CN107808157A - A kind of method and device of detonator coding positioning and identification - Google Patents
A kind of method and device of detonator coding positioning and identification Download PDFInfo
- Publication number
- CN107808157A CN107808157A CN201711092470.5A CN201711092470A CN107808157A CN 107808157 A CN107808157 A CN 107808157A CN 201711092470 A CN201711092470 A CN 201711092470A CN 107808157 A CN107808157 A CN 107808157A
- Authority
- CN
- China
- Prior art keywords
- detonator
- image
- network model
- coding
- positioning
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
- G06V30/14—Image acquisition
- G06V30/148—Segmentation of character regions
- G06V30/153—Segmentation of character regions using recognition of characters or words
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
This application discloses a kind of positioning of detonator coding and know method for distinguishing, including：Pretreatment operation is carried out to original image set, obtains training image collection；Default convolutional neural networks are trained according to training image collection, obtains positioning network model and identifies network model；Using positioning network model and identifying that network model is positioned and identified to detonator image, the detonator coding of detonator image is obtained.Convolutional neural networks are applied to the positioning of detonator coding character with identifying in work by this method, detonator image is positioned and identified using network model and identification network model is positioned, obtain detonator coding, whole process just can carry out end-to-end identification without detonator image is cut into monocase, realize being accurately positioned and identifying for detonator coding.The application additionally provides a kind of device, equipment and the computer-readable recording medium of detonator coding positioning and identification simultaneously, has above-mentioned beneficial effect.
Description
Technical field
The application is related to character locating and identification field, more particularly to a kind of detonator coding positioning and knowledge method for distinguishing, dress
Put, equipment and computer-readable recording medium.
Background technology
Detonator is a kind of ignition equipment generally used in engineering explosion.In order to strengthen the management to detonator, life is implemented
Production, sell, using the management responsibility of links, effectively prevent the wandering society of industrial detonator, safeguard public safety, to detonator reality
Numbering management is applied, is to implement management responsibility, effectively pre- anti-thunder tube is lost in, and reduces an important measures of explosion.
Detonator coding (being made up of digital and a small amount of letter, a string of coded strings share 13 characters) is record detonator tool
The unique mark of body service condition.At present, the coding form of domestic industry detonator is mainly laser scored and two kinds of mechanical indentation,
Because laser scored efficiency is higher, therefore most of domestic manufacturer uses laser scored coding.
It is existing at this stage to be all based on traditional image procossing for the positioning of detonator coding character and knowledge method for distinguishing
With the characterization method of engineer.In processing detonator image process, it is necessary to set substantial amounts of artificial threshold value, and artificial design is special
Take over the positioning and identification in character for use.Due to intensity of illumination, the shadow of the factor such as background is reflective, inhomogeneities illumination and detonator material
Ring, detonator image has larger noise, is unfavorable for the positioning and identification of detonator coding.In addition, detonator coding have font it is small and
The characteristics of stroke is thin, some are even distributed in discontinuous point-like, and the positioning and identification for detonator coding bring greatly tired
It is difficult.
Therefore, how to realize that being accurately positioned with identification for detonator coding is skill that those skilled in the art need to solve at present
Art problem.
The content of the invention
The purpose of the application is to provide the method, apparatus of a kind of positioning of detonator coding and identification, equipment and computer-readable
Storage medium, this method can realize being accurately positioned and identifying for detonator coding.
In order to solve the above technical problems, the application provides a kind of detonator coding positioning and knows method for distinguishing, this method includes：
Pretreatment operation is carried out to the original image set received, obtains training image collection；
Default convolutional neural networks are trained according to the training image collection, obtains positioning network model and identifies network mould
Type；
The detonator image collected is positioned using the positioning network model, obtains the detonator of the detonator image
Encode area-of-interest；
The detonator coding area-of-interest is identified using the identification network model, obtains the detonator image
Detonator coding.
Optionally, the detonator image collected is positioned using the positioning network model, obtains the detonator figure
The detonator coding area-of-interest of picture, including：
Convolution operation and deconvolution operation are carried out to the detonator image, obtains location feature figure；
The candidate frame of preset number defined in the location feature figure；
Convolution operation is carried out to the location feature figure using convolution kernel is positioned, the candidate frame is obtained and belongs to the detonator
The probability of area-of-interest is encoded, and the candidate frame is relative to the position offset of the detonator coding area-of-interest；
The candidate frame of the maximum probability is selected, and the detonator figure is calculated according to the corresponding position offset
The detonator coding area-of-interest of picture.
Optionally, convolution operation is carried out to the detonator image and deconvolution operates, including：
According to formulaConvolution operation is carried out to the detonator image；
Wherein,For n-th of characteristic pattern of l layers, flFor convolution function,For m-th of characteristic pattern of l-1 layers,
To carry out the convolution kernel of convolution operation on l layers,For offset,The characteristic pattern number of l layers is connected to for l-1 layers
Mesh.
Optionally, the detonator coding area-of-interest is identified using the identification network model, obtained described
The detonator coding of detonator image, including：
Process of convolution is carried out to the detonator coding area-of-interest, is identified feature；
The identification feature is classified using multi-tag full articulamentum, obtains corresponding detonator coding.
Optionally, the described pair of original image set received carries out pretreatment operation, obtains training image collection, including：
Label using the coding received as the area-of-interest training image；
The training image and the label are stored to the training image collection；
Wherein, the pretreatment operation include image map operation, edge detecting operation, closed operation, noise-removal operation,
At least one of in labeling operation, image cutting and combination operation.
