CN110443322A - Image processing method, device, server and readable storage medium storing program for executing - Google Patents
Image processing method, device, server and readable storage medium storing program for executing Download PDFInfo
- Publication number
- CN110443322A CN110443322A CN201910762042.1A CN201910762042A CN110443322A CN 110443322 A CN110443322 A CN 110443322A CN 201910762042 A CN201910762042 A CN 201910762042A CN 110443322 A CN110443322 A CN 110443322A
- Authority
- CN
- China
- Prior art keywords
- image
- processed
- power equipment
- feature
- learning model
- 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
- G06F18/217—Validation; Performance evaluation; Active pattern learning techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The application provides a kind of image processing method, device, server and readable storage medium storing program for executing, is related to computer image processing technology field.This method comprises: obtaining image to be processed, image to be processed includes the infrared image for acquiring power equipment and obtaining；By completing the deep learning model extraction image to be processed after training in the first characteristics of image of the first default convolutional layer, and image to be processed is extracted in the second characteristics of image of the second default convolutional layer；Fusion Features are carried out to the first characteristics of image, the second characteristics of image according to the first characteristics of image and the second characteristics of image, and feature extraction is carried out to the characteristics of image obtained after fusion, obtain target signature；The problem of image to be processed is identified according to target signature by deep learning model, the recognition result for indicating the type of power equipment in image to be processed is obtained, power equipment infrared image recognition result poor accuracy can be improved.
Description
Technical field
The present invention relates to computer image processing technology field, in particular to a kind of image processing method, device,
Server and readable storage medium storing program for executing.
Background technique
In technical field of electric power, need to detect the safety of equipment.For example, it is desired to which the temperature to equipment is examined
It surveys, is burned to avoid equipment because of high temperature.The mode of the temperature of detection device is usual at present are as follows: is arranged on measurement equipment to be checked
Temperature sensor detects the temperature of measurement equipment to be checked by way of temperature sensor contact arrangement.Equipment in electric power computer room
It is more, cause the layout difficulty of hardware circuit big.It is currently, there are and is identified in electric power computer room using infrared image by machine learning
Each equipment type, be then based on the temperature that infrared image carrys out detection device.And collected infrared image usually compares mould
Paste, so that the recognition result poor accuracy that identification equipment obtains.
Summary of the invention
The application provides a kind of image processing method, device, server and readable storage medium storing program for executing, can improve and set to electric power
The problem of standby infrared image recognition result poor accuracy.
To achieve the goals above, technical solution provided by the embodiment of the present application is as follows:
In a first aspect, the embodiment of the present application provides a kind of image processing method, which comprises
Image to be processed is obtained, the image to be processed includes the infrared image for acquiring power equipment and obtaining；Pass through completion
Image to be processed described in deep learning model extraction after training and is extracted in the first characteristics of image of the first default convolutional layer
Second characteristics of image of the image to be processed in the second default convolutional layer；According to the first image feature and second figure
Picture feature carries out Fusion Features to the first image feature, second characteristics of image, and special to the image obtained after fusion
Sign carries out feature extraction, obtains target signature；By the deep learning model according to the target signature to described to be processed
Image is identified, the recognition result that the type of power equipment is indicated in the image to be processed is obtained.
In the above-described embodiment, method is special in the first image of the first default convolutional layer by extracting image to be processed
It levies, in the second characteristics of image of the second default convolutional layer, then carries out Fusion Features, and fused characteristics of image is utilized to carry out
Feature extraction recycles the target signature extracted to identify to image to be processed.Based on this, compared to directly using to
The characteristics of image of image is handled, the characteristics of image in target signature is enhanced, so that identifying electricity using target signature
When the type of power equipment, help to utilize the accuracy improved with enhanced target signature to power equipment type identification.
With reference to first aspect, in some alternative embodiments, according to the first image feature and second figure
Picture feature carries out Fusion Features to the first image feature, second characteristics of image, and special to the image obtained after fusion
Sign carries out feature extraction, obtains target signature, comprising: the size based on the first image feature, second characteristics of image,
The size of the first image feature and/or second characteristics of image is adjusted, so that the first image feature and described the
The size of two characteristics of image is identical；The first image feature, second characteristics of image are overlaped, obtained fused
Characteristics of image；By fused characteristics of image described in the deep learning model extraction, the target signature is obtained.
In the above-described embodiment, it is adjusted by the size to the first characteristics of image, the second characteristics of image, so that adjusting
Size after section is identical, is easy to implement Fusion Features to obtain fused characteristics of image.Based on this, fused characteristics of image
Feature can be enhanced, the feature for extracting the target signature that fused characteristics of image obtains can also be enhanced, thus
Help to improve the accuracy of power equipment type in infrared image.
With reference to first aspect, in some alternative embodiments, by completing the deep learning model extraction after training
The image to be processed is in the first characteristics of image of the first default convolutional layer, and to extract the image to be processed default second
Second characteristics of image of convolutional layer, comprising: by the deep learning model, based on the corresponding volume of the described first default convolutional layer
Product parameter carries out convolution algorithm to the image to be processed and obtains the first image feature；By the deep learning model,
Convolution algorithm is carried out to the image to be processed based on the described second default convolutional layer corresponding deconvolution parameter and obtains described second
Characteristics of image.
