CN109359542A - The determination method and terminal device of vehicle damage rank neural network based - Google Patents
The determination method and terminal device of vehicle damage rank neural network based Download PDFInfo
- Publication number
- CN109359542A CN109359542A CN201811086124.0A CN201811086124A CN109359542A CN 109359542 A CN109359542 A CN 109359542A CN 201811086124 A CN201811086124 A CN 201811086124A CN 109359542 A CN109359542 A CN 109359542A
- Authority
- CN
- China
- Prior art keywords
- matrix
- preset
- vehicle damage
- damage
- parameter
- 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
- 230000006378 damage Effects 0.000 title claims abstract description 190
- 238000000034 method Methods 0.000 title claims abstract description 33
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 22
- 239000011159 matrix material Substances 0.000 claims abstract description 179
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 53
- 230000006870 function Effects 0.000 claims description 20
- 208000027418 Wounds and injury Diseases 0.000 claims description 15
- 208000014674 injury Diseases 0.000 claims description 15
- 238000004590 computer program Methods 0.000 claims description 13
- 238000011478 gradient descent method Methods 0.000 claims description 9
- 241000208340 Araliaceae Species 0.000 claims description 3
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 claims description 3
- 235000003140 Panax quinquefolius Nutrition 0.000 claims description 3
- 235000008434 ginseng Nutrition 0.000 claims description 3
- 238000003475 lamination Methods 0.000 claims 1
- 230000001537 neural effect Effects 0.000 claims 1
- 230000032258 transport Effects 0.000 claims 1
- 238000013473 artificial intelligence Methods 0.000 abstract description 2
- 230000008676 import Effects 0.000 description 6
- 238000010586 diagram Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 2
- 239000002184 metal Substances 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
- G06V20/58—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
- G06V20/584—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads of vehicle lights or traffic lights
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The present invention is suitable for field of artificial intelligence, provide the determination method and terminal device of a kind of vehicle damage rank neural network based, by obtaining vehicle damage image, and according to the position coordinates of each pixel and the corresponding relationship of rgb value in the vehicle damage image, the corresponding damage data matrix of the vehicle damage image is generated;The damage data matrix is imported into preset convolutional neural networks, obtains the eigenmatrix of the vehicle damage image;The eigenmatrix is imported in preset softmax classifier, the corresponding probability matrix of the eigenmatrix is calculated, the value of each element in the probability matrix represents the probability that the vehicle damage image belongs to the corresponding damage rank of the element;By the corresponding loss level of the maximum element of probability matrix intermediate value, exports and save time and human cost for the corresponding damage rank of the vehicle damage image to improve the accuracy and the degree of automation of setting loss.
Description
Technical field
The invention belongs to field of artificial intelligence more particularly to a kind of vehicle damage ranks neural network based really
Determine method and terminal device.
Background technique
Currently, during vehicle insurance is settled a claim, business personnel is needed to carry out setting loss to vehicle according to the extent of damage of vehicle, but
The judgment criteria of each business personnel is different, and experienced degree is different, it is easy to the accuracy of car damage identification is influenced, it is particularly difficult
Be the vehicle metal extent of damage classification, the classification of the metal plate extent of damage tends to obscure.Meanwhile with car ownership
Increase, the business of vehicle insurance Claims Resolution also greatly increases, and for traditional artificial setting loss step there are the setting loss time is slow, human cost is high
Problem.
Summary of the invention
In view of this, the embodiment of the invention provides a kind of determination method of vehicle damage rank neural network based and
Terminal device, to solve the problems, such as that setting loss inaccuracy and setting loss intelligence degree of the existing technology are low.
The first aspect of the embodiment of the present invention provides a kind of determination method of vehicle damage rank neural network based,
Include:
Vehicle damage image is obtained, and according to the position coordinates and rgb value of each pixel in the vehicle damage image
Corresponding relationship, generate the corresponding damage data matrix of the vehicle damage image;
The damage data matrix is imported into preset convolutional neural networks, obtains the feature square of the vehicle damage image
Battle array;
The eigenmatrix is imported in preset softmax classifier, the corresponding probability square of the eigenmatrix is calculated
Gust, the value of each element in the probability matrix represents the vehicle damage image and belongs to the corresponding damage rank of the element
Probability;
By the corresponding loss level of the maximum element of probability matrix intermediate value, exports and corresponded to for the vehicle damage image
Damage rank.
The second aspect of the embodiment of the present invention provides a kind of terminal device, including memory and processor, described to deposit
The computer program that can be run on the processor is stored in reservoir, when the processor executes the computer program,
Realize following steps:
Vehicle damage image is obtained, and according to the position coordinates and rgb value of each pixel in the vehicle damage image
Corresponding relationship, generate the corresponding damage data matrix of the vehicle damage image;
The damage data matrix is imported into preset convolutional neural networks, obtains the feature square of the vehicle damage image
Battle array;
The eigenmatrix is imported in preset softmax classifier, the corresponding probability square of the eigenmatrix is calculated
Gust, the value of each element in the probability matrix represents the vehicle damage image and belongs to the corresponding damage rank of the element
Probability;
By the corresponding loss level of the maximum element of probability matrix intermediate value, exports and corresponded to for the vehicle damage image
Damage rank.