Optionally, default convolutional neural networks are being trained according to the training image collection, obtain positioning network model and
After identifying network model, in addition to：
The data that convolution operation obtains are normalized, in order to accelerate the training of convolutional neural networks and prevention
There is over-fitting.
Optionally, default convolutional neural networks are being trained according to the training image collection, obtain positioning network model and
After identifying network model, in addition to：
According to loss functionTo measure the property of positioning network model
Can be fine or not, and then weigh the degree of accuracy of the positioning network model；
Wherein, N is the number of the candidate frame to match with the detonator coding area-of-interest, Lloc(x, l, g) is
Position loss function, Lconf(x, c) is confidence level loss function, and α is weight term.
The application also provides a kind of device of detonator coding positioning and identification, and the device includes：
Pretreatment unit, for carrying out pretreatment operation to the original image set received, obtain training image collection；
Training unit, for training default convolutional neural networks according to the training image collection, obtain positioning network mould
Type and identification network model；
Positioning unit, for being positioned using the positioning network model to the detonator image collected, obtain described
The detonator coding area-of-interest of detonator image；
Recognition unit, for the detonator coding area-of-interest to be identified using the identification network model, obtain
To the detonator coding of the detonator image.
The application also provides a kind of equipment of detonator coding positioning and identification, and the equipment includes：
Memory, for storing computer program；
Processor, any one of as described above the detonator coding positioning and identification are realized during for performing the computer program
Method the step of.
The application also provides a kind of computer-readable recording medium, and calculating is stored with the computer-readable recording medium
Machine program, the side of any one detonator coding positioning and identification as described above is realized when the computer program is executed by processor
The step of method.
A kind of detonator coding positioning provided herein and knowledge method for distinguishing, by entering to the original image set received
Row pretreatment operation, obtains training image collection；Default convolutional neural networks are trained according to training image collection, obtain positioning network
Model and identification network model；The detonator image collected is positioned using network model is positioned, obtains detonator image
Detonator coding area-of-interest；Using identifying that detonator coding area-of-interest is identified network model, detonator image is obtained
Detonator coding.
Technical scheme provided herein, the characteristic of its convolutional coding structure can be utilized automatically according to convolutional neural networks
The characteristics of abstract picture feature of high level is extracted from image, convolutional neural networks is applied to the positioning of detonator coding character
With in identification work, default convolutional neural networks are trained by training image collection, obtain positioning network model and identify network
Model, eliminate the process of artificial extraction feature；Detonator image is determined using network model and identification network model is positioned
Position and identification, detonator coding is obtained, whole process just can carry out end-to-end identification without detonator image is cut into monocase, know
Other process is quick and accuracy of identification is high, realizes being accurately positioned and identifying for detonator coding, good application prospect be present.This Shen
A kind of device, equipment and the computer-readable recording medium of detonator coding positioning and identification are please additionally provided simultaneously, are had above-mentioned
Beneficial effect, it will not be repeated here.
Brief description of the drawings
, below will be to embodiment or existing in order to illustrate more clearly of the embodiment of the present application or technical scheme of the prior art
There is the required accompanying drawing used in technology description to be briefly described, it should be apparent that, drawings in the following description are only this
The embodiment of application, for those of ordinary skill in the art, on the premise of not paying creative work, can also basis
The accompanying drawing of offer obtains other accompanying drawings.
A kind of detonator coding that Fig. 1 is provided by the embodiment of the present application positions and known the flow chart of method for distinguishing；
A kind of detonator coding that Fig. 2 is provided by Fig. 1 positions and known a kind of practical manifestation mode of S101 in method for distinguishing
Flow chart；
A kind of detonator coding that Fig. 3 is provided by Fig. 1 positions and known a kind of practical manifestation mode of S103 in method for distinguishing
Flow chart；
Fig. 4 is the signal for the method that a kind of positioning network model that the embodiment of the present application provides is positioned to detonator image
Figure；
Fig. 5-a are a kind of signal distribution map of the candidate frame of the embodiment of the present application offer on characteristic pattern；
Fig. 5-b are signal distribution map of another candidate frame of the embodiment of the present application offer on characteristic pattern；
A kind of detonator coding that Fig. 6 is provided by Fig. 1 positions and known a kind of practical manifestation mode of S104 in method for distinguishing
Flow chart；
Fig. 7 is what detonator coding area-of-interest was identified a kind of identification network model that the embodiment of the present application provides
The schematic diagram of method；
A kind of detonator coding that Fig. 8 is provided by the embodiment of the present application positions and the structure chart of the device of identification；
Another detonator coding that Fig. 9 is provided by the embodiment of the present application positions and the structure chart of the device of identification；
A kind of positioning of detonator coding and the structure chart of identification equipment that Figure 10 is provided by the embodiment of the present application.
Embodiment
The core of the application is to provide the method, apparatus of a kind of positioning of detonator coding and identification, equipment and computer-readable
Storage medium, this method can realize being accurately positioned and identifying for detonator coding.
To make the purpose, technical scheme and advantage of the embodiment of the present application clearer, below in conjunction with the embodiment of the present application
In accompanying drawing, the technical scheme in the embodiment of the present application is clearly and completely described, it is clear that described embodiment is
Some embodiments of the present application, rather than whole embodiments.Based on the embodiment in the application, those of ordinary skill in the art
The every other embodiment obtained under the premise of creative work is not made, belong to the scope of the application protection.
Fig. 1 is refer to, a kind of detonator coding that Fig. 1 is provided by the embodiment of the present application positions and known the flow of method for distinguishing
Figure.