In the above-described embodiment, convolution is carried out to characteristics of image by deconvolution parameter, respective roll product can be obtained
The characteristics of image of layer, convenient for carrying out Fusion Features using characteristics of image.
With reference to first aspect, in some alternative embodiments, the image to be processed is indicated in the recognition result
It is middle there are when the power equipment of specified type, the method also includes: the specified type is determined in the image to be processed
Image-region of the power equipment in the image to be processed；According to the infrared figure of image-region in the image to be processed
Picture determines the temperature data of the power equipment of the specified type.
In the above-described embodiment, the image-region by the power equipment using specified type in image to be processed,
Determine the temperature of the power equipment, help to improve determined by temperature accuracy, avoid by be not specified type electricity
The image-region of power equipment is as identification region, to determine the temperature of the power equipment, so that the temperature and electric power that determine
Equipment does not correspond to, and leads to the power equipment temperature inaccuracy calculated.
With reference to first aspect, in some alternative embodiments, the method also includes: be greater than in the temperature data
Or when being equal to preset threshold, the terminal device of Xiang Zhiding issues prompt information.
In the above-described embodiment, by sending out prompt information to terminal device, facilitate administrative staff and find electric power in time
The case where equipment is greater than or equal to preset threshold there are temperature data.When temperature data is greater than or equal to preset threshold, electric power
Usually there is the risk being burned by high temperature in equipment.Based on this, facilitates administrative staff and power equipment is safeguarded in time,
Improve the safety of power equipment.
With reference to first aspect, in some alternative embodiments, it is mentioned by the deep learning model after completing to train
Take the image to be processed before the first characteristics of image of the convolutional layer in the first default number of plies, the method also includes: it obtains
Take training image collection, the training image collection includes multiple training images, each training image include power equipment and with
The corresponding label of the type of the power equipment；According to the training image collection, training predetermined deep learning model is completed
The deep learning model after training.
In the above-described embodiment, by being trained to deep learning model, it can be improved deep learning model to red
The accuracy that the type of power equipment is identified in outer image.
With reference to first aspect, in some alternative embodiments, the power equipment includes breaker, transformer, electricity
At least one of machine.
Second aspect, the embodiment of the present application also provide a kind of image processing apparatus, and described device includes:
Image acquisition unit, for obtaining image to be processed, the image to be processed includes that acquisition power equipment obtains
Infrared image；
Feature extraction unit, for image to be processed described in the deep learning model extraction after being trained by completion first
First characteristics of image of default convolutional layer, and the image to be processed is extracted in the second image spy of the second default convolutional layer
Sign；
Fusion Features unit is used for according to the first image feature and second characteristics of image to the first image
Feature, second characteristics of image carry out Fusion Features, and carry out feature extraction to the characteristics of image obtained after fusion, obtain mesh
Mark feature；
Recognition unit, for being carried out according to the target signature to the image to be processed by the deep learning model
Identification obtains the recognition result that the type of power equipment is indicated in the image to be processed.
The third aspect, the embodiment of the present application also provide a kind of server, including memory, the processor to intercouple, institute
It states and stores computer program in memory, when the computer program is executed by the processor, so that the server is held
The above-mentioned method of row.
Fourth aspect, the embodiment of the present application also provide a kind of computer readable storage medium, in the readable storage medium storing program for executing
It is stored with computer program, when the computer program is run on computers, so that the computer executes above-mentioned side
Method.
Detailed description of the invention
Technical solution in ord to more clearly illustrate embodiments of the present application, below will be to needed in the embodiment attached
Figure is briefly described.It should be appreciated that the following drawings illustrates only some embodiments of the application, therefore it is not construed as pair
The restriction of range for those of ordinary skill in the art without creative efforts, can also be according to this
A little attached drawings obtain other relevant attached drawings.
Fig. 1 is the communication connection schematic diagram of server provided by the embodiments of the present application, terminal device.
Fig. 2 is the structural representation of server provided by the embodiments of the present application.
Fig. 3 is the flow diagram of image processing method provided by the embodiments of the present application.
Fig. 4 is the functional block diagram of image processing apparatus provided by the embodiments of the present application.
Icon: 10- server；11- processing module；12- memory module；13- communication module；20- terminal device；100- figure
As processing unit；110- image acquisition unit；120- feature extraction unit；130- Fusion Features unit；140- recognition unit.
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 is described.It needs
It is noted that term " first ", " second " etc. are only used for distinguishing description, it is not understood to indicate or imply relatively important
Property.
Fig. 1 is please referred to, the embodiment of the present application provides a kind of server 10, which can set at least one terminal
Standby 20 establish communication connection by network, to carry out data interaction.For example, the available power equipment of terminal device 20 is infrared
Image, and the infrared image that will acquire is sent to server 10, so that server 10 carries out identifying processing to the infrared image.
Server 10 can install or be stored in advance deep learning model, can be by deep learning model to the infrared of power equipment
Image is identified, to determine the type of electronic equipment.Wherein, power equipment includes but is not limited to breaker, transformer, power generation
The equipment such as mechanical, electrical motivation.
Understandably, the quantity of the terminal device 20 communicated to connect with server 10 can be one, or it is multiple,
Here the quantity of the terminal device 20 communicated to connect with server 10 is not especially limited.
