | 2021-05-10 | 2021-07-23 | 上海畅选科技合伙企业(有限合伙) | Apple appearance defect identification system and method |
| CN113591901A (en) * | 2021-06-10 | 2021-11-02 | 中国航天时代电子有限公司 | Target detection method based on anchor frame |
| CN113780480B (en) * | 2021-11-11 | 2022-02-22 | 深圳佑驾创新科技有限公司 | Method for constructing multi-target detection and category identification model based on YOLOv5 |
| CN113780480A (en) * | 2021-11-11 | 2021-12-10 | 深圳佑驾创新科技有限公司 | Method for constructing multi-target detection and category identification model based on YOLOv5 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110378420A (en) | A kind of image detecting method, device and computer readable storage medium | |
| CN109492643B (en) | Certificate identification method and device based on OCR, computer equipment and storage medium | |
| US20210192202A1 (en) | Recognizing text in image data | |
| Beijbom et al. | Menu-match: Restaurant-specific food logging from images | |
| Puri et al. | Recognition and volume estimation of food intake using a mobile device | |
| CN109447169A (en) | The training method of image processing method and its model, device and electronic system | |
| CN111340126A (en) | Article identification method and device, computer equipment and storage medium | |
| French et al. | Deep neural networks for analysis of fisheries surveillance video and automated monitoring of fish discards | |
| CN107808120A (en) | Glasses localization method, device and storage medium | |
| CN109740571A (en) | The method of Image Acquisition, the method, apparatus of image procossing and electronic equipment | |
| Aslan et al. | Benchmarking algorithms for food localization and semantic segmentation | |
| CN108229418B (en) | Human body key point detection method and apparatus, electronic device, storage medium, and program | |
| CN107918767B (en) | Object detection method, device, electronic equipment and computer-readable medium | |
| CN111061898A (en) | Image processing method, image processing device, computer equipment and storage medium | |
| US20190294863A9 (en) | Method and apparatus for face classification | |
| WO2021238548A1 (en) | Region recognition method, apparatus and device, and readable storage medium | |
| CN110222582B (en) | Image processing method and camera | |
| CN110533654A (en) | The method for detecting abnormality and device of components | |
| CN109857878B (en) | Article labeling method and device, electronic equipment and storage medium | |
| CN110765882A (en) | Video tag determination method, device, server and storage medium | |
| CN111382766A (en) | Equipment fault detection method based on fast R-CNN | |
| CN115115825B (en) | Method, device, computer equipment and storage medium for detecting object in image | |
| CN111124863A (en) | Intelligent equipment performance testing method and device and intelligent equipment | |
| CN113780116A (en) | Invoice classification method and device, computer equipment and storage medium | |
| Kumar et al. | Drone-based apple detection: Finding the depth of apples using YOLOv7 architecture with multi-head attention mechanism |
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 |