The third aspect of the embodiment of the present invention provides a kind of computer readable storage medium, the computer-readable storage
Media storage has computer program, and the computer program realizes following steps when being executed by processor:
Vehicle damage image is obtained, and according to the position coordinates and rgb value of each pixel in the vehicle damage image
Corresponding relationship, generate the corresponding damage data matrix of the vehicle damage image;The damage data matrix is imported default
Convolutional neural networks, obtain the eigenmatrix of the vehicle damage image;The eigenmatrix is imported into preset softmax
In classifier, the corresponding probability matrix of the eigenmatrix is calculated, described in the value of each element in the probability matrix represents
Vehicle damage image belongs to the probability of the corresponding damage rank of the element;The maximum element of probability matrix intermediate value is corresponding
Loss level exports as the corresponding damage rank of the vehicle damage image.
Optionally, the preset convolutional neural networks include the convolutional layer of the first preset quantity 3 × 3, each convolution
Layer is according to it in the corresponding convolutional layer number of the sequence of the convolutional neural networks from front to back;
It is described that the damage data matrix is imported into preset convolutional neural networks, obtain the spy of the vehicle damage image
Levy matrix, comprising:
The damage data matrix is imported into preset convolutional neural networks, and from the largest number of convolution of the convolutional layer
Layer starts, and numbers at interval of the convolutional layer of the second preset quantity, the data of convolutional layer output is extracted, as selected data;
The selected data of third preset quantity are subjected to global average pond, generate the pond vector of the third preset quantity;To institute
The pond vector for stating third preset quantity is spliced, and total pond vector is generated;Total pond vector is inputted into the convolution
The full articulamentum of neural network exports the eigenmatrix of the vehicle damage image.
It is optionally, described to calculate the corresponding probability matrix of the eigenmatrix, comprising:
Pass through formula:Calculate the corresponding probability matrix of the eigenmatrix;The σ (j) is
The corresponding probability value of j-th of element in the probability matrix;zjFor the corresponding parameter of j-th of element in preset parameter matrix;
The M is the number of element in the parameter matrix, the xiFor i-th of element in the eigenmatrix, the e is that nature is normal
Number.
Optionally, before the acquisition original image, further includes:
Transfer the training matrix and the corresponding damage rank of each training matrix of preset quantity;
Pass through preset rules:
Determine the corresponding calibration coefficients of training matrix t, the rtIndicate the corresponding calibration coefficients of training matrix t, the x (t) indicates institute
State the corresponding damage rank of training matrix t;
Obtain preset cost function:It is described
T (z) is the corresponding cost parameter of the parameter matrix, and the M is the number of element in the parameter matrix, and the P is described
The quantity of training matrix, the xtiFor i-th of element in the eigenmatrix t, the zjIt is j-th in preset parameter matrix
The corresponding parameter of element, the s are preset regularization coefficient, and the N is the number of the element of the probability matrix;
The minimum value of the cost parameter, output cost parameter corresponding institute when being minimized are solved by gradient descent method
Each element in parameter matrix is stated, to generate the parameter matrix.
Optionally, further includes:
Obtain multiple preset training-related injury data matrixes and the corresponding eigenmatrix of training-related injury data matrix;Instead
Following steps are executed again until the cross entropy loss function value of updated convolutional neural networks is less than preset loss threshold value: will
Output of the training-related injury data matrix as the convolutional neural networks, by stochastic gradient descent method to the convolution mind
Each layer parameter in full articulamentum through network is updated, and calculates the cross entropy loss function of updated convolutional neural networks
Value.
In embodiments of the present invention, by according to the position coordinates of pixel each in vehicle damage image and rgb value
Corresponding relationship, generates the corresponding damage data matrix of vehicle damage image, and damage data matrix is imported preset convolutional Neural
Network obtains the eigenmatrix of vehicle damage image, later imports eigenmatrix in preset softmax classifier, calculates
The corresponding probability matrix of eigenmatrix, and using the corresponding loss level of the maximum element of probability matrix intermediate value as damage figure of enjoying the cool
As corresponding damage rank saves time and human cost to improve the accuracy and the degree of automation of setting loss.
Detailed description of the invention
It to describe the technical solutions in the embodiments of the present invention more clearly, below will be to embodiment or description of the prior art
Needed in attached drawing be briefly described, it should be apparent that, the accompanying drawings in the following description is only of the invention some
Embodiment for those of ordinary skill in the art without any creative labor, can also be according to these
Attached drawing obtains other attached drawings.
Fig. 1 is the implementation process of the determination method of vehicle damage rank neural network based provided in an embodiment of the present invention
Figure;
Fig. 2 is the specific of the determination method S102 of vehicle damage rank neural network based provided in an embodiment of the present invention
Implementation flow chart;
Fig. 3 is the product process figure of parameter matrix provided in an embodiment of the present invention;
Fig. 4 is the structural frames of the determining device of vehicle damage rank neural network based provided in an embodiment of the present invention
Figure;
Fig. 5 is the schematic diagram of terminal device provided in an embodiment of the present invention.
Specific embodiment
In being described below, for illustration and not for limitation, the tool of such as particular system structure, technology etc is proposed
Body details, to understand thoroughly the embodiment of the present invention.However, it will be clear to one skilled in the art that there is no these specific
The present invention also may be implemented in the other embodiments of details.In other situations, it omits to well-known system, device, electricity
The detailed description of road and method, in case unnecessary details interferes description of the invention.
In order to illustrate technical solutions according to the invention, the following is a description of specific embodiments.
Fig. 1 shows the realization of the determination method of vehicle damage rank neural network based provided in an embodiment of the present invention
Process, this method process include step S101 to S104.The specific implementation principle of each step is as follows.
S101: obtain vehicle damage image, and according to the position coordinates of each pixel in the vehicle damage image with
The corresponding relationship of rgb value generates the corresponding damage data matrix of the vehicle damage image.
It in embodiments of the present invention, can be by camera installation by vehicle when user needs to carry out setting loss Claims Resolution to vehicle
The picture shooting of injury region gets off.Method involved in the embodiment of the present invention obtain user input vehicle damage photo after,
The damage data matrix for subsequent automated calculating can be converted into.
Specifically, the corresponding one group of rgb value of each pixel of a vehicle damage image, the embodiment of the present invention pass through each
The position coordinates of pixel and the corresponding relationship of rgb value, it is corresponding as the pixel in matrix by the corresponding pixel value of pixel
The value of target element is sat, so as to which two-dimensional vehicle damage image is converted to a damage data matrix.
Optionally, due to the corresponding one group of rgb value of each pixel of a vehicle damage image, so construction damage data
The specific steps of matrix can be with are as follows: step 1: constructs 3 picture element matrixs, i.e. R figure layer pair respectively based on three figure layers of medical image
Answer a matrix, the corresponding matrix of G figure layer, the corresponding matrix of B figure layer, in each picture element matrix the value of element be 0~
255.Step 2: the corresponding matrix of multiple figure layers is merged, and the mode of fusion can be with are as follows: the column retained in three matrixes are compiled
Number with vehicle loss image abscissa correspond, the row of the picture element matrix of R figure layer is expanded, fills two between every row
Line blank row, and each row of other two picture element matrix is successively imported into each blank line of expansion according to the order of row number,
To constitute the matrix of 3M*N, wherein M is the line number of the pixel of vehicle damage image, and N is the pixel of vehicle damage image
Columns, using the matrix after expansion as damage data matrix.
The damage data matrix is imported preset convolutional neural networks, obtains the vehicle damage image by S102
Eigenmatrix.
In embodiments of the present invention, a convolutional neural networks have been trained in advance, are calculating damage data square every time
After battle array, only damage data matrix need to be imported preset convolutional neural networks, the feature square of vehicle damage image can be calculated
Battle array, without all convolutional neural networks of re -training before each calculate.
In embodiments of the present invention, convolutional neural networks include convolutional layer, pond layer and full articulamentum, wherein preset
Convolutional neural networks include the convolutional layer of the first preset quantity 3 × 3, and each convolutional layer is according to it in the convolutional neural networks
The corresponding convolutional layer number of sequence from front to back.
As an embodiment of the present invention, as shown in Fig. 2, above-mentioned S102 includes:
S1021: the damage data matrix is imported into preset convolutional neural networks, and numbers maximum from the convolutional layer
Convolutional layer start, numbered at interval of the convolutional layer of the second preset quantity, extract the data of convolutional layer output, as selected
Data.
Optionally, which may include 11 convolutional layers, and the layers 1 and 2 of model is 3x3
Convolutional layer, the step-length of convolution is 1, and feature port number is 32;The 3rd layer of model is the convolutional layer of 3x3, and the step-length of convolution is 2,
Feature port number is 64;The the 4th and the 5th layer of model is the convolutional layer of 3x3, and the step-length of convolution is 1, and feature port number is 64;Mould
The 6th layer of type is the convolutional layer of 3x3, and the step-length of convolution is 2, and feature port number is 128;The the 7th and the 8th layer of model is 3x3's
Convolutional layer, the step-length of convolution are 1, and feature port number is 128;The 9th layer of model is the convolutional layer of 3x3, and the step-length of convolution is 2,
Feature port number is 256;The 10th of model and 11th layer be 3x3 convolutional layer, the step-length of convolution is 1, and feature port number is
256。
Optionally, the data that a convolutional layer exports are extracted every 3 convolutional layers forward from 11th layer convolutional layer, as quilt
Data are selected, therefore extract 11th layer, the 8th layer, the 5th layer and the 2nd layer of data exported, as selected data.
The selected data of third preset quantity are carried out global average pond, generate the third preset quantity by S1022
Pond vector.
Illustratively, it is assumed that third present count measures 3, then carries out the overall situation respectively to 11th layer, the 8th layer, the 5th layer of data
The operation in average pond obtains the pond vector of the pond vector of 64 dimensions, the pond vector of 128 dimensions and 256 dimensions.
S1023 splices the pond vector of the third preset quantity, generates total pond vector.
Illustratively, the pond vector of the pond vector of 64 above-mentioned dimensions, the pond vector of 128 dimensions and 256 dimensions is spelled
It is connected in total pond vector of one 448 dimension.
Total pond vector is inputted the full articulamentum of the convolutional neural networks, exports the vehicle damage by S1024
The eigenmatrix of image.
It is to be appreciated that the full articulamentum by preset convolutional neural networks can be by total pond of 448 dimensions of input
Vector is converted into an eigenmatrix, since the Computing Principle of the full articulamentum of convolutional neural networks is the prior art, so
This is without being described in detail and specifically limiting.
Optionally, in the embodiment of the present invention, the training process of specific convolutional neural networks includes:
The first step obtains multiple preset training-related injury data matrixes and the corresponding feature of training-related injury data matrix
Matrix.
Second step executes following steps repeatedly until the cross entropy loss function value of updated convolutional neural networks is less than
Preset loss threshold value: using the training-related injury data matrix as the output of the convolutional neural networks, by it is existing with
Machine gradient descent method is updated each layer parameter in the full articulamentum of the convolutional neural networks, calculates updated convolution
The cross entropy loss function value of neural network.
S103 imports the eigenmatrix in preset softmax classifier, and it is corresponding general to calculate the eigenmatrix
Rate matrix, the value of each element in the probability matrix represent the vehicle damage image and belong to the corresponding damage grade of the element
Other probability.
Optionally, the preset softmax classifier passes through formula:Calculate the feature
The corresponding probability matrix of matrix;The σ (j) is the corresponding probability value of j-th of element in the probability matrix;zjFor preset ginseng
The corresponding parameter of j-th of element in matrix number;The M is the number of element in the parameter matrix, the xiFor the feature
I-th of element in matrix, the e are natural constant.
It is to be appreciated that the parameter matrix is the data precomputed according to existing training data, it is specific to join
The product process of matrix number is as shown in Figure 3:
S301 transfers the training matrix and the corresponding damage rank of each training matrix of preset quantity.
S302 determines the corresponding calibration coefficients of training matrix by preset rules.
Optionally, the preset rules include:
S303 obtains preset cost function, and the minimum value of the cost parameter is solved by gradient descent method, exports generation
Each element of valence parameter when being minimized in the corresponding parameter matrix, to generate the parameter matrix.
Optionally, the preset cost function includes:
The T (z) is the parameter matrix
Corresponding cost parameter, the M are the number of element in the parameter matrix, and the P is the quantity of the training matrix, described
xtiFor i-th of element in the eigenmatrix t, the zjIt is described for the corresponding parameter of j-th of element in preset parameter matrix
S is preset regularization coefficient, and the N is the number of the element of the probability matrix.
It is to be appreciated that the value of each element in parameter matrix is most appropriate when cost parameter T (z) is minimized
, to train optimal parameter matrix.
S104 exports the corresponding loss level of the maximum element of probability matrix intermediate value for the vehicle damage figure
As corresponding damage rank.
It is to be appreciated that the value of each element is made according to the corresponding relationship of preset matrix element and damage rank
For the probability of corresponding damage rank, and the corresponding loss level of the maximum element of selected value is corresponding as vehicle damage image
Damage rank.
In embodiments of the present invention, by obtaining vehicle damage image, and according to each picture in the vehicle damage image
The position coordinates of vegetarian refreshments and the corresponding relationship of rgb value generate the corresponding damage data matrix of the vehicle damage image;It will be described
Damage data matrix imports preset convolutional neural networks, obtains the eigenmatrix of the vehicle damage image;By the feature
Matrix imports in preset softmax classifier, calculates the corresponding probability matrix of the eigenmatrix, in the probability matrix
The value of each element represents the probability that the vehicle damage image belongs to the corresponding damage rank of the element;By the probability matrix
The corresponding loss level of the maximum element of intermediate value exports as the corresponding damage rank of the vehicle damage image, to improve setting loss
Accuracy and the degree of automation, save time and human cost.
Corresponding to the determination method of vehicle damage neural network based rank described in foregoing embodiments, Fig. 4 is shown
The structural block diagram of the determining device of vehicle damage rank neural network based provided in an embodiment of the present invention, for the ease of saying
Bright, only parts related to embodiments of the present invention are shown.
Referring to Fig. 4, which includes:
Module 401 is obtained, for obtaining vehicle damage image, and according to each pixel in the vehicle damage image
The corresponding relationship of position coordinates and rgb value generates the corresponding damage data matrix of the vehicle damage image;
First computing module 402 obtains described for the damage data matrix to be imported preset convolutional neural networks
The eigenmatrix of vehicle damage image;
Second computing module 403, for importing the eigenmatrix in preset softmax classifier, described in calculating
The corresponding probability matrix of eigenmatrix, the value of each element in the probability matrix represent the vehicle damage image and belong to this
The probability of the corresponding damage rank of element;
Deciding grade and level module 404, for by the corresponding loss level of the maximum element of probability matrix intermediate value, it to be described for exporting
The corresponding damage rank of vehicle damage image.
Optionally, the preset convolutional neural networks include the convolutional layer of the first preset quantity 3 × 3, each convolution
Layer is according to it in the corresponding convolutional layer number of the sequence of the convolutional neural networks from front to back;
Optionally, described that the damage data matrix is imported into preset convolutional neural networks, obtain the vehicle damage
The eigenmatrix of image, comprising:
The damage data matrix is imported into preset convolutional neural networks, and from the largest number of convolution of the convolutional layer
Layer starts, and numbers at interval of the convolutional layer of the second preset quantity, the data of convolutional layer output is extracted, as selected data;
The selected data of third preset quantity are subjected to global average pond, generate the pond vector of the third preset quantity;To institute
The pond vector for stating third preset quantity is spliced, and total pond vector is generated;Total pond vector is inputted into the convolution
The full articulamentum of neural network exports the eigenmatrix of the vehicle damage image.
It is optionally, described to calculate the corresponding probability matrix of the eigenmatrix, comprising:
Pass through formula:Calculate the corresponding probability matrix of the eigenmatrix;The σ (j) is
The corresponding probability value of j-th of element in the probability matrix;zjFor the corresponding parameter of j-th of element in preset parameter matrix;
The M is the number of element in the parameter matrix, the xiFor i-th of element in the eigenmatrix, the e is that nature is normal
Number.
Optionally, before the acquisition original image, further includes:
Transfer the training matrix and the corresponding damage rank of each training matrix of preset quantity;
Pass through preset rules:Really
Determine the corresponding calibration coefficients of training matrix t, the rtThe corresponding calibration coefficients of training matrix t are indicated, described in x (t) expression
The corresponding damage rank of training matrix t;
Obtain preset cost function:The T
It (z) is the corresponding cost parameter of the parameter matrix, the M is the number of element in the parameter matrix, and the P is the instruction
Practice the quantity of matrix, the xtiFor i-th of element in the eigenmatrix t, the zjIt is j-th yuan in preset parameter matrix
The corresponding parameter of element, the s are preset regularization coefficient, and the N is the number of the element of the probability matrix;Pass through gradient
Descent method solves the minimum value of the cost parameter, and output cost parameter is each in the corresponding parameter matrix when being minimized
A element, to generate the parameter matrix.
Optionally, described device further include:
Training module, it is corresponding for obtaining multiple preset training-related injury data matrixes and training-related injury data matrix
Eigenmatrix;Following steps are executed repeatedly until the cross entropy loss function value of updated convolutional neural networks is less than default
Loss threshold value: using the training-related injury data matrix as the output of the convolutional neural networks, pass through stochastic gradient descent
Method is updated each layer parameter in the full articulamentum of the convolutional neural networks, calculates updated convolutional neural networks
Cross entropy loss function value.
In embodiments of the present invention, by obtaining vehicle damage image, and according to each picture in the vehicle damage image
The position coordinates of vegetarian refreshments and the corresponding relationship of rgb value generate the corresponding damage data matrix of the vehicle damage image;It will be described
Damage data matrix imports preset convolutional neural networks, obtains the eigenmatrix of the vehicle damage image;By the feature
Matrix imports in preset softmax classifier, calculates the corresponding probability matrix of the eigenmatrix, in the probability matrix
The value of each element represents the probability that the vehicle damage image belongs to the corresponding damage rank of the element;By the probability matrix
The corresponding loss level of the maximum element of intermediate value exports as the corresponding damage rank of the vehicle damage image, to improve setting loss
Accuracy and the degree of automation, save time and human cost.
Fig. 5 is the schematic diagram for the terminal device that one embodiment of the invention provides.As shown in figure 5, the terminal of the embodiment is set
Standby 5 include: processor 50, memory 51 and are stored in the meter that can be run in the memory 51 and on the processor 50
Calculation machine program 52, such as the determination program of vehicle damage rank neural network based.The processor 50 executes the calculating
The step in the determination embodiment of the method for above-mentioned each vehicle damage rank neural network based is realized when machine program 52, such as
Step 101 shown in FIG. 1 is to 104.Alternatively, the processor 50 realizes that above-mentioned each device is real when executing the computer program 52
Apply the function of each module/unit in example, such as the function of unit 401 to 404 shown in Fig. 4.
Illustratively, the computer program 52 can be divided into one or more module/units, it is one or
Multiple module/units are stored in the memory 51, and are executed by the processor 50, to complete the present invention.Described one
A or multiple module/units can be the series of computation machine program instruction section that can complete specific function, which is used for
Implementation procedure of the computer program 52 in the terminal device 5 is described.
The terminal device 5 can be the calculating such as desktop PC, notebook, palm PC and cloud server and set
It is standby.The terminal device may include, but be not limited only to, processor 50, memory 51.It will be understood by those skilled in the art that Fig. 5
The only example of terminal device 5 does not constitute the restriction to terminal device 5, may include than illustrating more or fewer portions
Part perhaps combines certain components or different components, such as the terminal device can also include input-output equipment, net
Network access device, bus etc..
Alleged processor 50 can be central processing unit (Central Processing Unit, CPU), can also be
Other general processors, digital signal processor (Digital Signal Processor, DSP), specific integrated circuit
(Application Specific Integrated Circuit, ASIC), ready-made programmable gate array (Field-
Programmable Gate Array, FPGA) either other programmable logic device, discrete gate or transistor logic,
Discrete hardware components etc..General processor can be microprocessor or the processor is also possible to any conventional processor
Deng.
The memory 51 can be the internal storage unit of the terminal device 5, such as the hard disk or interior of terminal device 5
It deposits.The memory 51 is also possible to the External memory equipment of the terminal device 5, such as be equipped on the terminal device 5
Plug-in type hard disk, intelligent memory card (Smart Media Card, SMC), secure digital (Secure Digital, SD) card dodge
Deposit card (Flash Card) etc..Further, the memory 51 can also both include the storage inside list of the terminal device 5
Member also includes External memory equipment.The memory 51 is for storing needed for the computer program and the terminal device
Other programs and data.The memory 51 can be also used for temporarily storing the data that has exported or will export.
It is apparent to those skilled in the art that for convenience of description and succinctly, only with above-mentioned each function
Can unit, module division progress for example, in practical application, can according to need and by above-mentioned function distribution by different
Functional unit, module are completed, i.e., the internal structure of described device is divided into different functional unit or module, more than completing
The all or part of function of description.Each functional unit in embodiment, module can integrate in one processing unit, can also
To be that each unit physically exists alone, 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 software functional units.In addition, each function list
Member, the specific name of module are also only for convenience of distinguishing each other, the protection scope being not intended to limit this application.Above system
The specific work process of middle unit, module, can refer to corresponding processes in the foregoing method embodiment, and details are not described herein.
In the above-described embodiments, it all emphasizes particularly on different fields to the description of each embodiment, is not described in detail or remembers in some embodiment
The part of load may refer to the associated description of other embodiments.
The unit as illustrated by the separation member may or may not be physically separated, aobvious as unit
The component shown may or may not be physical unit, it can and it is in one place, or may be distributed over multiple
In network unit.It can select some or all of unit therein according to the actual needs to realize the mesh of this embodiment scheme
's.
If the integrated module/unit be realized in the form of SFU software functional unit and as independent product sale or
In use, can store in a computer readable storage medium.Based on this understanding, the present invention realizes above-mentioned implementation
All or part of the process in example method, can also instruct relevant hardware to complete, the meter by computer program
Calculation machine program can be stored in a computer readable storage medium.
Embodiment described above is merely illustrative of the technical solution of the present invention, rather than its limitations;Although referring to aforementioned reality
Applying example, invention is explained in detail, those skilled in the art should understand that: it still can be to aforementioned each
Technical solution documented by embodiment is modified or equivalent replacement of some of the technical features;And these are modified
Or replacement, the spirit and scope for technical solution of various embodiments of the present invention that it does not separate the essence of the corresponding technical solution should all
It is included within protection scope of the present invention.
Claims (10)
1. a kind of determination method of vehicle damage rank neural network based characterized by comprising
Vehicle damage image is obtained, and according to the position coordinates of each pixel and pair of rgb value in the vehicle damage image
It should be related to, generate the corresponding damage data matrix of the vehicle damage image;
The damage data matrix is imported into preset convolutional neural networks, obtains the eigenmatrix of the vehicle damage image;
The eigenmatrix is imported in preset softmax classifier, the corresponding probability matrix of the eigenmatrix, institute are calculated
The value for stating each element in probability matrix represents the probability that the vehicle damage image belongs to the corresponding damage rank of the element;
The corresponding loss level of the maximum element of probability matrix intermediate value is exported for the corresponding damage of the vehicle damage image
Injury grade is other.
2. the determination method of vehicle damage rank neural network based as described in claim 1, which is characterized in that described pre-
If convolutional neural networks include the first preset quantity 3 × 3 convolutional layer, each convolutional layer is according to it in the convolutional Neural
The corresponding convolutional layer number of the sequence of network from front to back;
It is described that the damage data matrix is imported into preset convolutional neural networks, obtain the feature square of the vehicle damage image
Battle array, comprising:
The damage data matrix is imported into preset convolutional neural networks, and is opened from the largest number of convolutional layer of the convolutional layer
Begin, is numbered at interval of the convolutional layer of the second preset quantity, the data of convolutional layer output are extracted, as selected data;
The selected data of third preset quantity are subjected to global average pond, generate the pond vector of the third preset quantity;
The pond vector of the third preset quantity is spliced, total pond vector is generated;
The full articulamentum that total pond vector is inputted to the convolutional neural networks, exports the feature of the vehicle damage image
Matrix.
3. the determination method of vehicle damage rank neural network based as described in claim 1, which is characterized in that the meter
Calculate the corresponding probability matrix of the eigenmatrix, comprising:
Pass through formula:Calculate the corresponding probability matrix of the eigenmatrix;The σ (j) is described general
The corresponding probability value of j-th of element in rate matrix;zjFor the corresponding parameter of j-th of element in preset parameter matrix;The M is
The number of element in the parameter matrix, the xiFor i-th of element in the eigenmatrix, the e is natural constant.
4. the determination method of vehicle damage rank neural network based as claimed in claim 3, which is characterized in that described
Before acquisition original image, further includes:
Transfer the training matrix and the corresponding damage rank of each training matrix of preset quantity;
Pass through preset rules:Really
Determine the corresponding calibration coefficients of training matrix t, the rtThe corresponding calibration coefficients of training matrix t are indicated, described in x (t) expression
The corresponding damage rank of training matrix t;
Obtain preset cost function:The T (z)
For the corresponding cost parameter of the parameter matrix, the M is the number of element in the parameter matrix, and the P is the training
The quantity of matrix, the xtiFor i-th of element in the eigenmatrix t, the zjFor j-th of element in preset parameter matrix
Corresponding parameter, the s are preset regularization coefficient, and the N is the number of the element of the probability matrix;
The minimum value of the cost parameter, output cost parameter corresponding ginseng when being minimized are solved by gradient descent method
Each element in matrix number, to generate the parameter matrix.