Terminal device 20 may be, but not limited to, PC (Personal Computer, PC), tablet computer, individual
Digital assistants (Personal Digital Assistant, PDA), mobile internet surfing equipment (Mobile Internet Device,
MID) etc..Network may be, but not limited to, cable network or wireless network.
Referring to figure 2., in the present embodiment, server 10 may include processing module 11, memory module 12, communication module
13 and image processing apparatus 100, processing module 11, memory module 12, communication module 13 and image processing apparatus 100 it is each
It is directly or indirectly electrically connected between element, to realize the transmission or interaction of data.For example, these elements can lead between each other
It crosses one or more communication bus or signal wire is realized and is electrically connected.
Processing module 11 can be a kind of IC chip, the processing capacity with signal.Above-mentioned processing module 11 can
To be general processor.For example, the processor can be central processing unit (Central Processing Unit, CPU), figure
Shape processor (Graphics Processing Unit, GPU), network processing unit (Network Processor, NP) etc.；Also
It can be digital signal processor (Digital Signal Processing, DSP), specific integrated circuit (Application
Specific Integrated Circuit, ASIC), field programmable gate array (Field-Programmable Gate
Array, FPGA) either other programmable logic device, discrete gate or transistor logic, discrete hardware components, it can be with
Realize or execute disclosed each method, step and the logic diagram in the embodiment of the present application.
Communication module 13 is used to establish the communication connection of server 10 Yu terminal device 20 by network, and is received by network
Send out data.
Memory module 12 may be, but not limited to, random access memory, read-only memory, programmable read only memory,
Erasable Programmable Read Only Memory EPROM, electrically erasable programmable read-only memory etc..In the present embodiment, memory module 12 can be with
For storing infrared image, deep learning model.Certainly, memory module 12 can be also used for storage program, and processing module 11 exists
It receives after executing instruction, executes the program.
Image processing apparatus 100 includes that at least one can be stored in storage mould in the form of software or firmware (firmware)
In block 12 or the software function mould that is solidificated in 20 operating system of server 10 or terminal device (Operating System, OS)
Block.Processing module 11 is used to execute the executable module stored in memory module 12, such as included by image processing apparatus 100
Software function module and computer program etc..
It is understood that structure shown in Fig. 2 is only a kind of structural schematic diagram of server 10, server 10 can be with
Including than more components shown in Fig. 2.Each component shown in Fig. 2 can be realized using hardware, software, or its combination.
Referring to figure 3., the embodiment of the present application provides a kind of image processing method, can be applied to above-mentioned server 10,
Each step that method is executed by server 10 or terminal device 20, can improve low to power equipment infrared image recognition accuracy
The problem of.
In the present embodiment, image processing method may comprise steps of:
Step S210, obtains image to be processed, and the image to be processed includes the infrared image for acquiring power equipment and obtaining；
Step S220, by image to be processed described in the deep learning model extraction after completion training in the first default convolution
First characteristics of image of layer, and the image to be processed is extracted in the second characteristics of image of the second default convolutional layer；
Step S230, according to the first image feature and second characteristics of image to the first image feature, institute
It states the second characteristics of image and carries out Fusion Features, and feature extraction is carried out to the characteristics of image obtained after fusion, obtain target signature；
Step S240 knows the image to be processed according to the target signature by the deep learning model
Not, the recognition result that the type of power equipment is indicated in the image to be processed is obtained.
The infrared image edge of power equipment is relatively fuzzyyer, and characteristics of image does not protrude, therefore, in existing mode, in benefit
When identifying the type of power equipment with infrared image, there is a problem of that the accuracy of identification is low.In the above-mentioned implementation of the application
In mode, method is by extraction infrared image in the first characteristics of image of the first default convolutional layer, in the second default convolutional layer
Then second characteristics of image carries out Fusion Features, and carries out feature extraction using fused characteristics of image, recycling is extracted
Target signature image to be processed identified.Based on this, compared to the characteristics of image for directly utilizing image to be processed, mesh
Characteristics of image in mark feature is enhanced, so that facilitating when using the type of target signature identification power equipment
Utilize the accuracy improved with enhanced target signature to power equipment type identification.
Each step in Fig. 3 will be described in detail below:
Step S210, obtains image to be processed, and the image to be processed includes the infrared image for acquiring power equipment and obtaining.
In the present embodiment, when obtaining image to be processed, if method is executed by terminal device 20, the terminal device 20
Power equipment can be shot by self-contained infrared camera obtain the infrared image of power equipment；Alternatively, can by other
To shoot the infrared image of the photographic equipment shooting power equipment of infrared image, photographic equipment can be incited somebody to action after obtaining infrared image
Infrared image is sent to terminal device 20, or by infrared image storage to storage medium (including but not limited to USB flash disk, hard disk, deposit
Card storage etc.), the infrared image of power equipment is then got from the infrared image in storage medium reading by terminal device 20.
Wherein, infrared image accessed by terminal device 20 can be used as acquired image to be processed.
If method is executed by server 10, the mode for obtaining image to be processed can be with are as follows: firstly, terminal device 20 can lead to
It crosses above-mentioned mode and gets infrared image；Then, the infrared image for the power equipment that will acquire by terminal device 20 is sent
To server 10；Finally, server 10 is after receiving the infrared image of power equipment of the transmission of terminal device 20, it can should
Infrared image is as image to be processed.