It specifically comprises the following steps：
S101：Pretreatment operation is carried out to the original image set received, obtains training image collection；
For training convolutional neural networks, it is necessary to carry out pretreatment operation to original image set, training image collection is obtained；
Optionally, original image set can be user's input, can also be downloaded on network, the application is to original graph
The source of image set is not especially limited；
Optionally, pretreatment operation is carried out to the original image set received, obtains training image collection, can include：
The original image concentrated to the original image received carries out pretreatment operation, obtains interested with detonator coding
The corresponding training image of area coordinate；
Label using the coding received as the training image；
Training image and label are correspondingly stored to training image collection；
Referred to herein as pretreatment operation include image map operation, edge detecting operation, closed operation, noise remove behaviour
At least one of in work, labeling operation, image cutting and combination operation.
S102：Default convolutional neural networks are trained according to training image collection, obtains positioning network model and identifies network
Model；
Referred to herein as training image collection train default convolutional neural networks, following two parts training step can be included
Suddenly：
Position training step：Default positioning convolutional neural networks are trained using training image collection, can be in iteration 30
Deconditioning after ten thousand times, obtain positioning network model, and reach by the positioning network model, when receiving an original image
When, it can quickly determine the position of detonator coding and the effect by its detonator coding extracted region out；Because the present invention is base
In the mode discovery of training initial data its detonator coding position feature rule, therefore can be closed in the training process by setting
Suitable learning rate and batch size (number of pictures of each iteration), which reaches, makes the convergent purpose of model；
Optionally, place is normalized in the data obtained when can also position network model to training by convolution operation
Reason, in order to accelerate the training of convolutional neural networks and over-fitting occurs in prevention；Can be fixed to training according to below equation
The data obtained during the network model of position by convolution operation are normalized：
Wherein,For the output characteristic of kth layer, x(k)For input feature vector, E [x(k)] be input feature vector average value,For the variance of input feature vector, γ(k)For zoom factor, β(k)For displacement factor,y (k)For nonlinear characteristic；
By normalization operation feature can be made to zoom to [0,1] this section；And in test, learned automatically using kth layer
The zoom factor γ practised(k)With displacement factor β(k)Original nonlinear characteristic is recovered to representy (k)；
Optionally, the estimation to whole data set average and variance can be used as by the use of the average and variance of one batch, criticizing
Output result after normalization is input to activation primitive ReLU functions so that strengthens non-linear expression's ability of model；
Optionally, to reduce data dimension, the output after activation primitive can also be input to pond layer and is used to reduce spy
Figure resolution ratio is levied, to reduce data dimension；Pond layer can be maximum pond, can take maximum in the convolution kernel of one 2 × 2
Number；
Optionally, can also be according to loss functionIt is fixed to measure
The performance quality of position network model, and then weigh the degree of accuracy of positioning network model；
Wherein, N is the number of candidate frame to match with detonator coding area-of-interest, Lloc(x, l, g) loses for position
Function, the detonator coding area-of-interest for regression forecasting frame l relative to markgCenter and width and height,
Lconf(x, c) is confidence level loss function, and the confidence level c, α that detonator coding is navigated to for weighing each candidate frame are weight
, it could be arranged to 1；
Recognition training step：In identification convolutional neural networks default using training image collection training, schemed using training
As the one-to-one relation of detonator coding with receiving, default identification convolutional neural networks are trained, due to convolutional Neural net
Network can utilize the characteristic of its convolutional coding structure automatically to extract the abstract picture feature of high level from image and utilize mark more
Sign full articulamentum to classify to each position, it is hereby achieved that identification network model.
S103：The detonator image collected is positioned using network model is positioned, the detonator for obtaining detonator image is compiled
Code area-of-interest；
When receiving the detonator image collected, a series of volume is passed through to the detonator image using network model is positioned
After product, deconvolution operation, the detonator coding area-of-interest of detonator image is finally given.
S104：Using identifying that detonator coding area-of-interest is identified network model, the detonator of detonator image is obtained
Coding.
Based on above-mentioned technical proposal, a kind of detonator coding provided herein positioning and knowledge method for distinguishing, according to convolution
The characteristics of neutral net can utilize the characteristic of its convolutional coding structure automatically to extract the abstract picture feature of high level from image,
Convolutional neural networks are applied to the positioning of detonator coding character with identification work, default volume is trained by training image collection
Product neutral net, obtain positioning network model and identify network model, eliminate the process of artificial extraction feature；Using positioning net
Network model and identification network model are positioned and identified to detonator image, obtain detonator coding, and whole process need not be by detonator
Image is cut into monocase and just can be identified, and identification process is quick and accuracy of identification is high, realizes the accurate fixed of detonator coding
, good application prospect be present in position and identification.
Based on above-described embodiment, Fig. 2 is refer to, a kind of detonator coding that Fig. 2 is provided by Fig. 1 positions and known method for distinguishing
A kind of flow chart of middle S101 practical manifestation mode.
The present embodiment is the S101 for a upper embodiment, is to be made that specific implementation to the content of S101 descriptions
Description, it is below the flow chart shown in Fig. 2, it specifically includes following steps：
S201：Map operation is carried out to original image；
When original image background is dark, and detonator coding and background subtraction are away from unobvious, original image can be mapped
Operation, maps to [0,1] section so that original image brightens by value of the gradation of image in section [0,0.4].
S202：Edge detecting operation is carried out to original image；
Optionally, edge detecting operation can be carried out using Sobel detection to original image, because detonator coding is present
There can be more pixel compared with strong edge, therefore by the detonator coding region of the original image of edge detecting operation.
