CN110533654A - The method for detecting abnormality and device of components - Google Patents
The method for detecting abnormality and device of components Download PDFInfo
- Publication number
- CN110533654A CN110533654A CN201910817968.6A CN201910817968A CN110533654A CN 110533654 A CN110533654 A CN 110533654A CN 201910817968 A CN201910817968 A CN 201910817968A CN 110533654 A CN110533654 A CN 110533654A
- Authority
- CN
- China
- Prior art keywords
- picture
- neural network
- network model
- components
- twin neural
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
- G06T7/0008—Industrial image inspection checking presence/absence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
- G06T7/001—Industrial image inspection using an image reference approach
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/70—Determining position or orientation of objects or cameras
- G06T7/73—Determining position or orientation of objects or cameras using feature-based methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
- G06V10/507—Summing image-intensity values; Histogram projection analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
- G06T2207/30164—Workpiece; Machine component
Abstract
The present invention provides a kind of method for detecting abnormality of components and devices, the above method includes: the normal sample and exceptional sample for obtaining the first components, wherein, the normal sample includes: the first picture of first components, and the exceptional sample includes: the second picture that the second components of different location are in first components；Twin neural network model is trained using the normal sample and the exceptional sample, wherein the twin neural network model is for determining the similarity of the normal sample and exceptional sample that are input in the twin neural network model；The first components being input in the third picture of the twin neural network model using the twin neural network model detection after the completion of training are with the presence or absence of abnormal.By the invention it is possible to solve the problems, such as that the efficiency of the abnormality detection of components in the related technology is lower.
Description
Technical field
The present invention relates to the communications fields, in particular to the method for detecting abnormality and device of a kind of components.
Background technique
When the components to vehicle carry out abnormality detection, can be obtained by movement of the rail vehicle in front of line-scan digital camera
Entire vehicle picture, the picture of vehicle contains various components informations, and the related defects of various components can be with
It is reflected in the picture.For example, screw is that occur the relatively broad components of abnormal problem in the components of vehicle.Right
When the picture of vehicle is analyzed, the information such as texture, edge, the color in image can be utilized based on traditional images processing technique
Carry out the feature extraction work of screw abnormal.And this mode, have the following deficiencies: that the picture of vehicle is very big, and each screw
Occupied area is very small, it is difficult to directly extract the spy of screw in an entire vehicle pictures using the technology that traditional images are handled
Sign；The technology of traditional images processing is difficult to the abnormal problem of the screw expanded in the various different pieces of vehicle, needs to be every
Feature extraction work is separately provided in a part, so as to cause the inefficiency of the abnormality detection of screw.
For the problems such as in the related technology, the efficiency of the abnormality detection of components is lower, not yet proposition technical solution.
Summary of the invention
The embodiment of the invention provides a kind of method for detecting abnormality of components and devices, at least to solve in the related technology
The lower problem of the efficiency of the abnormality detection of components.
According to one embodiment of present invention, a kind of method for detecting abnormality of components is provided, comprising:
Obtain the normal sample and exceptional sample of the first components, wherein the normal sample includes: the described 1st
First picture of component, the exceptional sample include: be in first components different location the second components
Two pictures；
Twin neural network model is trained using the normal sample and the exceptional sample, wherein described
Twin neural network model is for determining the normal sample and the phase of exceptional sample that are input in the twin neural network model
Like degree；
The twin neural network model is input to using the twin neural network model detection after the completion of training
The first components in third picture are with the presence or absence of abnormal.
Optionally, described that twin neural network model is instructed using the normal sample and the exceptional sample
Practice, comprising:
Multiple groups training data is input in the twin neural network model, using the multiple groups training data to described
Twin neural network model is trained, wherein the multiple groups training data includes at least first group of training data and second group
Training data, wherein first group of training data includes the first normal sample and the first exceptional sample, second group of instruction
Practicing data includes the second normal sample and third normal sample.
Optionally, the twin neural network model detection using after the completion of training is input to the twin nerve
The first components in the third picture of network model are with the presence or absence of abnormal, comprising:
Obtain first picture and third picture；
First picture and third picture are input in the twin neural network model after the completion of the training, obtained
Take the twin neural network model to the similarity obtained after first picture and the third picture analyzing；
It is similar to being obtained after first picture and the third picture analyzing according to the twin neural network model
Degree determines the first components in the third picture with the presence or absence of abnormal.
Optionally, it is described according to the twin neural network model to first picture and the third picture analyzing after
Obtained similarity determines the first components in the third picture with the presence or absence of abnormal, comprising:
In the case where the similarity is less than first threshold, there are different for the first part for determining in the third picture
Often；
In the case where the similarity is greater than second threshold, it is different to determine that the first part in the third picture does not exist
Often.