Understandably, image to be processed for the obtained thermal infrared images of shooting power equipment, power equipment can include but
It is not limited to transformer, breaker, motor (including motor, generator) etc..
Step S220, by image to be processed described in the deep learning model extraction after completion training in the first default convolution
First characteristics of image of layer, and the image to be processed is extracted in the second characteristics of image of the second default convolutional layer.
Understandably, before identifying to image to be processed, itself be can store for server 10 or terminal device 20
There is the deep learning model after completing training, which can be used for carrying out feature extraction to image, and to figure
Object as in carries out type identification.
When extracting characteristics of image, deep learning model can extract the shallow-layer characteristics of image and deep layer figure of image to be processed
As feature.For example, the first default convolutional layer is shallow-layer convolutional layer, the second default convolutional layer is deep layer convolutional layer.Wherein, deep layer is rolled up
The number of plies of lamination is bigger than the number of plies of shallow-layer convolutional layer.Certainly, the first default convolutional layer may be deep layer convolutional layer, if first is pre-
If convolutional layer may be deep layer convolutional layer, then the second default convolutional layer is shallow-layer convolutional layer.
In the present embodiment, shallow-layer convolutional layer can be first any one layer into layer 5 convolutional layer, deep layer convolution
Layer can be any one layer in the 6th to the tenth layer of convolutional layer.For example, shallow-layer convolutional layer is third layer convolutional layer, deep layer convolution
Layer is the 8th layer of convolutional layer.Certainly, shallow-layer convolutional layer can also be the convolutional layer of other numbers of plies, as long as meeting deep layer convolutional layer
The number of plies is bigger than the number of plies of shallow-layer convolutional layer, is not especially limited here to the number of plies of deep layer convolutional layer, shallow-layer convolutional layer.Together
Sample, the first default convolutional layer, the second default convolutional layer can be configured according to the actual situation, be not especially limited here.
Step S220 may include: by the deep learning model, based on the corresponding volume of the described first default convolutional layer
Product parameter carries out convolution algorithm to the image to be processed and obtains the first image feature；By the deep learning model,
Convolution algorithm is carried out to the image to be processed based on the described second default convolutional layer corresponding deconvolution parameter and obtains described second
Characteristics of image.
In the present embodiment, different convolutional layers can be previously provided with deconvolution parameter corresponding with the convolutional layer, convolution
Parameter includes the parameter in the size and convolution kernel of convolution kernel.Wherein, the size of convolution kernel and parameters in convolution kernel can
To be configured according to the actual situation.It, can be based on each in image to be processed when carrying out convolution algorithm to image to be processed
The rgb value (rgb value refers to the value of three color of RGB, the i.e. value of Red Green Blue) of a pixel constitutes matrix, then to this
Matrix, convolution kernel carry out convolution algorithm, and the matrix obtained after operation can be used as the characteristics of image of first layer convolution.Certainly, it transports
The matrix obtained after calculation is also based on convolution kernel and carries out convolution algorithm again, obtains new matrix, which can be used as
Two layers of convolved image feature.Understandably, the characteristics of image being calculated can be used as the input of next layer of convolutional calculation, pass through
Convolution layer by layer can obtain the first characteristics of image, the second characteristics of image.
In the above-described embodiment, convolution is carried out to characteristics of image by deconvolution parameter, respective roll product can be obtained
The characteristics of image of layer, convenient for carrying out Fusion Features using characteristics of image.
In the present embodiment, if deep learning model is not completed to train, before step S230, method can also include
The step of deep learning model is trained.For example, method can also include: acquisition training image before step S230
Collection, the training image collection includes multiple training images, and each training image includes power equipment and sets with the electric power
The corresponding label of standby type；According to the training image collection, training predetermined deep learning model obtains completing the institute after training
State deep learning model.
In the present embodiment, when being trained to deep learning model, training image collection may include positive training sample
This collection and reverse train sample set；Wherein, the training image that positive training sample is concentrated is the power equipment for including target type
Infrared image, the training image in reverse train sample set be do not include target type power equipment infrared image.Its
In, the power equipment of target type can be selected according to the actual situation.For example, target type can refer to that power equipment is disconnected
This type of road device.
In the present embodiment, deep learning model can be FCNs (Fully Convolutional
NetworksforSemantic Segmentation, full convolutional network image, semantic segmentation) model.In model training, benefit
FCNs model is trained with positive training sample set and reverse train sample set, the FCNs mould after being trained
Type.
In the above-described embodiment, by being trained to deep learning model, it can be improved deep learning model to red
The accuracy that the type of power equipment is identified in outer image.
Step S230, according to the first image feature and second characteristics of image to the first image feature, institute
It states the second characteristics of image and carries out Fusion Features, and feature extraction is carried out to the characteristics of image obtained after fusion, obtain target signature.
In the present embodiment, the image after deep learning model carries out convolution to image to be processed, after available convolution
Feature, if the characteristics of image for obtaining different convolutional layers merges, just enables to merge for the same image to be processed
Characteristics of image afterwards is enhanced, to be conducive to the identification of power equipment, improves the accuracy of identification.Wherein, characteristics of image
It can be the rgb value of each pixel after being interpreted as convolution.
In the present embodiment, it (can be regarded as feature to return by being merged deep layer characteristics of image with shallow-layer characteristics of image
Infeed mechanism), be conducive to extract more abstract characteristics of image.When FCNs model carries out semantic segmentation to image, semantic segmentation
The ability of accuracy also available raising, the type of model identification power equipment can also enhance.