S203：Closed operation is carried out to edge image；
1 × 23 convolution kernel can be used to carry out closed operation to edge image, the operation can make detonator coding edges of regions
Connect, and the pixel in other regions can be then eliminated；
S204：The region for being more than 2000 pixels to join domain area carries out noise-removal operation；
The region that 2000 pixels can be more than to join domain area carries out noise-removal operation, so that because noise causes
Be not belonging to detonator coding edge and connected region is got rid of；
S205：The coordinate of candidate frame is labeled；
Detonator coding region minimum enclosed rectangle can be searched out and the square of detonator coding is will not belong to by length-width ratio
Shape removes, and draws final candidate frame, and the coordinate of the candidate frame is labeled；Optionally, annotation formatting can be detonator
The upper left corner of coding and bottom right angular coordinate；
It can substantially confirm the coordinate in detonator coding region after labeling operation, but it is big due to fuzzy or noise be present
Original image, the original image of marking error can also be sent to user so that user by the method that manually marks to mark
The detonator coding region for noting the original image of mistake is labeled；Its frame coordinate is finally write into XML file, and finally given
Training image；Original image is labeled by above method, can greatly reduce artificial mark cost；
Optionally, label of the coding that will can also be received as training image so that training image is with receiving
Coding is corresponded, and network model is identified for training；Referred to herein as receive be encoded to by user's identification and warp
Cross the consistent coding of the detonator coding of the area-of-interest of positioning network model output.
Based on above-described embodiment, it refer to a kind of detonator that Fig. 3, Fig. 4, Fig. 5-a and Fig. 5-b, Fig. 3 are provided by Fig. 1 and compile
Code positioning and the flow chart for knowing S103 a kind of practical manifestation mode in method for distinguishing；Fig. 4 is one that the embodiment of the present application provides
The schematic diagram for the method that kind positioning network model is positioned to detonator image；Fig. 5-a are one kind that the embodiment of the present application provides
Signal distribution map of the candidate frame on characteristic pattern；Fig. 5-b are another candidate frame of the embodiment of the present application offer on characteristic pattern
Signal distribution map.
The present embodiment is the S103 for a upper embodiment, is to be made that specific implementation to the content of S103 descriptions
Description, it is below the flow chart shown in Fig. 3, it specifically includes following steps：
S301：Convolution operation and deconvolution operation are carried out to detonator image, obtains location feature figure；
Optionally, positioning network model mentioned above can be made up of 13 convolutional layers, can after each convolution operation
The different characteristic pattern of number is generated, can be according to formulaConvolution operation is carried out to detonator image；
Wherein,For n-th of characteristic pattern of l layers, flFor convolution function,For m-th of characteristic pattern of l-1 layers,To carry out the convolution kernel of convolution operation on l layers,For offset,The characteristic pattern of l layers is connected to for l-1 layers
Number；
For example, as shown in figure 4, a series of convolution can be carried out to detonator image, after batch normalization, pondization operate, most
Throughout one's life into 64 passages of size, the convolutional layer 7 that size is 5 × 5, and convolutional layer 7 is carried out deconvolution operate to obtain 64 passages,
Fisrt feature figure and 64 passages that size is 10 × 10, the second feature figure that size is 20 × 20；In entirely positioning network mould
In type, high-rise convolutional layer has high abstraction feature, and the convolutional layer in bottom is then more sensitive to positional information, in order to melt
This two-part characteristic information is closed, the deconvolution operation of addition can expand the resolution ratio of the characteristic pattern of high-rise low resolution,
And then cause the characteristic pattern after deconvolution that there is expression high abstraction feature and the ability of position feature, to the position of detonator coding
It is more sensitive, further to strengthen the feature of characteristic pattern, fisrt feature figure can also be subjected to dot product operations with convolutional layer 6 and obtained
64 passages, the third feature figure that size is 10 × 10, second feature figure and convolutional layer 5 carried out to dot product operations obtain 64 leading to
Road, the fourth feature figure that size is 20 × 20.
S302：The candidate frame of preset number defined in location feature figure；
As shown in Fig. 5-a and Fig. 5-b, in order to navigate to detonator coding, if defined in two obtained characteristic patterns
Dry candidate frame, optionally, the yardstick of candidate frame can determine according to below equation：
Wherein, skFor the yardstick of candidate frame, m is above-mentioned two feature graph laplacian, sminFor the lowermost layer of the candidate frame
Yardstick, smaxFor the top yardstick of the candidate frame；In addition for the length-width ratio of each candidate frame, in order to adapt to detonator coding
Size, length-width ratio can be set to ar={ 5,6 }, then the length and width of each candidate frame can be calculated according to below equation：
Wherein,For the length of candidate frame,For the width of candidate frame.
S303：Convolution operation is carried out to location feature figure using convolution kernel, obtaining candidate frame, to belong to detonator coding interested
The probability in region, and candidate frame is relative to the position offset of detonator coding area-of-interest；
Convolution operation is carried out on features described above figure using convolution kernel, each candidate frame can be obtained and belong to detonator coding sense
The probability in interest region, and each candidate frame are sat relative to the position offset of detonator coding area-of-interest, including center
Mark, long and width；
Optionally, the positioning convolution kernel can be 1 × 3 convolution kernel.
S304：The maximum candidate frame of select probability, and the thunder of detonator image is calculated according to corresponding position offset
Pipe encodes area-of-interest.
