CN109993734A - Method and apparatus for output information - Google Patents
Method and apparatus for output information Download PDFInfo
- Publication number
- CN109993734A CN109993734A CN201910248418.7A CN201910248418A CN109993734A CN 109993734 A CN109993734 A CN 109993734A CN 201910248418 A CN201910248418 A CN 201910248418A CN 109993734 A CN109993734 A CN 109993734A
- Authority
- CN
- China
- Prior art keywords
- defect
- image
- detection model
- training
- defects detection
- 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
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
- G06T7/62—Analysis of geometric attributes of area, perimeter, diameter or volume
-
- 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
- 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
Abstract
Embodiment of the disclosure discloses the method and apparatus for output information.One specific embodiment of this method includes: the image for obtaining object to be detected；By the defects detection model based on Mask RCNN algorithm of image input training in advance, the corresponding defect classification of image and defect profile position are obtained；Defect information based on defect classification and defect profile position output described image.This embodiment improves the efficiency of object appearance quality testing and accuracy, convenient for the optimization and upgrading of subsequent quality evaluation and production line.
Description
Technical field
This disclosure relates to field of computer technology, and in particular to the method and apparatus for output information.
Background technique
In traditional industry manufacturing industry production scene, quality inspection is the key link in production procedure.In steel production, automobile
In the fields such as manufacture, papermaking, battery manufacture, solar panels manufacture, printed wiring board, chip, liquid crystal display, to product quality
A kind of important means controlled is detected to the surface state of product, to judge product with the presence or absence of flaw and lack
It falls into, and corresponding processing is done to product according to testing result.It is this to be based on product table planar in the production of traditional industry manufacturing industry
The quality inspection of state is mostly that manual inspection or semi-automatic optical instrument assist quality inspection, not only inefficiency, but also is easy to appear erroneous judgement,
In addition, the industrial data that this mode generates is not easy to store, manage and mining again recycles.
Summary of the invention
Embodiment of the disclosure proposes the method and apparatus for output information.
In a first aspect, embodiment of the disclosure provides a kind of method for output information, comprising: in response to receiving
Detection request, obtains the image of object to be detected；By the defect inspection based on Mask RCNN algorithm of image input training in advance
Model is surveyed, the corresponding defect classification of image and defect profile position are obtained；Based on defect classification and defect profile position output figure
The defect information of picture.
In some embodiments, obtain the image of object to be detected, comprising: according at least one of following parameter obtain to
At least one image of the object of detection: angle, filter, times mirror, focuses light.
In some embodiments, defects detection model includes Faster RCNN and example segmentation network, Faster RCNN
Using SE-ResNet as basic network.
In some embodiments, training obtains defects detection model as follows: obtaining training sample set, training
Sample includes sample image, defect classification corresponding with sample image and the defect profile position of object to be detected；It will train
The sample image of training sample in sample set is as input, by defect classification corresponding with the sample image of input and defect
Outline position obtains defects detection model as output, training.
In some embodiments, the penalty values of defects detection model include Classification Loss value, return penalty values and example
Divide the penalty values of the prediction of network.
Second aspect, embodiment of the disclosure provide a kind of device for output information, comprising: receiving unit, quilt
It is configured to receive detection request, obtains the image of object to be detected；Quality inspection unit is configured to input image
The defects detection model based on Mask RCNN algorithm of training in advance, obtains the corresponding defect classification of image and defect profile position
It sets；Processing unit is configured to the defect information based on defect classification and defect profile position output image.
In some embodiments, receiving unit is further configured to: being obtained according at least one of following parameter to be detected
Object at least one image: angle, light, filter, times mirror, focus.
In some embodiments, defects detection model includes Faster RCNN and example segmentation network, Faster RCNN
Using SE-ResNet as basic network.
In some embodiments, which further includes training unit, is configured to: obtaining training sample set, training sample
This includes sample image, defect classification corresponding with sample image and the defect profile position of object to be detected；It will training sample
The sample image of training sample in this set is as input, by defect classification corresponding with the sample image of input and defect wheel
Wide position obtains defects detection model as output, training.
In some embodiments, the penalty values of defects detection model include Classification Loss value, return penalty values and example
Divide the penalty values of the prediction of network.
The third aspect, embodiment of the disclosure provide a kind of electronic equipment, comprising: one or more processors；Storage
Device is stored thereon with one or more programs, when one or more programs are executed by one or more processors, so that one
Or multiple processors are realized such as method any in first aspect.
Fourth aspect, embodiment of the disclosure provide a kind of computer-readable medium, are stored thereon with computer program,
Wherein, it realizes when program is executed by processor such as method any in first aspect.
The method and apparatus for output information that embodiment of the disclosure provides, by being produced using image capture device
The image acquired in real time on product production line carries out detection judgement to the surface quality of product in real time, if detecting current process
There are quality problems for the product of image capture device, then judge the quality problems type and the box position gone wrong and
The outline position to go wrong.Improve the efficiency and accuracy of object appearance quality testing.And convenient for statistical shortcomings information from
And facilitate subsequent quality management &control.
Detailed description of the invention
By reading a detailed description of non-restrictive embodiments in the light of the attached drawings below, the disclosure is other
Feature, objects and advantages will become more apparent upon:
Fig. 1 is that the embodiment of the present application can be applied to exemplary system architecture figure therein；
Fig. 2 is the flow chart according to one embodiment of the method for output information of the application；
Fig. 3 a, 3b, 3c are the network knots of the defects detection model used according to the method for output information of the application
Composition；
Fig. 4 is the flow chart according to one embodiment of the defects detection model training method of the application；
Fig. 5 is the structural schematic diagram according to one embodiment of the device for output information of the application；
Fig. 6 is adapted for the structural schematic diagram for the computer system for realizing the electronic equipment of the embodiment of the present application.