In the present embodiment, step S230 may include: based on the first image feature, second characteristics of image
Size adjusts the size of the first image feature and/or second characteristics of image, so that the first image feature and institute
The size for stating the second characteristics of image is identical；The first image feature, second characteristics of image are overlaped, merged
Characteristics of image afterwards；By fused characteristics of image described in the deep learning model extraction, the target signature is obtained.
In this example, it is assumed that the first characteristics of image is the corresponding image of shallow-layer characteristics of image, it can be expressed as F1, greatly
It is small are as follows: a*b*c, a refer to the length of image, and b refers to that the width of image, c refer to that the port number of image, port number can be understood as height,
A, b, c are the integer greater than 0；Second characteristics of image is the corresponding image of deep layer characteristics of image, can be expressed as F2, size are as follows:
M*n*k, m refer to the length of image, and n refers to that the width of image, k refer to that the port number of image, m, n, k are the integer greater than 0, ordinary circumstance
Lower a>m, b>n, c<k.When merging to image, processing can be amplified to the second characteristics of image, so that m=a, n=
B obtains characteristic image F3, and characteristic image F3 is superimposed with the first characteristics of image F1, obtains characteristic image F4, size are as follows: a*b*
(k+c), characteristic image F4 is fused characteristics of image.At this point it is possible to (for example use 1*1 with the convolution kernel of specified size
The convolution kernel of the convolution kernel of size or other sizes), convolution is carried out to characteristic image F4, i.e., so that image F1 and image F3 not
Information with channel is mutually blended, and obtains the characteristic image F5 of characteristics of image enhancing, characteristic image F5 is just the first image
Feature and the second characteristics of image carry out the target signature extracted after Fusion Features.
It wherein, can be by linear interpolation algorithm to the second image spy when amplifying processing to the second characteristics of image
Sign amplifies, so that the size of the second characteristics of image is identical as the size of the first characteristics of image.
Certainly, in other embodiments, it is also possible to reduce the first characteristics of image, so that the first characteristics of image
Size it is identical as the size of the second characteristics of image.Or the size of the first characteristics of image, second characteristics of image are adjusted simultaneously
Size, as long as enabling to the first characteristics of image adjusted identical as the size of the second characteristics of image.Here to adjustment
The mode of picture size is not especially limited.
In the above-described embodiment, it is adjusted by the size to the first characteristics of image, the second characteristics of image, so that adjusting
Size after section is identical, is easy to implement Fusion Features to obtain fused characteristics of image.Based on this, fused characteristics of image
Feature can be enhanced, the feature for extracting the target signature that fused characteristics of image obtains can also be enhanced, thus
Help to improve the accuracy of power equipment type in infrared image.
Step S240 knows the image to be processed according to the target signature by the deep learning model
Not, the recognition result that the type of power equipment is indicated in the image to be processed is obtained.
In the present embodiment, server 10 or electronic equipment identify images to be recognized using deep learning model
When, deep learning model can identify image to be processed based on target signature.In identification process, it can will extract
To target signature and the obtained characteristics of image of power equipment infrared image of deep learning model training target type compared
To analysis, to obtain in images to be recognized with the presence or absence of the recognition result of the power equipment of target type.For example, figure to be identified
The target signature of picture is identical as the characteristics of image that the power equipment infrared image of training objective type obtains or similarity is greater than finger
Determine threshold value (can be configured according to the actual situation, such as can be 95%, 99% etc.), then it is assumed that exist in images to be recognized
The power equipment of specified type.
As an alternative embodiment, indicating that there are specified types in the images to be recognized in the recognition result
Power equipment when, the method can also include: the power equipment that the specified type is determined in the image to be processed
Image-region in the image to be processed；According to the infrared image of image-region in the image to be processed, determine described in
The temperature data of the power equipment of specified type.
In the present embodiment, the color characteristic for indicating objects in images temperature can be carried in image to be processed.True
There are after the power equipment of specified type (for example being breaker, transformer etc.) in fixed image to be processed, server 10 can
The characteristics of image of power equipment based on specified type determines the image outline of the power equipment from image to be processed.In
It has been determined that power equipment, can be based on the face of the infrared image in image outline region after the image outline in image to be processed
Color characteristic, to determine the temperature of specified type.For example, different colors and temperature data (i.e. temperature value) are closed in advance
Join, so that the different colours in the infrared image after imaging indicate different temperature values.
In the above-described embodiment, the image-region by the power equipment using specified type in image to be processed,
Determine the temperature of the power equipment, help to improve determined by temperature accuracy, avoid by be not specified type electricity
The image-region of power equipment is as identification region, to determine the temperature of the power equipment, so that the temperature and electric power that determine
Equipment does not correspond to, and leads to the power equipment temperature inaccuracy calculated.
As a kind of optional embodiment.The method can also include: to be greater than or equal in advance in the temperature data
If when threshold value, the terminal device 20 of Xiang Zhiding issues prompt information.
In the present embodiment, preset threshold can be configured according to the actual conditions of different types of power equipment.Example
Such as, preset threshold can exist for breaker is burned corresponding temperature when risk.Specified terminal device 20 can be regarded as pre-
The terminal device 20 first communicated to connect with server 10.The terminal device 20 can be smart phone or other equipment.