Optionally, can be by the candidate frame of non-maxima suppression algorithms selection maximum probability and inclined according to corresponding position
The detonator coding area-of-interest of detonator image is calculated in shifting amount.
It refer to a kind of detonator coding that Fig. 6 and Fig. 7, Fig. 6 are provided by Fig. 1 and position and know one of S104 in method for distinguishing
The flow chart of kind practical manifestation mode；Fig. 7 is that a kind of identification network model that the embodiment of the present application provides is emerging to detonator coding sense
The schematic diagram for the method that interesting region is identified.
The present embodiment is the S104 for a upper embodiment, is to be made that specific implementation to the content of S104 descriptions
Description, it is below the flow chart shown in Fig. 6, it specifically includes following steps：
S401：Process of convolution is carried out to detonator coding area-of-interest, is identified feature；
Optionally, as shown in fig. 7, the identification network model can have 5 convolutional layers, it can connect and criticize after each convolutional layer
The identification feature response that convolutional layer obtains is normalized to [0,1] section by normalization layer, to accelerate the receipts of the identification network model
Hold back, and improve generalization ability；
Optionally, can also the full connection of addition 13 after shared full articulamentum 1 in order to identify 13 characters
Network, the output layer of each network have 54 neurons to represent 54 characters being classified (10 digital 0-9 and English respectively
Except letter, lowercase c, o, s, u, v, w, x, z).
S402：The identification feature is classified using multi-tag full articulamentum, obtains corresponding detonator coding.
When using identifying that detonator coding area-of-interest is identified network model, to detonator coding area-of-interest
Process of convolution is carried out, feature is identified, identification feature is classified using multi-tag full articulamentum, obtains corresponding detonator
Coding.
Fig. 8 is refer to, a kind of detonator coding that Fig. 8 is provided by the embodiment of the present application positions and the structure of the device of identification
Figure.
The device can include：
Pretreatment unit 100, for carrying out pretreatment operation to the original image set received, obtain training image collection；
Training unit 200, for training default convolutional neural networks according to training image collection, obtain positioning network model
With identification network model；
Positioning unit 300, for being positioned to the detonator image collected using positioning network model, obtain detonator figure
The detonator coding area-of-interest of picture；
Recognition unit 400, for using identifying that detonator coding area-of-interest is identified network model, obtaining detonator
The detonator coding of image.
Fig. 9 is refer to, another detonator coding that Fig. 9 is provided by the embodiment of the present application positions and the knot of the device of identification
Composition.
The positioning unit 300 can include：
Feature obtains subelement, for carrying out convolution operation and deconvolution operation to detonator image, obtains location feature figure；
Define subelement, the candidate frame for the preset number defined in location feature figure；
Probability obtains subelement, for carrying out convolution operation to location feature figure using convolution kernel, obtains candidate frame and belongs to
The probability of detonator coding area-of-interest；
Deviation post subelement, for carrying out convolution operation to location feature figure using convolution kernel, it is relative to obtain candidate frame
In the position offset of detonator coding area-of-interest.
Subelement is selected, the candidate frame maximum for select probability, and calculated according to the corresponding position offset
To the detonator coding area-of-interest of the detonator image.
The recognition unit 400 can include：
Subelement is identified, for carrying out process of convolution to detonator coding area-of-interest, is identified feature；
Classification subelement, for classifying using sort program to identification feature, obtains corresponding detonator coding.
The pretreatment unit 100 can include：
Subelement is pre-processed, the original image for being concentrated to the original image received carries out pretreatment operation, obtains
Corresponding training image with detonator coding area-of-interest coordinate；
Label subelement, for the label using the coding received as training image；
Storing sub-units, for training image and label correspondingly to be stored to training image collection；
Wherein, pretreatment operation includes image map operation, edge detecting operation, closed operation, noise-removal operation, mark
At least one of in operation, image cutting and combination operation.
The device can also include：
Normalization unit, for place to be normalized to the data obtained during training positioning network model by convolution operation
Reason, in order to accelerate the training of convolutional neural networks and over-fitting occurs in prevention.
The device can also include：
Metric element, for according to loss functionIt is fixed to measure
The performance quality of position network model, and then weigh the degree of accuracy of positioning network model；
Wherein, N is the number of candidate frame to match with detonator coding area-of-interest, Lloc(x, l, g) loses for position
Function, Lconf(x, c) is confidence level loss function, and α is weight term.
Because the embodiment of device part and the embodiment of method part are mutually corresponding, therefore the embodiment of device part please
Referring to the description of the embodiment of method part, wouldn't repeat here.
Figure 10 is refer to, a kind of positioning of detonator coding and the structure of identification equipment that Figure 10 is provided by the embodiment of the present application
Figure.
The equipment can produce bigger difference because configuration or performance are different, can include one or more processing
Device (central processing units, CPU) 522 (for example, one or more processors) and memory 532, one
The individual or storage medium 530 of more than one storage application program 542 or data 544 (such as one or more mass memories
Equipment).Wherein, memory 532 and storage medium 530 can be of short duration storage or persistently storage.It is stored in storage medium 530
Program can include one or more modules (diagram does not mark), and each module can include to a series of fingers in device
Order operation.Further, central processing unit 522 could be arranged to communicate with storage medium 530, positions and knows in detonator coding
The series of instructions operation in storage medium 530 is performed in other equipment 500.
Detonator coding positions and identification equipment 500 can also include one or more power supplys 525, one or one with
Upper wired or wireless network interface 550, one or more input/output interfaces 558, and/or, one or more behaviour
Make system 541, such as Windows ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM etc..