5. the determination method of vehicle damage rank neural network based as described in claim 1, which is characterized in that also wrap
It includes:
Obtain multiple preset training-related injury data matrixes and the corresponding eigenmatrix of training-related injury data matrix;
Following steps are executed repeatedly until the cross entropy loss function value of updated convolutional neural networks is less than preset loss
Threshold value:
Using the training-related injury data matrix as the output of the convolutional neural networks, by stochastic gradient descent method to described
Each layer parameter in the full articulamentum of convolutional neural networks is updated, and calculates the cross entropy damage of updated convolutional neural networks
Lose functional value.
6. a kind of terminal device, including memory and processor, it is stored with and can transports on the processor in the memory
Capable computer program, which is characterized in that when the processor executes the computer program, realize following steps:
Vehicle damage image is obtained, and according to the position coordinates of each pixel and pair of rgb value in the vehicle damage image
It should be related to, generate the corresponding damage data matrix of the vehicle damage image;
The damage data matrix is imported into preset convolutional neural networks, obtains the eigenmatrix of the vehicle damage image;
The eigenmatrix is imported in preset softmax classifier, the corresponding probability matrix of the eigenmatrix, institute are calculated
The value for stating each element in probability matrix represents the probability that the vehicle damage image belongs to the corresponding damage rank of the element;
The corresponding loss level of the maximum element of probability matrix intermediate value is exported for the corresponding damage of the vehicle damage image
Injury grade is other.
7. terminal device as claimed in claim 6, which is characterized in that the preset convolutional neural networks include first default
The convolutional layer of quantity 3 × 3, each convolutional layer is according to it in the corresponding volume of the sequence of the convolutional neural networks from front to back
Lamination number;
It is described that the damage data matrix is imported into preset convolutional neural networks, obtain the feature square of the vehicle damage image
Battle array, comprising:
The damage data matrix is imported into preset convolutional neural networks, and is opened from the largest number of convolutional layer of the convolutional layer
Begin, is numbered at interval of the convolutional layer of the second preset quantity, the data of convolutional layer output are extracted, as selected data;
The selected data of third preset quantity are subjected to global average pond, generate the pond vector of the third preset quantity;
The pond vector of the third preset quantity is spliced, total pond vector is generated;
The full articulamentum that total pond vector is inputted to the convolutional neural networks, exports the feature of the vehicle damage image
Matrix.
8. terminal device as claimed in claim 6, which is characterized in that described to calculate the corresponding probability square of the eigenmatrix
Battle array, comprising:
Pass through formula:Calculate the corresponding probability matrix of the eigenmatrix;The σ (j) is described general
The corresponding probability value of j-th of element in rate matrix;zjFor the corresponding parameter of j-th of element in preset parameter matrix;The M is
The number of element in the parameter matrix, the xiFor i-th of element in the eigenmatrix, the e is natural constant.
9. terminal device as claimed in claim 8, which is characterized in that before the acquisition original image, further includes:
Transfer the training matrix and the corresponding damage rank of each training matrix of preset quantity;
Pass through preset rules:Really
Determine the corresponding calibration coefficients of training matrix t, the rtThe corresponding calibration coefficients of training matrix t are indicated, described in x (t) expression
The corresponding damage rank of training matrix t;
Obtain preset cost function:The T (z)
For the corresponding cost parameter of the parameter matrix, the M is the number of element in the parameter matrix, and the P is the training
The quantity of matrix, the xtiFor i-th of element in the eigenmatrix t, the zjFor j-th of element in preset parameter matrix
Corresponding parameter, the s are preset regularization coefficient, and the N is the number of the element of the probability matrix;
The minimum value of the cost parameter, output cost parameter corresponding ginseng when being minimized are solved by gradient descent method
Each element in matrix number, to generate the parameter matrix.
10. a kind of computer readable storage medium, the computer-readable recording medium storage has computer program, and feature exists
In when the computer program is executed by processor the step of any one of such as claim 1 to 5 of realization the method.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811086124.0A CN109359542A (en) | 2018-09-18 | 2018-09-18 | The determination method and terminal device of vehicle damage rank neural network based |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811086124.0A CN109359542A (en) | 2018-09-18 | 2018-09-18 | The determination method and terminal device of vehicle damage rank neural network based |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109359542A true CN109359542A (en) | 2019-02-19 |
Family
ID=65351160
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811086124.0A Pending CN109359542A (en) | 2018-09-18 | 2018-09-18 | The determination method and terminal device of vehicle damage rank neural network based |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109359542A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110033386A (en) * | 2019-03-07 | 2019-07-19 | 阿里巴巴集团控股有限公司 | Identification method and device, the electronic equipment of car accident |
CN110136009A (en) * | 2019-04-17 | 2019-08-16 | 阿里巴巴集团控股有限公司 | Accident vehicle Claims Resolution method and apparatus, electronic equipment |
WO2021151278A1 (en) * | 2020-05-28 | 2021-08-05 | 平安科技(深圳)有限公司 | Method and apparatus for vehicle damage assessment based on artificial intelligence, electronic device, and medium |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105930680A (en) * | 2016-05-27 | 2016-09-07 | 大连楼兰科技股份有限公司 | Damage degree model verification method and system |
CN106127747A (en) * | 2016-06-17 | 2016-11-16 | 史方 | Car surface damage classifying method and device based on degree of depth study |
CN106248801A (en) * | 2016-09-06 | 2016-12-21 | 哈尔滨工业大学 | A kind of Rail crack detection method based on many acoustie emission events probability |
CN106504248A (en) * | 2016-12-06 | 2017-03-15 | 成都通甲优博科技有限责任公司 | Vehicle damage method of discrimination based on computer vision |
CN106776817A (en) * | 2016-11-25 | 2017-05-31 | 中国平安财产保险股份有限公司 | A kind of time data processing method and system of vehicle insurance intelligence setting loss platform |
CN107092922A (en) * | 2017-03-13 | 2017-08-25 | 平安科技(深圳)有限公司 | Car damages recognition methods and server |
CN107194323A (en) * | 2017-04-28 | 2017-09-22 | 阿里巴巴集团控股有限公司 | Car damage identification image acquiring method, device, server and terminal device |
-
2018
- 2018-09-18 CN CN201811086124.0A patent/CN109359542A/en active Pending
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105930680A (en) * | 2016-05-27 | 2016-09-07 | 大连楼兰科技股份有限公司 | Damage degree model verification method and system |
CN106127747A (en) * | 2016-06-17 | 2016-11-16 | 史方 | Car surface damage classifying method and device based on degree of depth study |
CN106248801A (en) * | 2016-09-06 | 2016-12-21 | 哈尔滨工业大学 | A kind of Rail crack detection method based on many acoustie emission events probability |
CN106776817A (en) * | 2016-11-25 | 2017-05-31 | 中国平安财产保险股份有限公司 | A kind of time data processing method and system of vehicle insurance intelligence setting loss platform |
CN106504248A (en) * | 2016-12-06 | 2017-03-15 | 成都通甲优博科技有限责任公司 | Vehicle damage method of discrimination based on computer vision |
CN107092922A (en) * | 2017-03-13 | 2017-08-25 | 平安科技(深圳)有限公司 | Car damages recognition methods and server |
CN107194323A (en) * | 2017-04-28 | 2017-09-22 | 阿里巴巴集团控股有限公司 | Car damage identification image acquiring method, device, server and terminal device |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110033386A (en) * | 2019-03-07 | 2019-07-19 | 阿里巴巴集团控股有限公司 | Identification method and device, the electronic equipment of car accident |
CN110033386B (en) * | 2019-03-07 | 2020-10-02 | 阿里巴巴集团控股有限公司 | Vehicle accident identification method and device and electronic equipment |
CN110136009A (en) * | 2019-04-17 | 2019-08-16 | 阿里巴巴集团控股有限公司 | Accident vehicle Claims Resolution method and apparatus, electronic equipment |
WO2021151278A1 (en) * | 2020-05-28 | 2021-08-05 | 平安科技(深圳)有限公司 | Method and apparatus for vehicle damage assessment based on artificial intelligence, electronic device, and medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108229654B (en) | Neural network convolution operation device and method | |
US20200097806A1 (en) | Processing method and accelerating device | |
CN107358293B (en) | Neural network training method and device | |
CN109543832B (en) | Computing device and board card | |
CN109685819B (en) | Three-dimensional medical image segmentation method based on feature enhancement | |
CN106203617B (en) | A kind of acceleration processing unit and array structure based on convolutional neural networks | |
CN105512723B (en) | A kind of artificial neural networks apparatus and method for partially connected | |
CN111860398B (en) | Remote sensing image target detection method and system and terminal equipment | |
CN109902546A (en) | Face identification method, device and computer-readable medium | |
CN108009594B (en) | A kind of image-recognizing method based on change grouping convolution | |
CN109949255A (en) | Image rebuilding method and equipment | |
CN108197532A (en) | The method, apparatus and computer installation of recognition of face | |
CN107341542A (en) | Apparatus and method for performing Recognition with Recurrent Neural Network and LSTM computings | |
CN109949219B (en) | Reconstruction method, device and equipment of super-resolution image | |
CN108416437A (en) | The processing system and method for artificial neural network for multiply-add operation | |
CN106951926A (en) | The deep learning systems approach and device of a kind of mixed architecture | |
CN112529146B (en) | Neural network model training method and device | |
CN112163601B (en) | Image classification method, system, computer device and storage medium | |
CN106650928A (en) | Method and device for optimizing neural network | |
CN106780645A (en) | Dynamic MRI images method for reconstructing and device | |
CN109359542A (en) | The determination method and terminal device of vehicle damage rank neural network based | |
CN109754359A (en) | A kind of method and system that the pondization applied to convolutional neural networks is handled | |
CN108229536A (en) | Optimization method, device and the terminal device of classification prediction model | |
CN110009097A (en) | The image classification method of capsule residual error neural network, capsule residual error neural network | |
CN108320018A (en) | A kind of device and method of artificial neural network operation |
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 |