When temperature data is greater than or equal to preset threshold, also means that power equipment exists and be burned because of high temperature
Risk.If method is executed by server 10, which can send prompt information to terminal device 20, so that administrator
Member's discovery in time, in order to be handled (such as cooling processing or other processing) in time.If method is held by terminal device 20
Row, the terminal device 20 can issue prompt information with itself, or will be prompted to information and be sent to other equipment (usually to manage
The equipment that personnel hold, such as smart phone) so that administrative staff can have found high temperature wind present in power equipment in time
Danger.
Referring to figure 4., the embodiment of the present application also provides a kind of image processing apparatus 100, can be used for executing or realizing
Each step for the image processing method stated.The image processing apparatus 100 may include image acquisition unit 110, feature extraction list
Member 120, Fusion Features unit 130 and recognition unit 140.
Image acquisition unit 110, for obtaining image to be processed, the image to be processed includes that acquisition power equipment obtains
Infrared image.
Feature extraction unit 120, for being existed by image to be processed described in the deep learning model extraction after completion training
First characteristics of image of the first default convolutional layer, and the image to be processed is extracted in the second image of the second default convolutional layer
Feature.
Optionally, feature extraction unit 120 can be also used for: pre- based on described first by the deep learning model
If the corresponding deconvolution parameter of convolutional layer carries out convolution algorithm to the image to be processed and obtains the first image feature；Pass through institute
Deep learning model is stated, convolution fortune is carried out to the image to be processed based on the corresponding deconvolution parameter of the described second default convolutional layer
Calculation obtains second characteristics of image.
Fusion Features unit 130, according to the first image feature and second characteristics of image to the first image
Feature, second characteristics of image carry out Fusion Features, and carry out feature extraction to the characteristics of image obtained after fusion, obtain mesh
Mark feature.
Optionally, Fusion Features unit 130 can also be used in: be based on the first image feature, second characteristics of image
Size, adjust the size of the first image feature and/or second characteristics of image so that the first image feature with
The size of second characteristics of image is identical；The first image feature, second characteristics of image are overlaped, melted
Characteristics of image after conjunction；By fused characteristics of image described in the deep learning model extraction, the target signature is obtained.
Recognition unit 140 is used for through the deep learning model according to the target signature to the image to be processed
It is identified, obtains the recognition result for indicating the type of power equipment in the image to be processed.
Optionally, image processing apparatus 100 can also include temperature determining unit, indicate image to be processed in recognition result
It is middle there are when the power equipment of specified type, temperature determining unit is used for: the specified class is determined in the image to be processed
Image-region of the power equipment of type in the image to be processed；According to the infrared figure of image-region in the image to be processed
Picture determines the temperature data of the power equipment of the specified type.
Optionally, temperature determining unit is also used to: when the temperature data is greater than or equal to preset threshold, Xiang Zhiding's
Terminal device 20 issues prompt information.
Optionally, image processing apparatus 100 can also include training unit.In feature extraction unit 120 by completing instruction
First characteristics of image of convolutional layer of the image to be processed described in the deep learning model extraction after white silk in the first default number of plies it
Before, image acquisition unit 110 is also used to obtain training image collection, and the training image collection includes multiple training images, Mei Gesuo
Stating training image includes power equipment and label corresponding with the type of the power equipment.Training unit can be used for according to institute
Training image collection is stated, training predetermined deep learning model obtains completing the deep learning model after training.
It should be noted that it is apparent to those skilled in the art that, for convenience and simplicity of description, on
The server 10 of description, the specific work process of image processing apparatus 100 are stated, it can be corresponding with reference to each step in preceding method
Process no longer excessively repeats herein.
The embodiment of the present application also provides a kind of computer readable storage medium.Computer journey is stored in readable storage medium storing program for executing
Sequence, when computer program is run on computers, so that computer executes such as above-mentioned image processing method as described in the examples
Method.
Through the above description of the embodiments, those skilled in the art can be understood that the application can lead to
Hardware realization is crossed, the mode of necessary general hardware platform can also be added to realize by software, based on this understanding, this Shen
Technical solution please can be embodied in the form of software products, which can store in a non-volatile memories
In medium (can be CD-ROM, USB flash disk, mobile hard disk etc.), including some instructions are used so that a computer equipment (can be
Personal computer, server or network equipment etc.) execute method described in each implement scene of the application.
In conclusion the application provides a kind of image processing method, device, server and readable storage medium storing program for executing.This method
It include: to obtain image to be processed, image to be processed includes the infrared image for acquiring power equipment and obtaining；After completing training
Deep learning model extraction image to be processed and extracts image to be processed and exists in the first characteristics of image of the first default convolutional layer
Second characteristics of image of the second default convolutional layer；According to the first characteristics of image and the second characteristics of image to the first characteristics of image,
Two characteristics of image carry out Fusion Features, and carry out feature extraction to the characteristics of image obtained after fusion, obtain target signature；Pass through
Deep learning model identifies image to be processed according to target signature, obtains the class that power equipment is indicated in image to be processed
The recognition result of type.In the present solution, being then based on fusion by merging to the first characteristics of image, the second characteristics of image
Obtained target signature is identified, the feature for enhancing identified image is facilitated, to improve the standard identified to power equipment
Exactness.