Above-mentioned Fig. 1 to described by Fig. 7 detonator coding positioning and know method for distinguishing in step by detonator coding position and
The equipment of identification is realized based on the structure shown in the Figure 10.
It is apparent to those skilled in the art that for convenience and simplicity of description, the system of foregoing description,
The specific work process of device and unit, the corresponding process in preceding method embodiment is may be referred to, will not be repeated here.
In several embodiments provided herein, it should be understood that disclosed device, apparatus and method, can be with
Realize by another way.For example, device embodiment described above is only schematical, for example, the division of unit,
Only a kind of division of logic function, can there is an other dividing mode when actually realizing, such as multiple units or component can be with
With reference to or be desirably integrated into another system, or some features can be ignored, or not perform.It is another, it is shown or discussed
Mutual coupling or direct-coupling or communication connection can be by some interfaces, the INDIRECT COUPLING of device or unit or
Communication connection, can be electrical, mechanical or other forms.
The unit illustrated as separating component can be or may not be physically separate, be shown as unit
Part can be or may not be physical location, you can with positioned at a place, or can also be distributed to multiple networks
On unit.Some or all of unit therein can be selected to realize the purpose of this embodiment scheme according to the actual needs.
In addition, each functional unit in each embodiment of the application can be integrated in a processing unit, can also
That unit is individually physically present, can also two or more units it is integrated in a unit.Above-mentioned integrated list
Member can both be realized in the form of hardware, can also be realized in the form of SFU software functional unit.
If integrated unit is realized in the form of SFU software functional unit and is used as independent production marketing or in use, can
To be stored in a computer read/write memory medium.Based on such understanding, the technical scheme of the application substantially or
Saying all or part of the part to be contributed to prior art or the technical scheme can be embodied in the form of software product
Out, the computer software product is stored in a storage medium, including some instructions are causing a computer equipment
(can be personal computer, funcall device, or network equipment etc.) performs the whole of each embodiment method of the application
Or part steps.And foregoing storage medium includes：USB flash disk, mobile hard disk, read-only storage (Read-Only Memory,
ROM), random access memory (RandomAccess Memory, RAM), magnetic disc or CD etc. are various can be with storage program generation
The medium of code.
A kind of detonator coding provided herein is positioned above and method, apparatus, equipment and the computer of identification can
Storage medium is read to be described in detail.Specific case used herein is explained the principle and embodiment of the application
State, the explanation of above example is only intended to help and understands the present processes and its core concept.It should be pointed out that for this skill
For the those of ordinary skill in art field, on the premise of the application principle is not departed from, some change can also be carried out to the application
Enter and modify, these are improved and modification is also fallen into the application scope of the claims.
It should also be noted that, in this manual, such as first and second or the like relational terms be used merely to by
One entity or operation make a distinction with another entity or operation, and not necessarily require or imply these entities or operation
Between any this actual relation or order be present.Moreover, term " comprising ", "comprising" or its any other variant meaning
Covering including for nonexcludability, so that process, method, article or equipment including a series of elements not only include that
A little key elements, but also the other element including being not expressly set out, or also include for this process, method, article or
The intrinsic key element of equipment.In the absence of more restrictions, the key element limited by sentence "including a ...", is not arranged
Except other identical element in the process including key element, method, article or equipment being also present.
Claims (10)
1. a kind of detonator coding positioning and knowledge method for distinguishing, it is characterised in that including：
Pretreatment operation is carried out to the original image set received, obtains training image collection；
Default convolutional neural networks are trained according to the training image collection, obtains positioning network model and identifies network model；
The detonator image collected is positioned using the positioning network model, obtains the detonator coding of the detonator image
Area-of-interest；
The detonator coding area-of-interest is identified using the identification network model, obtains the thunder of the detonator image
Pipe encodes.
2. according to the method for claim 1, it is characterised in that using the positioning network model to the detonator figure that collects
As being positioned, the detonator coding area-of-interest of the detonator image is obtained, including：
Convolution operation and deconvolution operation are carried out to the detonator image, obtains location feature figure；
The candidate frame of preset number defined in the location feature figure；
Convolution operation is carried out to the location feature figure using convolution kernel is positioned, the candidate frame is obtained and belongs to the detonator coding
The probability of area-of-interest, and the candidate frame is relative to the position offset of the detonator coding area-of-interest；
The candidate frame of the maximum probability is selected, and the detonator image is calculated according to the corresponding position offset
Detonator coding area-of-interest.
3. according to the method for claim 2, it is characterised in that convolution operation is carried out to the detonator image and deconvolution is grasped
Make, including：
According to formulaConvolution operation is carried out to the detonator image；
Wherein,For n-th of characteristic pattern of l layers, flFor convolution function,For m-th of characteristic pattern of l-1 layers,For
The convolution kernel of convolution operation is carried out on l layers,For offset, Vn lThe feature map number of l layers is connected to for l-1 layers.
4. according to the method for claim 1, it is characterised in that using the identification network model to the detonator coding sense
Interest region is identified, and obtains the detonator coding of the detonator image, including：
Process of convolution is carried out to the detonator coding area-of-interest, is identified feature；
The identification feature is classified using multi-tag full articulamentum, obtains corresponding detonator coding.
5. according to the method for claim 1, it is characterised in that the described pair of original image set received carries out pretreatment behaviour
Make, obtain training image collection, including：
The original image concentrated to the original image received carries out pretreatment operation, obtains interested with detonator coding
The corresponding training image of area coordinate；
Label using the coding received as the training image；
The training image and the label are correspondingly stored to the training image collection；
Wherein, the pretreatment operation includes image map operation, edge detecting operation, closed operation, noise-removal operation, mark
At least one of in operation, image cutting and combination operation.