Specific embodiment
The disclosure is described in further detail with reference to the accompanying drawings and examples.It is understood that this place is retouched
The specific embodiment stated is used only for explaining related invention, rather than the restriction to the invention.It also should be noted that in order to
Convenient for description, part relevant to related invention is illustrated only in attached drawing.
It should be noted that in the absence of conflict, the feature in embodiment and embodiment in the disclosure can phase
Mutually combination.The disclosure is described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.
Fig. 1 is shown can be using the method for output information of the disclosure or the implementation of the device for output information
The exemplary system architecture 100 of example.
As shown in Figure 1, system architecture 100 may include object 101 to be detected, camera 1021,1022,1023,
1024, server 103.Network is to provide communication link between 1024, server 103 in camera 1021,1022,1023
Medium.Network may include various connection types, such as wired, wireless communication link or fiber optic cables etc..
Object 101 to be detected can be the product for needing to carry out appearance detection on production line.Especially wooden product,
For example, wooden toy etc..The appearance of these toys be limited to using timber, it is possible that it is some as caused by timber lack
Fall into, such as worm hole, cracking, collapse it is scarce, the defects of incrustation, it is therefore desirable to carry out surface quality detection to wooden toy.
Server 103 can include: image capturing system, console, defects detection model, training engine, control module, instruction
Practice each and every one several main modulars such as database, Production database, service response system.
Image capturing system, using high precision image acquisition camera, adjustment angle, filter, times mirror, is focused light, is adopted
Collect multiple images for subsequent processing.
The picture that image capturing system on production line generates in real time is converted detection request by console, and according to pre- on line
The deployment scenario real-time perfoming load balancing and scheduling for surveying model will test request and be sent to the optimal prediction model that carries
On server.Defects detection model when running really on the server, the model are completed via training engine training.Model pair
After image data in the detection request of arrival carries out preset pretreatment, object detection calculating is carried out, and provides representative and lacks
The outline position information of sunken classification information and defect, and result is sent to control module.Control module and business scenario knot
Design is closed, the response for meeting production environment scene requirement can be made to the prediction result that model provides, such as according to business demand
Alarm, storage log, control mechanical arm etc..Control module can be using the processing behavior of prediction result and response as production date on line
Will is stored into Production database.The core of system-defects detection model is by training engine according to the training of history labeled data
It obtains.
Defects detection model uses depth convolutional neural networks structure (Deep CNNs), is divided using Mask RCNN example
Based on algorithm.Its theory structure is as shown in Figure 3a.Input of the original image as model on production line, the classification of defect
With outline position as output.Mask RCNN algorithm increases the network of example segmentation on the basis of Faster RCNN calculates
Branch.It is big to be reverted to original image using the algorithm based on two points of interpolation by the characteristic pattern that the branch extracts in basic network for characteristic pattern
It is small, predict to each pixel the example belonging to it.For the prediction result of each element, cross entropy fortune is done with true value
It calculates, obtains its loss.Then the loss and the loss of Faster RCNN are combined, is combined training (joint
Training), optimize network model parameter.
Faster RCNN algorithm is Mask RCNN algorithm basis, and network structure is as shown in Figure 3a.The algorithm is sharp first
With the convolution operation of disaggregated model, its characteristic pattern is obtained.Then candidate region network (Region Proposal is utilized
Whether include specific object: if recycling convolutional network to carry out comprising object if Network) calculating in a certain region of original image
Then its object category and boundary box (bounding box) are predicted in feature extraction；Without dividing if not including object
Class.In this way, the loss of three network branches is combined, combined training, Optimized model parameter are done.When model output with
When error amount between label is less than the preset threshold value for meeting business need, deconditioning.
Trained model each time can gradually be replaced by the online mode of small flow just running on line it is old
Model is extended with to achieve the purpose that model with service dynamic extensive.
It should be noted that the method provided by embodiment of the disclosure for output information is generally by server 103
It executes, correspondingly, the device for output information is generally positioned in server 103.
It should be understood that object to be detected, camera in Fig. 1, the number of server are only schematical.According to reality
It now needs, can have any number of object to be detected, camera, server.
With continued reference to Fig. 2, the process of one embodiment of the method for output information according to the disclosure is shown
200.This is used for the method for output information, comprising the following steps:
Step 201, in response to receiving detection request, the image of object to be detected is obtained.
In the present embodiment, can lead to for the executing subject of the method for output information (such as server shown in FIG. 1)
It crosses wired connection mode or radio connection receives detection request, then obtained from the camera of shooting examined object
Image.The image that can be acquired to camera pre-processes, and obtains the image of object to be detected after removing background image.Example
Such as, detect the position of object to pluck out object to be detected from the image that camera is shot by edge detection algorithm
Image.Controllable camera adjusts the angle, or the image of the different angle of object is obtained by the camera of different location.Also
Rotatable examined object is to obtain the image of multi-angle.It optionally, can also be by adjusting at least one following ginseng of camera
Number obtains at least one image of object to be detected: angle, filter, times mirror, focuses light.For example, flash lamp can be obtained open
The image of flash lamp is not opened.
Step 202, by the defects detection model based on Mask RCNN algorithm of image input training in advance, image is obtained
Corresponding defect classification and defect profile position.
In the present embodiment, which can judge to whether there is defect and defect according to the image of input
Classification and defect profile position.The network structure of Mask RCNN includes the network structure and example segmentation net of Faster RCNN
Network.The defective locations that Faster RCNN can determine the classification of defect and be outlined with block form.Example divides network can be with
It identifies the outline position of defect, and then can determine the quantity and area of defect.
As shown in Figure 3a, Mask RCNN algorithm extracts characteristic pattern by feature extraction layer first, such as utilizes disaggregated model
As basic network, its characteristic pattern is obtained using the convolution operation of basic network.Then RPN (Region Proposal is utilized
Network, candidate region network) it whether calculates in a certain region of original image comprising specific object: if sharp again comprising object
Feature extraction is carried out with ROI (region of interest) pond layer, layer is then returned by classification and predicts its object category
With boundary box (bounding box)；Without classification if not including object.In this way, by the loss of three network branches
It combines, does combined training, Optimized model parameter.When the error amount between the output and true value of model is less than certain threshold
When value, deconditioning.
The basic network of feature extraction layer can be the networks such as AlexNet, VGG, GoogleNet, ResNet.
RPN network is mainly used for generating region proposals (region recommendation), firstly generates a pile Anchor box
(anchor point box), it is carried out after cutting filtering by softmax judge Anchors belong to prospect (foreground) or after
Scape (background), i.e. " being object " or " not being object ", so this is one two classification.Meanwhile another branch
Bounding box regression (recurrence of boundary box) corrects anchor box, forms more accurate region
proposals。
ROI Align (ROI alignment) layer utilizes last in the region proposals and feature extraction layer of RPN generation
The feature map (characteristic pattern) that layer obtains, obtains the proposal feature map (recommended characteristics figure) of fixed size, into
Target identification and positioning can be carried out using full attended operation by entering to below.RoI Align is introduced instead of in Faster RCNN
RoI Pooling (pond ROI).Because RoI Pooling is not (the pixel-to-pixel being aligned one by one according to pixel
Alignment), perhaps this influence to bounding box is not very big, but is had a significant impact for the precision of mask.Make
50% is improved significantly to from 10% with the precision of mask after RoI Align.
Classification, which returns layer, can form ROI aligned layer the full attended operation of feature map progress of fixed size, utilize
Softmax carries out the classification of specific category, meanwhile, bounding box regression, which is completed, using L1Loss returns operation
Obtain the exact position of object.
Mask RCNN algorithm increases the network branches of example segmentation on the basis of Faster RCNN calculates.The branch
Characteristic pattern is extracted by basic network, characteristic pattern is reverted into original image size using the algorithm based on two points of interpolation, to each
A pixel predict the example belonging to it.For the prediction result of each element, cross entropy operation is done with true value, obtains it
Loss.Then the loss and the loss of Faster RCNN are combined, is done combined training (joint training), it is excellent
Change network model parameter.
In some optional implementations of the present embodiment, basic network can be used SE-ResNet and (squeeze excitation residual error
Network).On the basis of ResNet, extruding (Squeeze) and excitation (Excitation) operation are increased, can make full use of
Relationship between characteristic pattern difference channel.Specifically, each feature channel is got automatically exactly by way of study
Then significance level goes to promote useful feature and inhibits the feature little to current task use according to this significance level.
Fig. 3 b is the schematic diagram for the SE-ResNet that the disclosure uses.An input x is given, feature port number is c1, by a series of
It is c that a feature port number is obtained after the General Transformations such as convolution2Feature.It is different with traditional CNN (convolutional neural networks)
, next we operate the feature being previously obtained come recalibration by three.
It is Squeeze operation first, we carry out Feature Compression along Spatial Dimension, by each two-dimensional feature channel
Become a real number, the feature of dimension and input that this real number has global receptive field in a way, and exports is logical
Road number matches.It characterizes the global distribution responded on feature channel, and the layer close to input can also be obtained
Global receptive field, this point is all highly useful in many tasks.
Followed by Excitation operation, it is the mechanism for being similar to door in Recognition with Recurrent Neural Network.By parameter w come
Weight is generated for each feature channel, wherein parameter w is learnt correlation for explicitly Modelling feature interchannel.
Finally the operation of Reweight (again weight), we by the weight of the output of Excitation regard as into
Then the importance in each feature channel after crossing feature selecting is completed by multiplication by channel weighting to previous feature
The recalibration to primitive character on channel dimension.Fig. 3 c is the structure example being embedded into SE-ResNet in ResNet.Side
Dimensional information beside frame represents the output of this layer.Here we use global pooling (pond Quan Pingjun) as
Squeeze operation.And then two FC (Fully Connected, complete to connect) layers form Bottleneck (bottleneck) knot
Structure goes the correlation of modeling interchannel, and exports and the same number of weight of input feature vector.We first reduce characteristic dimension
To the 1/16 of input, then pass through one again after ReLu (Rectified Linear Unit corrects linear unit) activation
Connected layers of Fully rise and return to original dimension.It does so than directly being existed with one Connected layers of Fully of benefit
In: 1) have more non-linear, can preferably be fitted the correlation of interchannel complexity；2) considerably reduce parameter amount and
Calculation amount.Then normalized weight between 0~1 is obtained by the door of a Sigmoid, finally by a Scale (contracting
Put) operation will normalize after Weight to the feature in each channel on.In module basis above, according to
The overall structure of ResNet is overlapped, available SE-ResNet.
Step 203, the defect information based on defect classification and defect profile position output image.
In the present embodiment, after step 202, each image can be obtained the defect classification of at least one defect and lack
Fall into outline position.After the corresponding defect classification of each image of same object and defect profile position grouping are analyzed, it can obtain
To the defect information of the object.Defect information may include at least one of following: defect classification and defect profile position, defect number
Amount, defect area, Level.It can be also that the object carries out quality grading according to the area of defect.Then according to different matter
Magnitude is not handled, is made again for example, primary defect is directly destroyed, three-level defect is recyclable.
Here output can be output in the display being connected with server, can also pass through voice prompting staff
There are defective products.Or defect information directly is sent to the device that mechanical arm etc. is used to screen substandard products.Mechanical arm
Defective object can be picked and, or even different recovery areas can be put into according to defect type, defect area, Level
Domain.
The method provided by the above embodiment of the disclosure passes through lacking the image input training in advance of object to be detected
Fall into detection model, obtained defect classification and position.Improve the efficiency and accuracy of object appearance quality testing.And defect
Information can carry out secondary use excavation, improve the quality of product and reduce production cost.
With further reference to Fig. 4, it illustrates according to one embodiment of the defects detection model training method of the application
Process 400.The process 400 of the defects detection model training method, comprising the following steps:
Step 401, training sample set is obtained.
In the present embodiment, electronic equipment (such as the clothes shown in FIG. 1 of defects detection model training method operation thereon
Business) available training sample set, wherein training sample includes the sample image of object to be detected and corresponding with sample image
Defect type and defect profile position.The defects of training sample type and defect profile position can be to be marked by hand.
Step 402, the sample image for each training sample that training sample is concentrated is sequentially input to initialization defect inspection
Model is surveyed, prediction defect type and defect profile position corresponding to each sample image are obtained.
In the present embodiment, the sample image based on object to be detected acquired in step 401, electronic equipment can incite somebody to action
The sample image for each training sample that training sample is concentrated is sequentially input to initialization defects detection model, to obtain each
Prediction defect type and defect profile position corresponding to sample image.Here, electronic equipment can by each sample image from
The input side input of defects detection model is initialized, successively by the place of the parameter of each layer in initialization defects detection model
Reason, and from the outlet side output of initialization defects detection model, the information of outlet side output is corresponding to the sample image
Predict defect type and defect profile position.Wherein, initial imperfection detection model can be unbred defects detection model
Or the defects detection model that training is not completed, each layer are provided with initiation parameter, initiation parameter is in defects detection model
It can be continuously adjusted in training process.Initialize defects detection model can using VGG, ResNet, SE-ResNet,
Mask RCNN even depth convolutional neural networks structure.
Step 403, by training sample concentrate each training sample in sample image corresponding to defect type and lack
It falls into outline position to be compared with the prediction defect type of the sample image and defect profile position, obtains initialization defects detection
The predictablity rate of model.
In the present embodiment, the prediction defect type based on the obtained sample image of step 402, electronic equipment can incite somebody to action
It the prediction defect type of defect type corresponding to sample image and defect profile position and sample image and is lacked in training sample
Sunken outline position is compared, to obtain the predictablity rate of initialization defects detection model.Specifically, if a trained sample
The prediction defect type and defect wheel of defect type and defect profile position and the sample image corresponding to sample image in this
Wide position is same or similar, then it is correct to initialize defects detection model prediction；If in a training sample corresponding to sample image
Defect type and defect profile position it is different or not close from the prediction defect type of the sample image and defect profile position,
Then initialize defects detection model prediction mistake.Here, electronic equipment can calculate the correct number of prediction and total sample number
Ratio, and the predictablity rate as initialization defects detection model.
In some optional implementations of the present embodiment, the penalty values of defects detection model include Classification Loss value,
Return the penalty values of the prediction of penalty values and example segmentation network.Classification Loss value, recurrence penalty values are to respectively correspond
Softmax, smooth L1 calculate loss function, and example divides the penalty values of the prediction of network.For each ROI, mask points
The matrix of mono- K*m*2 dimension of Zhi Dingyi indicates that K different classification for the region of each m*m, there is each class
One.All it is to be carried out seeking relative entropy with sigmod function for each pixel, obtains average relative entropy error Lmask (example
Divide the penalty values of the prediction of network).Each ROI is just only used if detection obtains which classification ROI belongs to
The relative entropy error of which branch is calculated as error amount.(for example: classification has 3 classes (cat, dog, people), detects
Belong to " people " this kind to current ROI, then used Lmask is the mask of " people " this branch.) such definition so that
Our network does not need to distinguish which kind of each pixel belongs to, it is only necessary to go to distinguish the different difference in this class
Group.Finally two-value mask can be exported by making comparisons with threshold value 0.5.This avoid the competitions between class, by appointing for classification
The classification branch of profession is given in business.And Lmask intersects entropy loss using the sigmoid of two-value for each pixel.
Step 404, determine whether predictablity rate is greater than default accuracy rate threshold value.
In the present embodiment, the predictablity rate based on the obtained initialization defects detection model of step 403, electronics are set
It is standby to be compared the predictablity rate for initializing defects detection model with default accuracy rate threshold value, if more than default accurate
Threshold value is spent, thens follow the steps 405.If more than default accuracy threshold value, 406 are thened follow the steps.
Step 405, the defects detection model completed defects detection model is initialized as training.
In the present embodiment, the case where the prediction accuracy for initializing defects detection model is greater than default accuracy threshold value
Under, illustrate that the defects detection model training is completed, at this point, electronic equipment can will initialization defects detection model as having trained
At defects detection model.
Step 406, the parameter of adjustment initialization defects detection model.
In the present embodiment, the feelings of default accuracy threshold value are not more than in the prediction accuracy of initialization defects detection model
Under condition, the parameter of the adjustable initialization defects detection model of electronic equipment, and 402 are returned to step, until training energy
Enough characterize the defect inspection of the corresponding relationship between the sample image of object to be detected and defect type corresponding with sample image
Until surveying model.
With further reference to Fig. 5, as the realization to method shown in above-mentioned each figure, present disclose provides one kind for exporting letter
One embodiment of the device of breath, the Installation practice is corresponding with embodiment of the method shown in Fig. 2, which can specifically answer
For in various electronic equipments.
As shown in figure 5, the device 500 for output information of the present embodiment includes: receiving unit 501, quality inspection unit 502
With processing unit 503.Wherein, receiving unit 501 are configured in response to receive detection request, obtain object to be detected
Image；Quality inspection unit 502 is configured to input image the defects detection mould based on Mask RCNN algorithm of training in advance
Type obtains the corresponding defect classification of image and defect profile position；Processing unit 503 is configured to based on defect classification and lacks
Fall into the defect information of outline position output image.
In the present embodiment, for the receiving unit 501 of the device of output information 500, quality inspection unit 502 and processing unit
503 specific processing can be with reference to step 201, the step 202, step 203 in Fig. 2 corresponding embodiment.
In some optional implementations of the present embodiment, receiving unit is further configured to: according to it is following at least
One parameter obtains at least one image of object to be detected: angle, filter, times mirror, focuses light.
In some optional implementations of the present embodiment, defects detection model includes Faster RCNN and example point
Network is cut, Faster RCNN is using SE-ResNet as basic network.
In some optional implementations of the present embodiment, device 500 further includes training unit 504, is configured to: obtaining
Take training sample set, training sample include the sample image of object to be detected, defect classification corresponding with sample image and
Defect profile position；Using the sample image of the training sample in training sample set as input, by the sample image with input
As output, training obtains defects detection model for corresponding defect classification and defect profile position.
In some optional implementations of the present embodiment, the penalty values of defects detection model include Classification Loss value,
Return the penalty values of the prediction of penalty values and example segmentation network.
Below with reference to Fig. 6, it illustrates the electronic equipment that is suitable for being used to realize embodiment of the disclosure, (example is as shown in figure 1
Server) 600 structural schematic diagram.Server shown in Fig. 6 is only an example, should not be to the function of embodiment of the disclosure
Any restrictions can be brought with use scope.
As shown in fig. 6, electronic equipment 600 may include processing unit (such as central processing unit, graphics processor etc.)
601, random access can be loaded into according to the program being stored in read-only memory (ROM) 602 or from storage device 608
Program in memory (RAM) 603 and execute various movements appropriate and processing.In RAM 603, it is also stored with electronic equipment
Various programs and data needed for 600 operations.Processing unit 601, ROM 602 and RAM603 are connected with each other by bus 604.
Input/output (I/O) interface 605 is also connected to bus 604.
In general, following device can connect to I/O interface 605: including such as touch screen, touch tablet, keyboard, mouse, taking the photograph
As the input unit 606 of head, microphone, accelerometer, gyroscope etc.；Including such as liquid crystal display (LCD), loudspeaker, vibration
The output device 607 of dynamic device etc.；Storage device 608 including such as tape, hard disk etc.；And communication device 609.Communication device
609, which can permit electronic equipment 600, is wirelessly or non-wirelessly communicated with other equipment to exchange data.Although Fig. 6 shows tool
There is the electronic equipment 600 of various devices, it should be understood that being not required for implementing or having all devices shown.It can be with
Alternatively implement or have more or fewer devices.Each box shown in Fig. 6 can represent a device, can also root
According to needing to represent multiple devices.
Particularly, in accordance with an embodiment of the present disclosure, it may be implemented as computer above with reference to the process of flow chart description
Software program.For example, embodiment of the disclosure includes a kind of computer program product comprising be carried on computer-readable medium
On computer program, which includes the program code for method shown in execution flow chart.In such reality
It applies in example, which can be downloaded and installed from network by communication device 609, or from storage device 608
It is mounted, or is mounted from ROM 602.When the computer program is executed by processing unit 601, the implementation of the disclosure is executed
The above-mentioned function of being limited in the method for example.It should be noted that computer-readable medium described in embodiment of the disclosure can be with
It is computer-readable signal media or computer readable storage medium either the two any combination.It is computer-readable
Storage medium for example may be-but not limited to-the system of electricity, magnetic, optical, electromagnetic, infrared ray or semiconductor, device or
Device, or any above combination.The more specific example of computer readable storage medium can include but is not limited to: have
The electrical connection of one or more conducting wires, portable computer diskette, hard disk, random access storage device (RAM), read-only memory
(ROM), erasable programmable read only memory (EPROM or flash memory), optical fiber, portable compact disc read-only memory (CD-
ROM), light storage device, magnetic memory device or above-mentioned any appropriate combination.In embodiment of the disclosure, computer
Readable storage medium storing program for executing can be any tangible medium for including or store program, which can be commanded execution system, device
Either device use or in connection.And in embodiment of the disclosure, computer-readable signal media may include
In a base band or as the data-signal that carrier wave a part is propagated, wherein carrying computer-readable program code.It is this
The data-signal of propagation can take various forms, including but not limited to electromagnetic signal, optical signal or above-mentioned any appropriate
Combination.Computer-readable signal media can also be any computer-readable medium other than computer readable storage medium, should
Computer-readable signal media can send, propagate or transmit for by instruction execution system, device or device use or
Person's program in connection.The program code for including on computer-readable medium can transmit with any suitable medium,
Including but not limited to: electric wire, optical cable, RF (radio frequency) etc. or above-mentioned any appropriate combination.
Above-mentioned computer-readable medium can be included in above-mentioned electronic equipment；It is also possible to individualism, and not
It is fitted into the electronic equipment.Above-mentioned computer-readable medium carries one or more program, when said one or more
When a program is executed by the electronic equipment, so that the electronic equipment: obtaining the image of object to be detected；Image input is preparatory
The trained defects detection model based on Mask RCNN algorithm obtains the corresponding defect classification of image and defect profile position；Base
In the defect information of defect classification and defect profile position output described image.
The behaviour for executing embodiment of the disclosure can be write with one or more programming languages or combinations thereof
The computer program code of work, described program design language include object oriented program language-such as Java,
Smalltalk, C++ further include conventional procedural programming language-such as " C " language or similar program design language
Speech.Program code can be executed fully on the user computer, partly be executed on the user computer, as an independence
Software package execute, part on the user computer part execute on the remote computer or completely in remote computer or
It is executed on server.In situations involving remote computers, remote computer can pass through the network of any kind --- packet
It includes local area network (LAN) or wide area network (WAN)-is connected to subscriber computer, or, it may be connected to outer computer (such as benefit
It is connected with ISP by internet).
Flow chart and block diagram in attached drawing are illustrated according to the system of the various embodiments of the disclosure, method and computer journey
The architecture, function and operation in the cards of sequence product.In this regard, each box in flowchart or block diagram can generation
A part of one module, program segment or code of table, a part of the module, program segment or code include one or more use
The executable instruction of the logic function as defined in realizing.It should also be noted that in some implementations as replacements, being marked in box
The function of note can also occur in a different order than that indicated in the drawings.For example, two boxes succeedingly indicated are actually
It can be basically executed in parallel, they can also be executed in the opposite order sometimes, and this depends on the function involved.Also it to infuse
Meaning, the combination of each box in block diagram and or flow chart and the box in block diagram and or flow chart can be with holding
The dedicated hardware based system of functions or operations as defined in row is realized, or can use specialized hardware and computer instruction
Combination realize.
Being described in unit involved in embodiment of the disclosure can be realized by way of software, can also be passed through
The mode of hardware is realized.Described unit also can be set in the processor, for example, can be described as: a kind of processor
Including receiving unit, quality inspection unit and processing unit.Wherein, the title of these units is not constituted under certain conditions to the list
The restriction of member itself, for example, receiving unit is also described as " in response to receiving detection request, obtaining object to be detected
The unit of the image of body ".
Above description is only the preferred embodiment of the disclosure and the explanation to institute's application technology principle.Those skilled in the art
Member is it should be appreciated that invention scope involved in the disclosure, however it is not limited to technology made of the specific combination of above-mentioned technical characteristic
Scheme, while should also cover in the case where not departing from the inventive concept, it is carried out by above-mentioned technical characteristic or its equivalent feature
Any combination and the other technical solutions formed.Such as features described above has similar function with (but being not limited to) disclosed in the disclosure
Can technical characteristic replaced mutually and the technical solution that is formed.
Claims (12)
1. a kind of method for output information, comprising:
In response to receiving detection request, the image of object to be detected is obtained；
By the defects detection model based on Mask RCNN algorithm of described image input training in advance, it is corresponding to obtain described image
Defect classification and defect profile position；
Defect information based on the defect classification and defect profile position output described image.
2. according to the method described in claim 1, wherein, the image for obtaining object to be detected, comprising:
At least one image of object to be detected is obtained according at least one of following parameter:
Angle, filter, times mirror, focuses light.
3. according to the method described in claim 1, wherein, the defects detection model includes Faster RCNN and example segmentation
Network, Faster RCNN is using SE-ResNet as basic network.
4. method described in one of -3 according to claim 1, wherein the defects detection model is trained as follows
It arrives:
Training sample set is obtained, training sample includes the sample image of object to be detected, defect corresponding with sample image
Classification and defect profile position；
It, will be corresponding with the sample image of input using the sample image of the training sample in the training sample set as input
As output, training obtains the defects detection model for defect classification and defect profile position.
5. according to the method described in claim 4, wherein, the penalty values of the defects detection model include Classification Loss value, return
Penalty values and example are returned to divide the penalty values of the prediction of network.
6. a kind of device for output information, comprising:
Receiving unit is configured in response to receive detection request, obtains the image of object to be detected；
Quality inspection unit is configured to input described image the defects detection model based on Mask RCNN algorithm of training in advance,
Obtain the corresponding defect classification of described image and defect profile position；
Processing unit is configured to the defect letter based on the defect classification and defect profile position output described image
Breath.
7. device according to claim 6, wherein the receiving unit is further configured to:
At least one image of object to be detected is obtained according at least one of following parameter:
Angle, filter, times mirror, focuses light.
8. device according to claim 6, wherein the defects detection model includes Faster RCNN and example segmentation
Network, Faster RCNN is using SE-ResNet as basic network.
9. the device according to one of claim 6-8, wherein described device further includes training unit, is configured to:
Training sample set is obtained, training sample includes the sample image of object to be detected, defect corresponding with sample image
Classification and defect profile position；
It, will be corresponding with the sample image of input using the sample image of the training sample in the training sample set as input
As output, training obtains the defects detection model for defect classification and defect profile position.
10. device according to claim 9, wherein the penalty values of the defects detection model include Classification Loss value, return
Penalty values and example are returned to divide the penalty values of the prediction of network.
11. a kind of electronic equipment, comprising:
One or more processors；
Storage device is stored thereon with one or more programs,
When one or more of programs are executed by one or more of processors, so that one or more of processors are real
Now such as method as claimed in any one of claims 1 to 5.
12. a kind of computer-readable medium, is stored thereon with computer program, wherein real when described program is executed by processor
Now such as method as claimed in any one of claims 1 to 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910248418.7A CN109993734A (en) | 2019-03-29 | 2019-03-29 | Method and apparatus for output information |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910248418.7A CN109993734A (en) | 2019-03-29 | 2019-03-29 | Method and apparatus for output information |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109993734A true CN109993734A (en) | 2019-07-09 |
Family
ID=67131767
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910248418.7A Pending CN109993734A (en) | 2019-03-29 | 2019-03-29 | Method and apparatus for output information |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109993734A (en) |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110992318A (en) * | 2019-11-19 | 2020-04-10 | 上海交通大学 | Special metal flaw detection system based on deep learning |
| CN111060518A (en) * | 2019-12-20 | 2020-04-24 | 重庆大学 | Stamping part defect identification method based on instance segmentation |
| CN111179261A (en) * | 2019-12-31 | 2020-05-19 | 深圳市太赫兹科技创新研究院 | Defect detection method, system, terminal device and storage medium |
| CN111288890A (en) * | 2020-02-13 | 2020-06-16 | 福建农林大学 | Road sign dimension and height automatic measurement method based on binocular photogrammetry technology |
| CN111325713A (en) * | 2020-01-21 | 2020-06-23 | 浙江省北大信息技术高等研究院 | Wood defect detection method, system and storage medium based on neural network |
| CN111415106A (en) * | 2020-04-29 | 2020-07-14 | 上海东普信息科技有限公司 | Truck loading rate identification method, device, equipment and storage medium |
| CN111415330A (en) * | 2020-02-27 | 2020-07-14 | 苏州杰锐思智能科技股份有限公司 | Copper foil appearance defect detection method based on deep learning |
| CN111428815A (en) * | 2020-04-16 | 2020-07-17 | 重庆理工大学 | Mechanical arm grabbing detection method based on Anchor angle mechanism |
| CN111652852A (en) * | 2020-05-08 | 2020-09-11 | 浙江华睿科技有限公司 | Method, device and equipment for detecting surface defects of product |
| CN111753692A (en) * | 2020-06-15 | 2020-10-09 | 珠海格力电器股份有限公司 | Target object extraction method, product detection method, device, computer and medium |
| CN112014399A (en) * | 2020-07-22 | 2020-12-01 | 湖北工业大学 | Belt type grain breakage rate and impurity content rate detection device and method in grain bin |
| CN112396575A (en) * | 2019-07-31 | 2021-02-23 | 由田新技股份有限公司 | Image-based classification system |
| CN112611757A (en) * | 2020-12-23 | 2021-04-06 | 深兰人工智能芯片研究院（江苏）有限公司 | AI auxiliary detection method, device, electronic equipment and computer readable storage medium |
| CN112734671A (en) * | 2021-01-11 | 2021-04-30 | 哈尔滨市科佳通用机电股份有限公司 | Vehicle image deformation repairing method based on deep learning |
| CN113469944A (en) * | 2021-06-03 | 2021-10-01 | 厦门宏泰智能制造有限公司 | Product quality inspection method and device and electronic equipment |
| WO2021232611A1 (en) * | 2020-05-22 | 2021-11-25 | 五邑大学 | Wine bottle surface defect model evaluation method, electronic device, and storage medium |
| EP4153533A4 (en) * | 2020-05-22 | 2023-12-20 | Fraunhofer USA, Inc. | Systems and methods for synthesizing a diamond using machine learning |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108846841A (en) * | 2018-07-02 | 2018-11-20 | 北京百度网讯科技有限公司 | Display screen quality determining method, device, electronic equipment and storage medium |
| CN108921839A (en) * | 2018-07-02 | 2018-11-30 | 北京百度网讯科技有限公司 | Continuous casting billet quality detection method, device, electronic equipment and storage medium |
| CN109086780A (en) * | 2018-08-10 | 2018-12-25 | 北京百度网讯科技有限公司 | Method and apparatus for detecting electrode piece burr |
-
2019
- 2019-03-29 CN CN201910248418.7A patent/CN109993734A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108846841A (en) * | 2018-07-02 | 2018-11-20 | 北京百度网讯科技有限公司 | Display screen quality determining method, device, electronic equipment and storage medium |
| CN108921839A (en) * | 2018-07-02 | 2018-11-30 | 北京百度网讯科技有限公司 | Continuous casting billet quality detection method, device, electronic equipment and storage medium |
| CN109086780A (en) * | 2018-08-10 | 2018-12-25 | 北京百度网讯科技有限公司 | Method and apparatus for detecting electrode piece burr |
Non-Patent Citations (3)
| Title |
|---|
| 李月景 编著: "《图像识别技术及其应用》", 31 March 1985, 北京：机械工业出版社 * |
| 陈百明: "《铸造工艺及设计》", 30 November 2016, 北京：北京理工大学出版社 * |
| 陈禄如 等: "《建筑钢结构施工手册》", 31 May 2002, 北京：中国计划出版社 * |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112396575A (en) * | 2019-07-31 | 2021-02-23 | 由田新技股份有限公司 | Image-based classification system |
| CN110992318A (en) * | 2019-11-19 | 2020-04-10 | 上海交通大学 | Special metal flaw detection system based on deep learning |
| CN111060518A (en) * | 2019-12-20 | 2020-04-24 | 重庆大学 | Stamping part defect identification method based on instance segmentation |
| CN111179261A (en) * | 2019-12-31 | 2020-05-19 | 深圳市太赫兹科技创新研究院 | Defect detection method, system, terminal device and storage medium |
| CN111325713A (en) * | 2020-01-21 | 2020-06-23 | 浙江省北大信息技术高等研究院 | Wood defect detection method, system and storage medium based on neural network |
| CN111288890A (en) * | 2020-02-13 | 2020-06-16 | 福建农林大学 | Road sign dimension and height automatic measurement method based on binocular photogrammetry technology |
| CN111415330A (en) * | 2020-02-27 | 2020-07-14 | 苏州杰锐思智能科技股份有限公司 | Copper foil appearance defect detection method based on deep learning |
| CN111428815A (en) * | 2020-04-16 | 2020-07-17 | 重庆理工大学 | Mechanical arm grabbing detection method based on Anchor angle mechanism |
| CN111415106A (en) * | 2020-04-29 | 2020-07-14 | 上海东普信息科技有限公司 | Truck loading rate identification method, device, equipment and storage medium |
| CN111652852A (en) * | 2020-05-08 | 2020-09-11 | 浙江华睿科技有限公司 | Method, device and equipment for detecting surface defects of product |
| CN111652852B (en) * | 2020-05-08 | 2024-03-29 | 浙江华睿科技股份有限公司 | Product surface defect detection method, device and equipment |
| WO2021232611A1 (en) * | 2020-05-22 | 2021-11-25 | 五邑大学 | Wine bottle surface defect model evaluation method, electronic device, and storage medium |
| EP4153533A4 (en) * | 2020-05-22 | 2023-12-20 | Fraunhofer USA, Inc. | Systems and methods for synthesizing a diamond using machine learning |
| CN111753692A (en) * | 2020-06-15 | 2020-10-09 | 珠海格力电器股份有限公司 | Target object extraction method, product detection method, device, computer and medium |
| CN112014399A (en) * | 2020-07-22 | 2020-12-01 | 湖北工业大学 | Belt type grain breakage rate and impurity content rate detection device and method in grain bin |
| CN112014399B (en) * | 2020-07-22 | 2023-08-25 | 湖北工业大学 | Belt-type grain crushing rate and impurity content detection device and method in grain tank |
| CN112611757A (en) * | 2020-12-23 | 2021-04-06 | 深兰人工智能芯片研究院（江苏）有限公司 | AI auxiliary detection method, device, electronic equipment and computer readable storage medium |
| CN112734671A (en) * | 2021-01-11 | 2021-04-30 | 哈尔滨市科佳通用机电股份有限公司 | Vehicle image deformation repairing method based on deep learning |
| CN113469944A (en) * | 2021-06-03 | 2021-10-01 | 厦门宏泰智能制造有限公司 | Product quality inspection method and device and electronic equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109993734A (en) | Method and apparatus for output information | |
| CN110009614A (en) | Method and apparatus for output information | |
| CN109978870A (en) | Method and apparatus for output information | |
| CN109949286A (en) | Method and apparatus for output information | |
| CN107423398A (en) | Exchange method, device, storage medium and computer equipment | |
| CN111784685A (en) | Power transmission line defect image identification method based on cloud edge cooperative detection | |
| CN109584227A (en) | A kind of quality of welding spot detection method and its realization system based on deep learning algorithm of target detection | |
| CN105574550A (en) | Vehicle identification method and device | |
| CN109741328A (en) | A kind of automobile apparent mass detection method based on production confrontation network | |
| CN108564049A (en) | A kind of fast face detection recognition method based on deep learning | |
| CN109902573A (en) | Multiple-camera towards video monitoring under mine is without mark pedestrian's recognition methods again | |
| CN109767422A (en) | Pipe detection recognition methods, storage medium and robot based on deep learning | |
| CN109145766A (en) | Model training method, device, recognition methods, electronic equipment and storage medium | |
| CN111259779B (en) | Video motion detection method based on center point track prediction | |
| CN109145759A (en) | Vehicle attribute recognition methods, device, server and storage medium | |
| CN108681746A (en) | A kind of image-recognizing method, device, electronic equipment and computer-readable medium | |
| CN109409327B (en) | RRU module object pose detection method based on end-to-end deep neural network | |
| CN109284779A (en) | Object detecting method based on the full convolutional network of depth | |
| CN105210085A (en) | Image labeling using geodesic features | |
| CN109241808A (en) | Two dimensional code auto-screening method in a kind of fashion images | |
| CN109886356A (en) | A kind of target tracking method based on three branch's neural networks | |
| CN115409069A (en) | Village and town building identification method, classification method, device, electronic equipment and medium | |
| CN111028216A (en) | Image scoring method and device, storage medium and electronic equipment | |
| CN110059646A (en) | The method and Target Searching Method of training action plan model | |
| CN109740567A (en) | Key point location model training method, localization method, device and equipment |
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 |