In embodiment provided herein, it should be understood that disclosed devices, systems, and methods can also lead to
Other modes are crossed to realize.Devices, systems, and methods embodiment described above is only schematical, for example, in attached drawing
Flow chart and block diagram show that the system of multiple embodiments according to the application, the possibility of method and computer program product are real
Existing architecture, function and operation.In this regard, each box in flowchart or block diagram can represent module, a journey
A part of sequence section or code, a part of the module, section or code include one or more for realizing defined
The executable instruction of logic function.It should also be noted that in some implementations as replacement, function marked in the box
It can also occur in a different order than that indicated in the drawings.For example, two continuous boxes can actually be substantially in parallel
It executes, they can also be executed in the opposite order sometimes, and this depends on the function involved.It is also noted that block diagram and/
Or the combination of each box in flow chart and the box in block diagram and or flow chart, can with execute as defined in function or
The dedicated hardware based system of movement is realized, or can be realized using a combination of dedicated hardware and computer instructions.
In addition, each functional module in each embodiment of the application can integrate one independent part of formation together, it can also be with
It is modules individualism, an independent part can also be integrated to form with two or more modules.
The foregoing is merely preferred embodiment of the present application, are not intended to limit this application, for the skill of this field
For art personnel, various changes and changes are possible in this application.Within the spirit and principles of this application, made any to repair
Change, equivalent replacement, improvement etc., should be included within the scope of protection of this application.
Claims (10)
1. a kind of image processing method, which is characterized in that the described method includes:
Image to be processed is obtained, the image to be processed includes the infrared image for acquiring power equipment and obtaining；
By image to be processed described in the deep learning model extraction after completion training in the first image of the first default convolutional layer
Feature, and the image to be processed is extracted in the second characteristics of image of the second default convolutional layer；
According to the first image feature and second characteristics of image to the first image feature, second characteristics of image
Fusion Features are carried out, and feature extraction is carried out to the characteristics of image obtained after fusion, obtain target signature；
The image to be processed is identified according to the target signature by the deep learning model, is obtained described wait locate
Manage the recognition result that the type of power equipment is indicated in image.
2. the method according to claim 1, wherein special according to the first image feature and second image
Sign to the first image feature, second characteristics of image carry out Fusion Features, and to the characteristics of image obtained after fusion into
Row feature extraction, obtains target signature, comprising:
Size based on the first image feature, second characteristics of image adjusts the first image feature and/or described
The size of second characteristics of image, so that the first image feature is identical as the size of second characteristics of image；
The first image feature, second characteristics of image are overlaped, fused characteristics of image is obtained；
By fused characteristics of image described in the deep learning model extraction, the target signature is obtained.
3. the method according to claim 1, wherein by completing described in the deep learning model extraction after training
Image to be processed and extracts the image to be processed in the second default convolution in the first characteristics of image of the first default convolutional layer
Second characteristics of image of layer, comprising:
By the deep learning model, based on the corresponding deconvolution parameter of the described first default convolutional layer to the image to be processed
It carries out convolution algorithm and obtains the first image feature；
By the deep learning model, based on the corresponding deconvolution parameter of the described second default convolutional layer to the image to be processed
It carries out convolution algorithm and obtains second characteristics of image.
4. the method according to claim 1, wherein indicating to deposit in the image to be processed in the recognition result
In the power equipment of specified type, the method also includes:
Image-region of the power equipment of the specified type in the image to be processed is determined in the image to be processed；
According to the infrared image of image-region in the image to be processed, the temperature number of the power equipment of the specified type is determined
According to.
5. according to the method described in claim 4, it is characterized in that, the method also includes:
When the temperature data is greater than or equal to preset threshold, the terminal device of Xiang Zhiding issues prompt information.
6. the method according to claim 1, wherein passing through the deep learning model extraction institute after completing training
Image to be processed is stated before the first characteristics of image of the convolutional layer in the first default number of plies, the method also includes:
Training image collection is obtained, the training image collection includes multiple training images, and each training image includes that electric power is set
Label standby and corresponding with the type of the power equipment；
According to the training image collection, training predetermined deep learning model obtains completing the deep learning model after training.
7. the method according to claim 1, wherein the power equipment includes breaker, transformer, in motor
At least one.
8. a kind of image processing apparatus, which is characterized in that described device includes:
Image acquisition unit, for obtaining image to be processed, the image to be processed include acquisition power equipment obtain it is infrared
Image；
Feature extraction unit, for default first by completing image to be processed described in the deep learning model extraction after training
First characteristics of image of convolutional layer, and the image to be processed is extracted in the second characteristics of image of the second default convolutional layer；
Fusion Features unit, for special to the first image according to the first image feature and second characteristics of image
Sign, second characteristics of image carry out Fusion Features, and carry out feature extraction to the characteristics of image obtained after fusion, obtain target
Feature；
Recognition unit, for being known according to the target signature to the image to be processed by the deep learning model
Not, the recognition result that the type of power equipment is indicated in the image to be processed is obtained.
9. a kind of server, which is characterized in that including memory, the processor to intercouple, storage is calculated in the memory
Machine program, when the computer program is executed by the processor, so that the server is executed as appointed in claim 1-7
Method described in one.
10. a kind of computer readable storage medium, which is characterized in that it is stored with computer program in the readable storage medium storing program for executing,
When the computer program is run on computers, so that the computer is executed such as any one of claim 1-7 institute
The method stated.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910762042.1A CN110443322A (en) | 2019-08-16 | 2019-08-16 | Image processing method, device, server and readable storage medium storing program for executing |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910762042.1A CN110443322A (en) | 2019-08-16 | 2019-08-16 | Image processing method, device, server and readable storage medium storing program for executing |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110443322A true CN110443322A (en) | 2019-11-12 |
Family
ID=68436389
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910762042.1A Pending CN110443322A (en) | 2019-08-16 | 2019-08-16 | Image processing method, device, server and readable storage medium storing program for executing |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110443322A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111062448A (en) * | 2019-12-24 | 2020-04-24 | 北京知道创宇信息技术股份有限公司 | Equipment type recognition model training method, equipment type recognition method and device |
| CN114648478A (en) * | 2022-03-29 | 2022-06-21 | 北京小米移动软件有限公司 | Image processing method, device, chip, electronic equipment and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107607207A (en) * | 2017-08-22 | 2018-01-19 | 中国科学院深圳先进技术研究院 | A kind of power equipment thermal fault detection method, system and electronic equipment |
| CN108830322A (en) * | 2018-06-15 | 2018-11-16 | 联想(北京)有限公司 | A kind of image processing method and device, equipment, storage medium |
| CN108830280A (en) * | 2018-05-14 | 2018-11-16 | 华南理工大学 | A kind of small target detecting method based on region nomination |
| US20190005657A1 (en) * | 2017-06-30 | 2019-01-03 | Baidu Online Network Technology (Beijing) Co., Ltd . | Multiple targets-tracking method and apparatus, device and storage medium |
| CN110033003A (en) * | 2019-03-01 | 2019-07-19 | 华为技术有限公司 | Image partition method and image processing apparatus |
-
2019
- 2019-08-16 CN CN201910762042.1A patent/CN110443322A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190005657A1 (en) * | 2017-06-30 | 2019-01-03 | Baidu Online Network Technology (Beijing) Co., Ltd . | Multiple targets-tracking method and apparatus, device and storage medium |
| CN107607207A (en) * | 2017-08-22 | 2018-01-19 | 中国科学院深圳先进技术研究院 | A kind of power equipment thermal fault detection method, system and electronic equipment |
| CN108830280A (en) * | 2018-05-14 | 2018-11-16 | 华南理工大学 | A kind of small target detecting method based on region nomination |
| CN108830322A (en) * | 2018-06-15 | 2018-11-16 | 联想(北京)有限公司 | A kind of image processing method and device, equipment, storage medium |
| CN110033003A (en) * | 2019-03-01 | 2019-07-19 | 华为技术有限公司 | Image partition method and image processing apparatus |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111062448A (en) * | 2019-12-24 | 2020-04-24 | 北京知道创宇信息技术股份有限公司 | Equipment type recognition model training method, equipment type recognition method and device |
| CN111062448B (en) * | 2019-12-24 | 2024-05-28 | 北京知道创宇信息技术股份有限公司 | Equipment type recognition model training method, equipment type recognition method and device |
| CN114648478A (en) * | 2022-03-29 | 2022-06-21 | 北京小米移动软件有限公司 | Image processing method, device, chip, electronic equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109508681B (en) | Method and device for generating human body key point detection model | |
| CN110705405B (en) | Target labeling method and device | |
| TWI619080B (en) | Method for calculating fingerprint overlapping region and electronic device | |
| CN109284729B (en) | Method, device and medium for acquiring face recognition model training data based on video | |
| CN109886928B (en) | Target cell marking method, device, storage medium and terminal equipment | |
| CN110493595B (en) | Camera detection method and device, storage medium and electronic device | |
| CN108734185B (en) | Image verification method and device | |
| US11455831B2 (en) | Method and apparatus for face classification | |
| CN111866392B (en) | Shooting prompting method and device, storage medium and electronic equipment | |
| CN105574910A (en) | Electronic Device and Method for Providing Filter in Electronic Device | |
| CN108564082A (en) | Image processing method, device, server and medium | |
| CN113642639B (en) | Living body detection method, living body detection device, living body detection equipment and storage medium | |
| US9560272B2 (en) | Electronic device and method for image data processing | |
| CN110826646A (en) | Robot vision testing method and device, storage medium and terminal equipment | |
| CN110443322A (en) | Image processing method, device, server and readable storage medium storing program for executing | |
| CN113255516A (en) | Living body detection method and device and electronic equipment | |
| CN106327531A (en) | Panorama video identification method and device, and video playing method and device | |
| CN113158773B (en) | Training method and training device for living body detection model | |
| CN109919164B (en) | User interface object identification method and device | |
| CN116048682A (en) | Terminal system interface layout comparison method and electronic equipment | |
| CN116993654A (en) | Camera module defect detection method, device, equipment, storage medium and product | |
| CN113065521B (en) | Object identification method, device, equipment and medium | |
| CN106402717B (en) | A kind of AR control method for playing back and intelligent desk lamp | |
| CN111124862B (en) | Intelligent device performance testing method and device and intelligent device | |
| CN108446693A (en) | Convenient labels method, system, equipment and the storage medium of target to be identified |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191112 |
|
| RJ01 | Rejection of invention patent application after publication |