6. according to the method for claim 5, it is characterised in that default convolution god is being trained according to the training image collection
Through network, obtain after positioning network model and identifying network model, in addition to：
The data that convolution operation obtains are normalized, in order to accelerate the training of convolutional neural networks and prevent to occur
Over-fitting.
7. according to the method for claim 5, it is characterised in that default convolution god is being trained according to the training image collection
Through network, obtain after positioning network model and identifying network model, in addition to：
According to loss functionIt is good to measure the performance of positioning network model
It is bad, and then weigh the degree of accuracy of the positioning network model；
Wherein, N is the number of the candidate frame to match with the detonator coding area-of-interest, Lloc(x, l, g) is position
Loss function, Lconf(x, c) is confidence level loss function, and α is weight term.
8. a kind of detonator coding positioning and the device of identification, it is characterised in that including：
Pretreatment unit, for carrying out pretreatment operation to the original image set received, obtain training image collection；
Training unit, for training default convolutional neural networks according to the training image collection, obtain positioning network model and
Identify network model；
Positioning unit, for being positioned using the positioning network model to the detonator image collected, obtain the detonator
The detonator coding area-of-interest of image；
Recognition unit, for the detonator coding area-of-interest to be identified using the identification network model, obtain institute
State the detonator coding of detonator image.
9. a kind of detonator coding positioning and the equipment of identification, it is characterised in that including：
Memory, for storing computer program；
Processor, realize during for performing the computer program detonator coding as described in any one of claim 1 to 7 position and
The step of knowing method for distinguishing.
10. a kind of computer-readable recording medium, it is characterised in that be stored with computer on the computer-readable recording medium
Program, the detonator coding positioning as described in any one of claim 1 to 7 is realized when the computer program is executed by processor and is known
The step of method for distinguishing.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711092470.5A CN107808157B (en) | 2017-11-08 | 2017-11-08 | Detonator code positioning and identifying method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711092470.5A CN107808157B (en) | 2017-11-08 | 2017-11-08 | Detonator code positioning and identifying method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107808157A true CN107808157A (en) | 2018-03-16 |
| CN107808157B CN107808157B (en) | 2021-08-13 |
Family
ID=61591924
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711092470.5A Active CN107808157B (en) | 2017-11-08 | 2017-11-08 | Detonator code positioning and identifying method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107808157B (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108416327A (en) * | 2018-03-28 | 2018-08-17 | 京东方科技集团股份有限公司 | A kind of object detection method, device, computer equipment and readable storage medium storing program for executing |
| CN108564035A (en) * | 2018-04-13 | 2018-09-21 | 杭州睿琪软件有限公司 | The method and system for the information recorded on identification document |
| CN108564570A (en) * | 2018-03-29 | 2018-09-21 | 哈尔滨工业大学（威海） | A kind of method and apparatus of intelligentized pathological tissues positioning |
| CN108985305A (en) * | 2018-06-26 | 2018-12-11 | 武汉科技大学 | A kind of positioning of laser-induced thermal etching industrial detonator coded image and bearing calibration |
| CN109035666A (en) * | 2018-08-29 | 2018-12-18 | 深圳市中电数通智慧安全科技股份有限公司 | A kind of fire-smoke detection method, apparatus and terminal device |
| CN109657671A (en) * | 2018-12-19 | 2019-04-19 | 深圳供电局有限公司 | Nameplate character recognition method and device, computer equipment and storage medium |
| CN110619325A (en) * | 2018-06-20 | 2019-12-27 | 北京搜狗科技发展有限公司 | Text recognition method and device |
| CN113269190A (en) * | 2021-07-21 | 2021-08-17 | 中国平安人寿保险股份有限公司 | Data classification method and device based on artificial intelligence, computer equipment and medium |
| CN114004807A (en) * | 2021-10-29 | 2022-02-01 | 推想医疗科技股份有限公司 | Method and device for identifying positioning patch |
| CN114708234A (en) * | 2022-04-12 | 2022-07-05 | 北京优创新港科技股份有限公司 | Method and device for identifying number of detonators on automatic bayonet coding all-in-one machine |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103295006A (en) * | 2013-01-18 | 2013-09-11 | 扬州群有信息科技有限公司 | High-performance detonator coded image recognition device |
| CN103530642A (en) * | 2012-08-21 | 2014-01-22 | 深圳市中金岭南有色金属股份有限公司凡口铅锌矿 | Detonator automatic recognition machine and detonator coding image recognition method |
| CN204790511U (en) * | 2015-05-20 | 2015-11-18 | 郑州长河电子工程有限公司 | A equipment is write in into to information that is used for digital electron detonator production line |
| CN106339591A (en) * | 2016-08-25 | 2017-01-18 | 汤平 | Breast cancer prevention self-service health cloud service system based on deep convolutional neural network |
| CN106372390A (en) * | 2016-08-25 | 2017-02-01 | 姹ゅ钩 | Deep convolutional neural network-based lung cancer preventing self-service health cloud service system |
-
2017
- 2017-11-08 CN CN201711092470.5A patent/CN107808157B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103530642A (en) * | 2012-08-21 | 2014-01-22 | 深圳市中金岭南有色金属股份有限公司凡口铅锌矿 | Detonator automatic recognition machine and detonator coding image recognition method |