According to another embodiment of the invention, a kind of abnormal detector of components is provided, comprising:
Module is obtained, for obtaining the normal sample and exceptional sample of the first components, wherein the normal sample packet
Include: the first picture of first components, the exceptional sample include: to be in the of different location with first components
The second picture of two components；
Training module, for being instructed using the normal sample and the exceptional sample to twin neural network model
Practice, wherein the twin neural network model for determine the normal sample that is input in the twin neural network model with
The similarity of exceptional sample；
Detection module, for being input to the twin mind using the twin neural network model detection after the completion of training
The first components in third picture through network model are with the presence or absence of abnormal.
Optionally, the training module, comprising:
Training unit uses the multiple groups for multiple groups training data to be input in the twin neural network model
Training data is trained the twin neural network model, wherein the multiple groups training data includes at least first group of instruction
Practice data and second group of training data, wherein first group of training data includes the first normal sample and the first abnormal sample
This, second group of training data includes the second normal sample and third normal sample.
Optionally, the detection module, comprising:
Acquiring unit, for obtaining first picture and third picture；
Input unit, the twin nerve for being input to first picture and third picture after the completion of the training
In network model, the twin neural network model is obtained to the phase obtained after first picture and the third picture analyzing
Like degree；
Determination unit is used for according to the twin neural network model to first picture and the third picture analyzing
The similarity obtained afterwards determines the first components in the third picture with the presence or absence of abnormal.
Optionally, the determination unit, comprising:
First determines subelement, for determining the third picture in the case where the similarity is less than first threshold
In the first part exist it is abnormal；
Second determines subelement, for determining the third picture in the case where the similarity is greater than second threshold
In the first part not there is no exception.
Optionally, according to another embodiment of the invention, a kind of storage medium is provided, is stored in the storage medium
There is computer program, wherein the computer program is arranged to execute the above method when operation.
Optionally, according to another embodiment of the invention, a kind of electronic device, including memory and processing are provided
Device, is stored with computer program in the memory, the processor be arranged to run the computer program to execute on
State method.
Through the invention, the normal sample and exceptional sample of the first components are obtained, wherein the normal sample packet
Include: the first picture of first components, the exceptional sample include: to be in the of different location with first components
The second picture of two components；Twin neural network model is instructed using the normal sample and the exceptional sample
Practice, wherein the twin neural network model for determine the normal sample that is input in the twin neural network model with
The similarity of exceptional sample；The twin nerve net is input to using the twin neural network model detection after the completion of training
The first components in the third picture of network model are with the presence or absence of abnormal.Therefore, it can solve the different of components in the related technology
The lower problem of the efficiency often detected, improves the efficiency of the abnormality detection of components, and can be improved the accurate of abnormality detection
Property.
Detailed description of the invention
The drawings described herein are used to provide a further understanding of the present invention, constitutes part of this application, this hair
Bright illustrative embodiments and their description are used to explain the present invention, and are not constituted improper limitations of the present invention.In the accompanying drawings:
Fig. 1 is the flow chart according to the method for detecting abnormality of the components of the embodiment of the present invention；
Fig. 2 is the flow chart according to the localization method of the wheel of the embodiment of the present invention；
Fig. 3 is the flow chart according to the localization method of the screw of the embodiment of the present invention；
Fig. 4 is the flow chart according to the method for detecting abnormality of the components of another embodiment of the present invention；
Fig. 5 is the structural block diagram of the abnormal detector of components according to an embodiment of the present invention.
Specific embodiment
Hereinafter, the present invention will be described in detail with reference to the accompanying drawings and in combination with Examples.It should be noted that not conflicting
In the case of, the features in the embodiments and the embodiments of the present application can be combined with each other.
It should be noted that description and claims of this specification and term " first " in above-mentioned attached drawing, "
Two " etc. be to be used to distinguish similar objects, without being used to describe a particular order or precedence order.
Embodiment 1
The embodiment of the invention provides a kind of method for detecting abnormality of components.Fig. 1 is according to the zero of the embodiment of the present invention
The flow chart of the method for detecting abnormality of component, as shown in Figure 1, comprising:
Step S102 obtains the normal sample and exceptional sample of the first components, wherein the normal sample includes:
First picture of first components, the exceptional sample include: to be in the second of different location with first components
The second picture of components；
Step S104 is trained twin neural network model using the normal sample and the exceptional sample,
Wherein, the twin neural network model is for determining the normal sample and exception that are input in the twin neural network model
The similarity of sample；
Step S106 is input to the twin nerve net using the twin neural network model detection after the completion of training
The first components in the third picture of network model are with the presence or absence of abnormal.
Through the invention, the normal sample and exceptional sample of the first components are obtained, wherein the normal sample packet
Include: the first picture of first components, the exceptional sample include: to be in the of different location with first components
The second picture of two components；Twin neural network model is instructed using the normal sample and the exceptional sample
Practice, wherein the twin neural network model for determine the normal sample that is input in the twin neural network model with
The similarity of exceptional sample；The twin nerve net is input to using the twin neural network model detection after the completion of training
The first components in the third picture of network model are with the presence or absence of abnormal.Therefore, it can solve the different of components in the related technology
The lower problem of the efficiency often detected, improves the efficiency of the abnormality detection of components, and can be improved the accurate of abnormality detection
Property.
It should be noted that in embodiments of the present invention, due to using the first picture of the first components as normal sample,
And the second picture that the second components of different location will be in first components is upper as exceptional sample, and use
It states normal sample and exceptional sample to be trained twin neural network model, without collecting vehicle in actual motion mistake
Exceptional sample in journey.Since the difficulty for collecting exceptional sample in vehicle actual moving process is larger, and also it is difficult to collect
To the exceptional sample that can satisfy quantity required, and if by the exceptional sample picture of machine generation screw, then it is given birth to by machine
At screw abnormal samples pictures be not inconsistent with actual conditions, thus if being trained using the exceptional sample picture to model not
It can accurately identify the exception that screw generates in reality.It therefore, through the above embodiments of the present invention, can be further
It solves the problems, such as that exceptional sample collects difficulty, reduces the cost and model management cost of model training, and can be improved instruction
In the abnormality detection of experienced reliability and the components that can be extended in various components.
It is described to use the normal sample and the exceptional sample to twin mind in an alternative embodiment of the invention
It is trained through network model, comprising: multiple groups training data is input in the twin neural network model, using described more
Group training data is trained the twin neural network model, wherein the multiple groups training data includes at least first group
Training data and second group of training data, wherein first group of training data includes that the first normal sample and first are abnormal
Sample, second group of training data include the second normal sample and third normal sample.
It should be noted that the normal sample includes first normal sample, second normal sample and institute
Third normal sample is stated, the exceptional sample includes first exceptional sample；The one zero in second normal sample
Part is corresponding at the first time；The first components corresponding second time in the third normal sample.Optionally, the first exceptional sample
It can be the exceptional sample obtained after the first normal sample is modified or adjusted.It is alternatively possible to modify first just
The position or surface of the first components obtain the first exceptional sample in normal sample, for example, to the 1st in the first normal sample
The position of component is adjusted and (modifies the position of the first components in the first normal sample), in the first normal sample
It adds dirty trace etc. and obtains the first exceptional sample in the surface of first components.
Optionally, in the above embodiment of the invention, when being trained to twin neural network model, by training number
It is input in twin neural network model in a pair wise manner according to (for example, picture).It should be noted that twin neural network mould
It include two neural networks in type, for two pictures being input in the twin neural network model in pairs, wherein a figure
Piece is input into a neural network in above-mentioned two neural network, and another picture is input into above-mentioned two nerve net
In another neural network in network.Optionally, each neural network can calculate the feature vector of the picture of input, twin
Raw neural network model is based on calculated feature vector, determines two pictures being input in the twin neural network model
Similitude.
In the above embodiment of the invention, a normal sample and an exceptional sample are input to the twin mind in pairs
It through in network model, and is also input to two normal samples are pairs of in the twin neural network model, to the nerve
Network model is trained.It should be noted that two normal samples being input in twin neural network in a pair wise manner,
It can be two pictures of the first components corresponding to different time in normal condition (that is, no exceptions), example
Such as, a normal sample is the picture of first components on the same day, and another normal sample is the figure of the first components of the previous day
Piece.
In the above embodiment of the invention, normal sample and exceptional sample are input to twin nerve in a pair wise manner
In network model, which is trained.
In an alternative embodiment of the invention, the twin neural network model using after the completion of training is detected
The first components in the third picture of the twin neural network model are input to the presence or absence of abnormal, comprising: described in acquisition
First picture and third picture；First picture and third picture are input to the twin nerve after the completion of the training
In network model, the twin neural network model is obtained to the phase obtained after first picture and the third picture analyzing
Like degree；It is true to the similarity obtained after first picture and the third picture analyzing according to the twin neural network model
The first components in the fixed third picture are with the presence or absence of abnormal.
In an alternative embodiment of the invention, it is described according to the twin neural network model to first picture and
The similarity obtained after the third picture analyzing determines the first components in the third picture with the presence or absence of abnormal, packet
It includes: in the case where the similarity is less than first threshold, it is abnormal to determine that the first part in the third picture exists；Institute
Similarity is stated greater than in the case where second threshold, determines that the first part in the third picture does not have exception.
Optionally, above-mentioned second threshold can be threshold value identical with first threshold.
It is explained below in conjunction with method for detecting abnormality of the example to above-mentioned components, but is not used in restriction originally
The technical solution of inventive embodiments.Below by taking components are the screw on wheel as an example, to the exemplary technical solution of the present invention into
Row is described in detail, and the exemplary technical solution of the present invention is as follows:
(1), component locating module, for extracting the component where components.
The step mainly first positions wheel, and the relative position then tightened up a screw according to wheel again carries out screw
Positioning.
Optionally, in the above-described embodiments, wheel alignment can be realized by the following method, refer to Fig. 2, Fig. 2 is root
According to the flow chart of the localization method of the wheel of the embodiment of the present invention.As shown in Fig. 2, the localization method of the wheel of the embodiment of the present invention
It is as follows:
Step 1: inputting the figure of whole train first；Later, step 2: being handled by convolutional neural networks, candidate region mentions
It takes, wheel detection, coordinate return scheduling algorithm and position to wheel position；Step 3, using the higher region of confidence level as wheel
Picture has finally determined wheel position；Step 4 will can individually be divided in the figure of whole train and extract each wheel
Picture.Since the wheel that train speed difference causes line-scan digital camera to acquire is not necessarily positive round, it can be to the vehicle being partitioned into
The picture of wheel is corrected.
Optionally, in the above-described embodiments, screw positioning can be realized by the following method, refer to Fig. 3, Fig. 3 is root
According to the flow chart of the localization method of the screw of the embodiment of the present invention.As shown in figure 3, the localization method of the screw of the embodiment of the present invention
It is as follows:
It is step 1, similar with wheel alignment, the figure of entire wheel is inputted first；Step 2, by convolutional neural networks handle,
Candidate region extraction, screw detecting, coordinate return scheduling algorithm and position to screw position；It is step 3, finally that confidence level is higher
Region as screw picture, the position of screw has finally been determined；The screw relative position information that step 4, basis have been set in advance,
The ownership of each screw in screw picture is confirmed.
It should be noted that in the above-described embodiments, the positioning of wheel and screw is all made of the convolution based on candidate region
Neural network.Optionally, convolutional neural networks can be using in advance by the positive and negative sample data manually extracted (for example, wheel figure
Piece and non-wheel picture, screw picture and non-screw picture) it is trained.
(2), the metric learning module compared based on historical data；
In view of in some cases, actually there is abnormal defect screw sample less, if largely using analogue data pair
Twin neural network model is trained the reliability that will affect result, therefore proposes the measurement being compared based on historical data
Study (metric learning obtains the similitude of two objects that is, by two objects of comparison).Detailed process is, by with same
Position screw be normal sample training data, different location screw be exceptional sample training data, by as unit of by spiral shell
Silk picture, which is input in twin neural network model, to be trained, and the similitude (i.e. similarity) to screw is finally exported.
(3), screw abnormal detection module.
In abnormality detection, to history image (the history figure of the same screw of the screw picture and the position that newly detect
Picture when seeming the screw no exceptions, i.e. normal picture) it is compared, thought if similar without abnormal, if not
It is similar, think that abnormal (for example, fall off, loosen, is dirty etc.) occurs in screw, and alarm.Fig. 4 is referred to, Fig. 4 is according to this hair
The flow chart of the method for detecting abnormality of the components of bright another embodiment.As shown in figure 4, the components of the embodiment of the present invention is different
Normal detection method is as follows:
The current image of screw and history picture are input in twin neural network model by step 1；Wherein, this is twin
Neural network model includes convolutional neural networks 1 and convolutional neural networks 2；
Step 2, the similarity measurement (i.e. similitude) that two pictures inputted are calculated by twin neural network model.
It can also include the data processing module based on classification in an alternative embodiment of the invention, it, can in the module
By using the exception of detection of classifier screw.Since defective data (i.e. exceptional sample) is less, in embodiments of the present invention,
It can be made with the picture (for example, the position of the screw in the picture of normal screw is modified) of manually generated defect screw
For exceptional sample, then using the picture of normal screw as normal sample, classifier (i.e. disaggregated model) is trained.It is optional
Ground, the method construct classifier that can be combined using HOG and SVM.When being carried out abnormality detection using the classifier, with zero
Part picture is as input, using the output of classifier as the foundation of determining defects (i.e. abnormality detection).
Wherein, histograms of oriented gradients (Histogram of Oriented Gradient, referred to as HOG) is characterized in one
Kind is used to carry out the Feature Descriptor of object detection in computer vision and image procossing.HOG feature is by calculating and counting
The gradient orientation histogram of image local area carrys out constitutive characteristic.Support vector machines (Support Vector Machine,
Referred to as SVM also known as support vector network) be classification in regression analysis analysis data supervised learning model with it is related
Learning algorithm.
It optionally, can when the screw in the screw picture that newly detects of twin neural network model confirmation is deposited when abnormal
Abnormal label is stamped with the screw picture newly detected to this.When abnormal picture is enough, will have in system
The exceptional sample that the normal sample of tape label and detection obtain can use this two parts data to carry out pair of a step at this time
Disaggregated model is trained.
Embodiment 2
According to another embodiment of the invention, a kind of abnormal detector of components is provided, the device is for real
Existing above-described embodiment and preferred embodiment, the descriptions that have already been made will not be repeated.As used below, term " module "
The combination of the software and/or hardware of predetermined function may be implemented.Although device described in following embodiment is preferably with software
It realizes, but the realization of the combination of hardware or software and hardware is also that may and be contemplated.
Fig. 5 is the structural block diagram of the abnormal detector of components according to an embodiment of the present invention, which includes:
Module 502 is obtained, for obtaining the normal sample and exceptional sample of the first components, wherein the normal sample
It originally include: the first picture of first components, the exceptional sample includes: to be in different location with first components
The second components second picture；
Training module 504, for using the normal sample and the exceptional sample to twin neural network model into
Row training, wherein the twin neural network model is for determining the normal sample being input in the twin neural network model
The similarity of this and exceptional sample；
Detection module 506, it is described twin for being input to using the twin neural network model detection after the completion of training
The first components in the third picture of raw neural network model are with the presence or absence of abnormal.
Through the invention, the normal sample and exceptional sample of the first components are obtained, wherein the normal sample packet
Include: the first picture of first components, the exceptional sample include: to be in the of different location with first components
The second picture of two components；Twin neural network model is instructed using the normal sample and the exceptional sample
Practice, wherein the twin neural network model for determine the normal sample that is input in the twin neural network model with
The similarity of exceptional sample；The twin nerve net is input to using the twin neural network model detection after the completion of training
The first components in the third picture of network model are with the presence or absence of abnormal.Therefore, it can solve the different of components in the related technology
The lower problem of the efficiency often detected, improves the efficiency of the abnormality detection of components, and can be improved the accurate of abnormality detection
Property.
In an alternative embodiment of the invention, the training module 504, comprising: training unit, for multiple groups to be trained
Data are input in the twin neural network model, using the multiple groups training data to the twin neural network model into
Row training, wherein the multiple groups training data includes at least first group of training data and second group of training data, wherein described
First group of training data includes the first normal sample and the first exceptional sample, and second group of training data includes second normal
Sample and third normal sample.
In an alternative embodiment of the invention, the detection module 506, comprising: acquiring unit, for obtaining described
One picture and third picture；Input unit is completed for first picture and third picture to be input to the training
In twin neural network model afterwards, the twin neural network model is obtained to first picture and the third picture point
The similarity obtained after analysis；Determination unit, for according to the twin neural network model to first picture and described the
The similarity obtained after three picture analyzings determines the first components in the third picture with the presence or absence of abnormal.
In an alternative embodiment of the invention, the determination unit, comprising: first determines subelement, for described
In the case that similarity is less than first threshold, it is abnormal to determine that the first part in the third picture exists；Second determines that son is single
Member, for it is different to determine that the first part in the third picture does not exist in the case where the similarity is greater than second threshold
Often.
Embodiment 3
The embodiments of the present invention also provide a kind of storage medium, which includes the program of storage, wherein above-mentioned
The method of any of the above-described is executed when program is run.
Optionally, in the present embodiment, above-mentioned storage medium can be set to store the journey for executing following steps
Sequence code:
S1 obtains the normal sample and exceptional sample of the first components, wherein the normal sample includes: described the
First picture of one components, the exceptional sample include: the second components that different location is in first components
Second picture；
S2 is trained twin neural network model using the normal sample and the exceptional sample, wherein institute
State twin neural network model for determine the normal sample that is input in the twin neural network model and exceptional sample
Similarity；
S3 is input to the twin neural network model using the twin neural network model detection after the completion of training
Third picture in the first components with the presence or absence of abnormal.
Optionally, in the present embodiment, above-mentioned storage medium can include but is not limited to: USB flash disk, read-only memory (Read-
Only Memory, referred to as ROM), it is random access memory (Random Access Memory, referred to as RAM), mobile hard
The various media that can store program code such as disk, magnetic or disk.
Optionally, the specific example in the present embodiment can be with reference to described in above-described embodiment and optional embodiment
Example, details are not described herein for the present embodiment.
Embodiment 4
The embodiments of the present invention also provide a kind of electronic device, including memory and processor, stored in the memory
There is computer program, which is arranged to run computer program to execute the step in any of the above-described embodiment of the method
Suddenly.
Optionally, above-mentioned electronic device can also include transmission device and input-output equipment, wherein the transmission device
It is connected with above-mentioned processor, which connects with above-mentioned processor.
Optionally, in the present embodiment, above-mentioned processor can be set to execute following steps by computer program:
S1 obtains the normal sample and exceptional sample of the first components, wherein the normal sample includes: described the
First picture of one components, the exceptional sample include: the second components that different location is in first components
Second picture；
S2 is trained twin neural network model using the normal sample and the exceptional sample, wherein institute
State twin neural network model for determine the normal sample that is input in the twin neural network model and exceptional sample
Similarity；
S3 is input to the twin neural network model using the twin neural network model detection after the completion of training
Third picture in the first components with the presence or absence of abnormal.
Optionally, the specific example in the present embodiment can be with reference to described in above-described embodiment and optional embodiment
Example, details are not described herein for the present embodiment.
Obviously, those skilled in the art should be understood that each module of the above invention or each step can be with general
Computing device realize that they can be concentrated on a single computing device, or be distributed in multiple computing devices and formed
Network on, optionally, they can be realized with the program code that computing device can perform, it is thus possible to which they are stored
It is performed by computing device in the storage device, and in some cases, it can be to be different from shown in sequence execution herein
Out or description the step of, perhaps they are fabricated to each integrated circuit modules or by them multiple modules or
Step is fabricated to single integrated circuit module to realize.In this way, the present invention is not limited to any specific hardware and softwares to combine.
The foregoing is only a preferred embodiment of the present invention, is not intended to restrict the invention, for the skill of this field
For art personnel, the invention may be variously modified and varied.It is all within principle of the invention, it is made it is any modification, etc.
With replacement, improvement etc., should all be included in the protection scope of the present invention.
Claims (10)
1. a kind of method for detecting abnormality of components characterized by comprising
Obtain the normal sample and exceptional sample of the first components, wherein the normal sample includes: first components
The first picture, the exceptional sample includes: the second figure that the second components of different location are in first components
Piece；
Twin neural network model is trained using the normal sample and the exceptional sample, wherein described twin
Neural network model is for determining the similarity of the normal sample and exceptional sample that are input in the twin neural network model；
The third of the twin neural network model is input to using the twin neural network model detection after the completion of training
The first components in picture are with the presence or absence of abnormal.
2. the method according to claim 1, wherein described use the normal sample and the exceptional sample
Twin neural network model is trained, comprising:
Multiple groups training data is input in the twin neural network model, using the multiple groups training data to described twin
Neural network model is trained, wherein the multiple groups training data includes at least first group of training data and second group of training
Data, wherein first group of training data includes the first normal sample and the first exceptional sample, described second group trained number
According to including the second normal sample and third normal sample.
3. the method according to claim 1, wherein the twin neural network using after the completion of training
Model inspection is input to the first components in the third picture of the twin neural network model with the presence or absence of abnormal, comprising:
Obtain first picture and third picture；
First picture and third picture are input in the twin neural network model after the completion of the training, institute is obtained
Twin neural network model is stated to the similarity obtained after first picture and the third picture analyzing；
It is true to the similarity obtained after first picture and the third picture analyzing according to the twin neural network model
The first components in the fixed third picture are with the presence or absence of abnormal.
4. according to the method described in claim 3, it is characterized in that, it is described according to the twin neural network model to described
The similarity obtained after one picture and the third picture analyzing determines that the first components in the third picture whether there is
It is abnormal, comprising:
In the case where the similarity is less than first threshold, it is abnormal to determine that the first part in the third picture exists；
In the case where the similarity is greater than second threshold, determine that the first part in the third picture does not have exception.
5. a kind of abnormal detector of components characterized by comprising
Module is obtained, for obtaining the normal sample and exceptional sample of the first components, wherein the normal sample includes:
First picture of first components, the exceptional sample include: to be in the second of different location with first components
The second picture of components；
Training module, for being trained using the normal sample and the exceptional sample to twin neural network model,
Wherein, the twin neural network model is for determining the normal sample and exception that are input in the twin neural network model
The similarity of sample；
Detection module, for being input to the twin nerve net using the twin neural network model detection after the completion of training
The first components in the third picture of network model are with the presence or absence of abnormal.
6. device according to claim 5, which is characterized in that the training module, comprising:
Training unit uses multiple groups training for multiple groups training data to be input in the twin neural network model
Data are trained the twin neural network model, wherein the multiple groups training data includes at least first group of trained number
According to second group of training data, wherein first group of training data include the first normal sample and the first exceptional sample, institute
Stating second group of training data includes the second normal sample and third normal sample.
7. device according to claim 5, which is characterized in that the detection module, comprising:
Acquiring unit, for obtaining first picture and third picture；
Input unit, the twin neural network for being input to first picture and third picture after the completion of the training
In model, it is similar to obtaining after first picture and the third picture analyzing to obtain the twin neural network model
Degree；
Determination unit, for being obtained according to the twin neural network model to after first picture and the third picture analyzing
To similarity determine the first components in the third picture with the presence or absence of abnormal.
8. device according to claim 7, which is characterized in that the determination unit, comprising:
First determines subelement, for determining in the third picture in the case where the similarity is less than first threshold
First part exists abnormal；
Second determines subelement, for determining in the third picture in the case where the similarity is greater than second threshold
Not there is no exception in the first part.
9. a kind of storage medium, which is characterized in that be stored with computer program in the storage medium, wherein the computer
Program is arranged to execute method described in any one of Claims 1-4 when operation.
10. a kind of electronic device, including memory and processor, which is characterized in that be stored with computer journey in the memory
Sequence, the processor are arranged to run the computer program to execute side described in any one of Claims 1-4
Method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910817968.6A CN110533654A (en) | 2019-08-30 | 2019-08-30 | The method for detecting abnormality and device of components |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910817968.6A CN110533654A (en) | 2019-08-30 | 2019-08-30 | The method for detecting abnormality and device of components |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110533654A true CN110533654A (en) | 2019-12-03 |
Family
ID=68665651
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910817968.6A Pending CN110533654A (en) | 2019-08-30 | 2019-08-30 | The method for detecting abnormality and device of components |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110533654A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111445459A (en) * | 2020-03-27 | 2020-07-24 | 广东工业大学 | Image defect detection method and system based on depth twin network |
| CN111460991A (en) * | 2020-03-31 | 2020-07-28 | 科大讯飞股份有限公司 | Anomaly detection method, related device and readable storage medium |
| CN111709920A (en) * | 2020-06-01 | 2020-09-25 | 深圳市深视创新科技有限公司 | Template defect detection method |
| CN111768387A (en) * | 2020-06-30 | 2020-10-13 | 创新奇智(青岛)科技有限公司 | Flaw detection method, twin neural network training method and device and electronic equipment |
| CN112308148A (en) * | 2020-11-02 | 2021-02-02 | 创新奇智(青岛)科技有限公司 | Defect category identification and twin neural network training method, device and storage medium |
| CN113655348A (en) * | 2021-07-28 | 2021-11-16 | 国网湖南省电力有限公司 | Power equipment partial discharge fault diagnosis method based on deep twin network, system terminal and readable storage medium |
| CN114708724A (en) * | 2021-06-28 | 2022-07-05 | 天津大学 | Man-vehicle-road cooperative multi-vehicle energy efficiency optimization algorithm |
| CN115041544A (en) * | 2022-08-11 | 2022-09-13 | 深圳市信润富联数字科技有限公司 | Method and device for detecting abnormality of stamping part and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108961236A (en) * | 2018-06-29 | 2018-12-07 | 国信优易数据有限公司 | Training method and device, the detection method and device of circuit board defect detection model |
| CN109559301A (en) * | 2018-11-20 | 2019-04-02 | 江苏第二师范学院（江苏省教育科学研究院） | A kind of EMU end to end enters the twin network method of institute's defects detection |
| US20190108396A1 (en) * | 2017-10-11 | 2019-04-11 | Aquifi, Inc. | Systems and methods for object identification |
| CN110136101A (en) * | 2019-04-17 | 2019-08-16 | 杭州数据点金科技有限公司 | A kind of tire X-ray defect detection method compared based on twin distance |
-
2019
- 2019-08-30 CN CN201910817968.6A patent/CN110533654A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190108396A1 (en) * | 2017-10-11 | 2019-04-11 | Aquifi, Inc. | Systems and methods for object identification |
| CN108961236A (en) * | 2018-06-29 | 2018-12-07 | 国信优易数据有限公司 | Training method and device, the detection method and device of circuit board defect detection model |
| CN109559301A (en) * | 2018-11-20 | 2019-04-02 | 江苏第二师范学院（江苏省教育科学研究院） | A kind of EMU end to end enters the twin network method of institute's defects detection |
| CN110136101A (en) * | 2019-04-17 | 2019-08-16 | 杭州数据点金科技有限公司 | A kind of tire X-ray defect detection method compared based on twin distance |
Non-Patent Citations (1)
| Title |
|---|
| 赵鑫欣: "基于卷积神经网络的铁轨扣件异常状态检测研究", 《中国优秀硕士学位论文全文数据库工程科技Ⅱ辑》 * |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111445459A (en) * | 2020-03-27 | 2020-07-24 | 广东工业大学 | Image defect detection method and system based on depth twin network |
| CN111445459B (en) * | 2020-03-27 | 2023-07-28 | 广东工业大学 | Image defect detection method and system based on depth twin network |
| CN111460991A (en) * | 2020-03-31 | 2020-07-28 | 科大讯飞股份有限公司 | Anomaly detection method, related device and readable storage medium |
| CN111709920A (en) * | 2020-06-01 | 2020-09-25 | 深圳市深视创新科技有限公司 | Template defect detection method |
| CN111768387A (en) * | 2020-06-30 | 2020-10-13 | 创新奇智(青岛)科技有限公司 | Flaw detection method, twin neural network training method and device and electronic equipment |
| CN112308148A (en) * | 2020-11-02 | 2021-02-02 | 创新奇智(青岛)科技有限公司 | Defect category identification and twin neural network training method, device and storage medium |
| CN114708724A (en) * | 2021-06-28 | 2022-07-05 | 天津大学 | Man-vehicle-road cooperative multi-vehicle energy efficiency optimization algorithm |
| CN113655348A (en) * | 2021-07-28 | 2021-11-16 | 国网湖南省电力有限公司 | Power equipment partial discharge fault diagnosis method based on deep twin network, system terminal and readable storage medium |
| CN113655348B (en) * | 2021-07-28 | 2023-12-08 | 国网湖南省电力有限公司 | Power equipment partial discharge fault diagnosis method, system terminal and readable storage medium based on deep twin network |
| CN115041544A (en) * | 2022-08-11 | 2022-09-13 | 深圳市信润富联数字科技有限公司 | Method and device for detecting abnormality of stamping part and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110533654A (en) | The method for detecting abnormality and device of components | |
| CN110060237B (en) | Fault detection method, device, equipment and system | |
| US5933519A (en) | Cytological slide scoring apparatus | |
| US11704888B2 (en) | Product onboarding machine | |
| JP6791864B2 (en) | Barcode tag detection in side view sample tube images for laboratory automation | |
| CN104615986B (en) | The method that pedestrian detection is carried out to the video image of scene changes using multi-detector | |
| CN108830332A (en) | A kind of vision vehicle checking method and system | |
| CN113111844B (en) | Operation posture evaluation method and device, local terminal and readable storage medium | |
| CN107808126A (en) | Vehicle retrieval method and device | |
| CN116188475B (en) | Intelligent control method, system and medium for automatic optical detection of appearance defects | |
| CN111161295B (en) | Dish image background stripping method | |
| CN111242899B (en) | Image-based flaw detection method and computer-readable storage medium | |
| CN112257660B (en) | Method, system, equipment and computer readable storage medium for removing invalid passenger flow | |
| CN113723157B (en) | Crop disease identification method and device, electronic equipment and storage medium | |
| CN109886932A (en) | Gear ring of wheel speed sensor detection method of surface flaw based on SVM | |
| EP3360077A1 (en) | Method and system for classifying objects from a stream of images | |
| CN110163109A (en) | A kind of lane line mask method and device | |
| CN113763348A (en) | Image quality determination method and device, electronic equipment and storage medium | |
| CN108182444A (en) | The method and device of video quality diagnosis based on scene classification | |
| CN112966687B (en) | Image segmentation model training method and device and communication equipment | |
| CN112686122B (en) | Human body and shadow detection method and device, electronic equipment and storage medium | |
| CN117218633A (en) | Article detection method, device, equipment and storage medium | |
| CN110866917A (en) | Tablet type and arrangement mode identification method based on machine vision | |
| CN115984759A (en) | Substation switch state identification method and device, computer equipment and storage medium | |
| CN114494355A (en) | Trajectory analysis method and device based on artificial intelligence, terminal equipment and medium |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20191203 |
|
| WD01 | Invention patent application deemed withdrawn after publication |