| CN103295006A (en) * | 2013-01-18 | 2013-09-11 | 扬州群有信息科技有限公司 | High-performance detonator coded image recognition device |
| CN204790511U (en) * | 2015-05-20 | 2015-11-18 | 郑州长河电子工程有限公司 | A equipment is write in into to information that is used for digital electron detonator production line |
| CN106339591A (en) * | 2016-08-25 | 2017-01-18 | 汤平 | Breast cancer prevention self-service health cloud service system based on deep convolutional neural network |
| CN106372390A (en) * | 2016-08-25 | 2017-02-01 | 姹ゅ钩 | Deep convolutional neural network-based lung cancer preventing self-service health cloud service system |
Non-Patent Citations (2)
| Title |
|---|
| SHAOQING REN 等: "Faster R-CNN Towards Real-Time Object Detection with Region Proposal Networks", 《IEEE》 * |
| 孙小虎: "金属***编码字符识别方法研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108416327A (en) * | 2018-03-28 | 2018-08-17 | 京东方科技集团股份有限公司 | A kind of object detection method, device, computer equipment and readable storage medium storing program for executing |
| CN108564570A (en) * | 2018-03-29 | 2018-09-21 | 哈尔滨工业大学（威海） | A kind of method and apparatus of intelligentized pathological tissues positioning |
| CN108564035A (en) * | 2018-04-13 | 2018-09-21 | 杭州睿琪软件有限公司 | The method and system for the information recorded on identification document |
| US10977513B2 (en) | 2018-04-13 | 2021-04-13 | Hangzhou Glorify Software Limited | Method, system and computer readable storage medium for identifying information carried on sheet |
| CN108564035B (en) * | 2018-04-13 | 2020-09-25 | 杭州睿琪软件有限公司 | Method and system for identifying information recorded on document |
| CN110619325A (en) * | 2018-06-20 | 2019-12-27 | 北京搜狗科技发展有限公司 | Text recognition method and device |
| CN110619325B (en) * | 2018-06-20 | 2024-03-08 | 北京搜狗科技发展有限公司 | Text recognition method and device |
| CN108985305A (en) * | 2018-06-26 | 2018-12-11 | 武汉科技大学 | A kind of positioning of laser-induced thermal etching industrial detonator coded image and bearing calibration |
| CN109035666B (en) * | 2018-08-29 | 2020-05-19 | 深圳市中电数通智慧安全科技股份有限公司 | Fire and smoke detection method and device and terminal equipment |
| CN109035666A (en) * | 2018-08-29 | 2018-12-18 | 深圳市中电数通智慧安全科技股份有限公司 | A kind of fire-smoke detection method, apparatus and terminal device |
| CN109657671A (en) * | 2018-12-19 | 2019-04-19 | 深圳供电局有限公司 | Nameplate character recognition method and device, computer equipment and storage medium |
| CN113269190A (en) * | 2021-07-21 | 2021-08-17 | 中国平安人寿保险股份有限公司 | Data classification method and device based on artificial intelligence, computer equipment and medium |
| CN114004807A (en) * | 2021-10-29 | 2022-02-01 | 推想医疗科技股份有限公司 | Method and device for identifying positioning patch |
| CN114708234A (en) * | 2022-04-12 | 2022-07-05 | 北京优创新港科技股份有限公司 | Method and device for identifying number of detonators on automatic bayonet coding all-in-one machine |
| CN114708234B (en) * | 2022-04-12 | 2022-12-02 | 北京优创新港科技股份有限公司 | Method and device for identifying number of detonators on automatic bayonet coding all-in-one machine |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107808157B (en) | 2021-08-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107808157A (en) | A kind of method and device of detonator coding positioning and identification | |
| CN113160192B (en) | Visual sense-based snow pressing vehicle appearance defect detection method and device under complex background | |
| CN103049763B (en) | Context-constraint-based target identification method | |
| CN106295646A (en) | A kind of registration number character dividing method based on degree of depth study and device | |
| CN108564088A (en) | Licence plate recognition method, device, equipment and readable storage medium storing program for executing | |
| CN110738207A (en) | character detection method for fusing character area edge information in character image | |
| CN109063706A (en) | Verbal model training method, character recognition method, device, equipment and medium | |
| CN105574550A (en) | Vehicle identification method and device | |
| CN109815979B (en) | Weak label semantic segmentation calibration data generation method and system | |
| CN109583483A (en) | A kind of object detection method and system based on convolutional neural networks | |
| CN107944450A (en) | A kind of licence plate recognition method and device | |
| CN111753828A (en) | Natural scene horizontal character detection method based on deep convolutional neural network | |
| CN106295502A (en) | A kind of method for detecting human face and device | |
| CN111257341A (en) | Underwater building crack detection method based on multi-scale features and stacked full convolution network | |
| CN107203775A (en) | A kind of method of image classification, device and equipment | |
| CN113822116A (en) | Text recognition method and device, computer equipment and storage medium | |
| CN108537286A (en) | A kind of accurate recognition methods of complex target based on key area detection | |
| CN111222545B (en) | Image classification method based on linear programming incremental learning | |
| CN110288033B (en) | Sugarcane top feature identification and positioning method based on convolutional neural network | |
| CN109934095A (en) | A kind of remote sensing images Clean water withdraw method and system based on deep learning | |
| CN112232411B (en) | HarDNet-Lite optimization method in embedded platform | |
| CN115136209A (en) | Defect detection system | |
| CN112364974A (en) | Improved YOLOv3 algorithm based on activation function | |
| CN112241736A (en) | Text detection method and device | |
| Zhang et al. | Traffic Sign Detection and Recognition Based onDeep Learning. |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |