CN207181307U - A kind of stereoscopic image acquisition device - Google Patents
A kind of stereoscopic image acquisition device Download PDFInfo
- Publication number
- CN207181307U CN207181307U CN201720382189.4U CN201720382189U CN207181307U CN 207181307 U CN207181307 U CN 207181307U CN 201720382189 U CN201720382189 U CN 201720382189U CN 207181307 U CN207181307 U CN 207181307U
- Authority
- CN
- China
- Prior art keywords
- module
- image
- signal
- acquisition device
- preset value
- 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.)
- Expired - Fee Related
Links
Landscapes
- Image Analysis (AREA)
Abstract
The utility model discloses a kind of stereoscopic image acquisition device, including CCD camera, image pre-processing module, image segmentation module, feature extraction module, data preprocessing module, neural network learning module and the neutral net detection module being sequentially connected；A kind of stereoscopic image acquisition device of the present utility model, the image of product is obtained using CCD camera, the image of product is handled using controller, wherein the process handled can improve product quality detection efficiency.
Description
Technical field
Automated machine equipment technical field is the utility model is related to, more particularly to a kind of stereo-picture collection
Device.
Background technology
In modern industry, the production of many products (such as electronics and device, hardware device parts) is complete on transfer matic
Into, each production link is directed to the quality testing of product, and some enterprises put into a large amount of manpowers, by the way of naked eyes detect come
Product quality is controlled, but because some human factors, product quality are difficult to ensure that.To improve detection efficiency and ensureing product matter
Amount, the automatic detection that the vision progress product quality of people is replaced using machine vision technique are the effective ways solved the problems, such as.
Machine vision due to can quick obtaining bulk information, and be easy to same design information and control information and integrated, because
In modern manufacturing production process, machine vision is widely used in quality testing, production control etc. for this.Regarded relative to human eye
Feel, machine vision has very big advantage and development prospect, therefore machine vision is developed rapidly in recent years, its extensive use
In every field such as medical treatment, industry, agricultural, military affairs, traffic.
In China, until middle and later periods nineties machine vision technique is just gradually recognized and understood, at present, used in system
Make machine vision in industry and still suffer from the problem of product quality detection efficiency is low, therefore, it is difficult to promote.
Utility model content
In view of the deficienciess of the prior art, the purpose of this utility model is to provide a kind of stereoscopic image acquisition device,
The image of product is obtained using CCD camera, the image of product is handled using controller, wherein the process handled can carry
High product quality testing efficiency.
To achieve the above object, the utility model provides following technical scheme：
A kind of stereoscopic image acquisition device, including be sequentially connected CCD camera, image pre-processing module, feature extraction mould
Block, data preprocessing module and neutral net detection module；
The CCD camera is used for the original image for obtaining product, and sends to described and state image pre-processing module；
Described image pretreatment module is used to correct the original image when the gray level of original image is more than preset value
Gray level, and send corresponding revise signal to the feature extraction module；
The feature extraction module is used to receive the revise signal, and is more than in the gray scale of the revised image
Corresponding characteristic vector signal is sent during preset value to the data preprocessing module；
The data preprocessing module is used to receive the characteristic vector signal, by the characteristic vector signal and preset value
It is compared, and sends corresponding preprocessed signal to the neutral net detection module；
The neutral net detection module is used to classify to product in response to the preprocessed signal.
As a kind of enforceable mode, described image pretreatment module includes comparing unit, and the comparing unit is used for
The gray level of the original image is corrected when the gray level of original image is more than preset value, and sends corresponding revise signal
To the feature extraction module.
As a kind of enforceable mode, the feature extraction module includes comparing unit, and the comparing unit is used to connect
The revise signal is received, and corresponding characteristic vector signal is sent when the gray scale of the revised image is more than preset value
To the data preprocessing module.
As a kind of enforceable mode, the data preprocessing module includes comparing unit, and the comparing unit is used for
The characteristic vector signal is received, the characteristic vector signal and preset value are compared, and sends corresponding pretreatment
Signal is to the neutral net detection module.
As a kind of enforceable mode, the neutral net detection module includes comparing unit, and the comparing unit is used
It is compared in by the preprocessed signal and preset value, and product is classified according to result of the comparison.
The utility model is compared to the beneficial effect of prior art：
The utility model provides a kind of stereoscopic image acquisition device, including CCD camera, image pre-processing module, feature
Abstraction module, data preprocessing module and neutral net detection module；Wherein CCD camera is that image obtains terminal, and other four
One key element and preset value are compared by individual module respectively, for example, image pre-processing module by the gray level of original image and
Preset value is compared；Finally product is classified, so as to improve product quality detection efficiency.
Brief description of the drawings
Fig. 1 is a block diagram of stereoscopic image acquisition device provided by the utility model；
Fig. 2 is another block diagram of stereoscopic image acquisition device provided by the utility model.
In figure：100th, CCD camera；200th, image pre-processing module；300th, image segmentation module；400th, feature extraction mould
Block；500th, data preprocessing module；600th, neural network learning module；700th, neutral net detection module；800th, servounit
Hand.
Embodiment
Below in conjunction with accompanying drawing, and other technical characteristic above-mentioned to the utility model and advantage are carried out clearly and completely
Description, it is clear that described embodiment is only section Example of the present utility model, rather than whole embodiments.
Reference picture 2, the utility model provide a kind of stereoscopic image acquisition device, and it includes the CCD camera being sequentially connected
100th, image pre-processing module 200, image segmentation module 300, feature extraction module 400, data preprocessing module 500, nerve
E-learning module 600 and neutral net detection module 700；It also includes servo manipulator 800.
Wherein, CCD camera 100 is used for the original image for obtaining product, and sends to image pre-processing module 200；Image
Pretreatment module 200 is used for the gray level that original image is corrected when the gray level of original image is more than preset value, and sends
Corresponding revise signal is to feature extraction module 400；Feature extraction module 400 is used to receive revise signal, and after amendment
The gray scale of image send corresponding characteristic vector signal to data preprocessing module 500 when being more than preset value；Data prediction
Module 500 is used to receive characteristic vector signal, and characteristic vector signal and preset value are compared, and sends corresponding pre- place
Signal is managed to neutral net detection module 700；Neutral net detection module 700 is used to carry out product in response to preprocessed signal
Classification.
Here, image pre-processing module 200, feature extraction module 400, data preprocessing module 500 and neutral net
Detection module 700 has a key element compared with preset value respectively, is the gray level of original image, revised image respectively
Gray scale, characteristic vector signal and preprocessed signal.Wherein, any link in this four processes, be all by one it is default will
Then element exports a result of the comparison with preset value compared with, principle therein is similar to voltage comparator, by voltage and
Preset value is compared and exports a result of the comparison, belongs to prior art.
Reference picture 1, in another embodiment, image pre-processing module 200 are used for the original got to CCD camera 100
Beginning image carries out gray level correcting process and noise smoothing processing, and sends to image segmentation module 300；Image segmentation module
300 are used to split the image after amendment, smoothing processing, and send to feature extraction module 400；Feature extraction module
400 are used to measure the gray scale of the image after segmentation, produce corresponding characteristic vector, and send to data preprocessing module
500；Between data preprocessing module 500 by characteristic vector for normalizing to 0 to 1, and input to neural network learning module
600；Neural network learning module 600 is used to obtain the characteristic vector after normalization according to original image, and generates corresponding defeated
Enter network mode；Neutral net detection module 700 is used to input network mode is identified and classified, to distinguish qualified products
And substandard product.Here, image pre-processing module 200 uses and the gray level amendment under environmental light intensity difference and noise is put down
Sliding processing, improves the grey-scale contrast of image, realizes the matching of detection image and template image；Image segmentation module 300 uses
Threshold sementation, threshold segmentation is carried out by defining to system area-of-interest AOI, makes quality testing region
It is more targeted；Feature extraction module 400 defines respective algorithms by product quality defect species and extracts image feature vector,
Improve product quality detection efficiency.
In addition, according to the characteristic vector after normalizing between 0 to 1, it can be determined that go out the current shooting of CCD camera 100
Angle, therefore servo manipulator 800 and data preprocessing module 500 can be connected, it is being located further forward servo manipulator 800
Module in obtain correlation shooting angle information, to adjust the shooting angle of CCD camera 100.
Hereinafter, to image pre-processing module 200, image segmentation module 300, feature extraction module 400, data prediction mould
Block 500, neural network learning module 600 and neutral net detection module 700 illustrate one by one.
Stereoscopic image acquisition device provided by the utility model, the image preprocessing gray level amendment of use, smooth noise
Method image is pre-processed.This process is realized by image pre-processing module 200.
Stereoscopic image acquisition device provided by the utility model, image is split using thresholding method, first root
According to the appropriate gray level thresholding (threshold value) of the Feature Selection one of detected product, then by each pixel ash in product image
Degree and it be compared, more than redistributing with maximum gray scale (255) for thresholding, less than thresholding distribution with minimal gray (0),
A new bianry image can be thus formed, and successfully object is revealed from background.This process is by image point
Module 300 is cut to realize.
Stereoscopic image acquisition device provided by the utility model, gradation of image information is carried out on the basis of image segmentation
Measurement, one group of feature is produced, these combinations of features together, is formed characteristic vector, the binary map of examined product image
And the information content that artwork includes is generally very big, it is impossible to directly provides it to BP neural network and judges to detect, it is therefore necessary to from
Some features are extracted in binary map and artwork, system, which defines, is referred to as the rectangular area for the scope being likely to occur comprising quality problems
Area-of-interest AOI (area of interest, AOI), extracts background light level value, the size of AOI hot spots, in AOI most
Hot spot forms the input feature value of BP neural network from 4 features with a distance from bias light in high-gray level value and AOI.This process
Realized by feature extraction module 400.
Stereoscopic image acquisition device provided by the utility model, the data that characteristic extracting module obtains use S function f (x)
=1/ (1+e-x) be normalized between 0 to 1, to input Processing with Neural Network.This process is by data preprocessing module
500 realize.
Stereoscopic image acquisition device provided by the utility model, product quality defect species is more, true defect and Artifact
It is not easy to distinguish.Therefore Feature Selection need to be depending on specific requirement.Quality inspection personnel is according to the species of defect such as in glass production
Needed with product, glass blocks (product) usually is divided into two stages is checked.First stage is to single glass defect
Detection, its target be find glass the defects of, including：Bubble, be mingled with, light distortion, viscous tin, scuffing, drawing lines.Second stage
It is that monolithic glass is classified on the basis of detecting in the first stage.Therefore the detection of individual defect is the basis of classification classification.Matter
Inspection person extremely paid close attention to the size of individual defect, is generally represented with major diameter.It is maximum axial distance for bubble, it is right
It is irregular be mingled with for be point-to-point transmission in heart defect curl ultimate range, and line length is then referred to linear discontinuities.Inspection
Major diameter must be calculated during survey.The neural network learning of software first has to determine the optimum structure of BP neural network.Wherein,
Input layer number depends on the dimension of input feature value, and input feature value is tieed up for 4 in the system, institute's input layer number
For 4.Output layer nodes can typically be equal to pattern class number, it is also possible to each pattern class of coded representation of output node.By
In typically only two kinds of the quality problems type of product：It is qualified with it is unqualified.Accordingly, it can be determined that output layer unit number is 1, its is defeated
Go out for 0 when represent product it is qualified, export for 1 when represent product it is unqualified.It is requirement of the hidden layer node number typically with problem, defeated
Enter the number of output unit and number of training have a direct relation, the system use node in hidden layer evaluation method for：
Wherein, NHFor optimal node in hidden layer；NIFor input layer number；NOFor output layer nodes；NPTo train sample
This number.
The system uses additional guide vanes, and in the change of each weights and threshold value plus the next item up is proportional to previous change
The value of amount, and new weights and threshold value are produced according to back propagation, BP neural network can be avoided to be fallen into learning process
In local minimum, accelerate pace of learning.
Δ W (k+1)=(1-mc) α+mc Δ W (k)；
Δ θ (k+1)=(1-mc) d+mc Δ θ (k)；
Wherein, k is frequency of training；Mc is factor of momentum, typically takes 0.95 or so.
In the case where given accuracy requires MSE≤104, BP neural network restrains to 28 samples of selection by 503 study
Afterwards, obtained weights and threshold value write-in file are saved, this is the learning outcome of network.
System is by being handled the image of detected product to obtain characteristic vector, and this vector is as input network
New model.This process is realized by neural network learning module 600.
Stereoscopic image acquisition device provided by the utility model, neutral net detection are exactly by calling BP neural network
Learning outcome to the new model for inputting network is identified and classified, i.e., product characteristics of image is detected, and export inspection
Result is surveyed, is finally handled accordingly by execution machine to being detected vial.
Particular embodiments described above, the purpose of this utility model, technical scheme and beneficial effect are carried out to enter one
The detailed description of step, it will be appreciated that the foregoing is only specific embodiment of the utility model, be not used to limit this reality
With new protection domain.Particularly point out, to those skilled in the art, it is all the spirit and principles of the utility model it
It is interior, any modification, equivalent substitution and improvements done etc., it should be included within the scope of protection of the utility model.
Claims (5)
1. a kind of stereoscopic image acquisition device, it is characterised in that including the CCD camera being sequentially connected（100）, image preprocessing mould
Block（200）, feature extraction module（400）, data preprocessing module（500）And neutral net detection module（700）；
The CCD camera（100）For obtaining the original image of product, and send to described image pretreatment module（200）；
Described image pretreatment module（200）For correcting the original image when the gray level of original image is more than preset value
Gray level, and send corresponding revise signal to the feature extraction module（400）；
The feature extraction module（400）For receiving the revise signal, and it is big in the gray scale of the revised image
Corresponding characteristic vector signal is sent when preset value to the data preprocessing module（500）；
The data preprocessing module（500）For receiving the characteristic vector signal, by the characteristic vector signal and preset
Value is compared, and sends corresponding preprocessed signal to the neutral net detection module（700）；
The neutral net detection module（700）For classifying in response to the preprocessed signal to product.
2. stereoscopic image acquisition device according to claim 1, it is characterised in that described image pretreatment module（200）
Including comparing unit, the comparing unit is used to correct the original image when the gray level of original image is more than preset value
Gray level, and corresponding revise signal is sent to the feature extraction module（400）.
3. stereoscopic image acquisition device according to claim 1, it is characterised in that the feature extraction module（400）Bag
Comparing unit is included, the comparing unit is used to receive the revise signal, and is more than in the gray scale of the revised image
Corresponding characteristic vector signal is sent during preset value to the data preprocessing module（500）.
4. stereoscopic image acquisition device according to claim 1, it is characterised in that the data preprocessing module（500）
Including comparing unit, the comparing unit is used to receive the characteristic vector signal, by the characteristic vector signal and preset value
It is compared, and sends corresponding preprocessed signal to the neutral net detection module（700）.
5. stereoscopic image acquisition device according to claim 1, it is characterised in that the neutral net detection module
（700）Including comparing unit, the comparing unit be used for the preprocessed signal and preset value are compared, and according to than
Compared with result product is classified.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201720382189.4U CN207181307U (en) | 2017-04-12 | 2017-04-12 | A kind of stereoscopic image acquisition device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201720382189.4U CN207181307U (en) | 2017-04-12 | 2017-04-12 | A kind of stereoscopic image acquisition device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN207181307U true CN207181307U (en) | 2018-04-03 |
Family
ID=61733410
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201720382189.4U Expired - Fee Related CN207181307U (en) | 2017-04-12 | 2017-04-12 | A kind of stereoscopic image acquisition device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN207181307U (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111251296A (en) * | 2020-01-17 | 2020-06-09 | 温州职业技术学院 | Visual detection system suitable for pile up neatly electric motor rotor |
-
2017
- 2017-04-12 CN CN201720382189.4U patent/CN207181307U/en not_active Expired - Fee Related
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111251296A (en) * | 2020-01-17 | 2020-06-09 | 温州职业技术学院 | Visual detection system suitable for pile up neatly electric motor rotor |
| CN111251296B (en) * | 2020-01-17 | 2021-05-18 | 温州职业技术学院 | Visual detection system suitable for pile up neatly electric motor rotor |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113160192B (en) | Visual sense-based snow pressing vehicle appearance defect detection method and device under complex background | |
| CN110148130B (en) | Method and device for detecting part defects | |
| CN107230203B (en) | Casting defect identification method based on human eye visual attention mechanism | |
| CN107564002A (en) | Plastic tube detection method of surface flaw, system and computer-readable recording medium | |
| CN106875381A (en) | A kind of phone housing defect inspection method based on deep learning | |
| Fang et al. | Crop leaf disease grade identification based on an improved convolutional neural network | |
| CN108765412A (en) | A kind of steel strip surface defect sorting technique | |
| CN110148162A (en) | A kind of heterologous image matching method based on composition operators | |
| CN106372622A (en) | Facial expression classification method and device | |
| CN102567744B (en) | Method for determining quality of iris image based on machine learning | |
| CN111815564B (en) | Method and device for detecting silk ingots and silk ingot sorting system | |
| CN102928435A (en) | Aircraft skin damage identification method and device based on image and ultrasound information fusion | |
| CN111539957B (en) | Image sample generation method, system and detection method for target detection | |
| CN108428231A (en) | A kind of multi-parameter Part Surface Roughness learning method based on random forest | |
| CN111415339B (en) | Image defect detection method for complex texture industrial product | |
| CN115082719B (en) | Wood quality grading method | |
| CN104680185B (en) | Hyperspectral image classification method based on boundary point reclassification | |
| CN111161237A (en) | Fruit and vegetable surface quality detection method, storage medium and sorting device thereof | |
| CN109145846A (en) | Material microstructure intelligent recognition analysis system and analysis method | |
| CN106997590A (en) | A kind of image procossing and detecting system based on detection product performance | |
| CN206897873U (en) | A kind of image procossing and detecting system based on detection product performance | |
| CN108664839A (en) | A kind of image processing method and equipment | |
| CN107024480A (en) | A kind of stereoscopic image acquisition device | |
| CN110321890B (en) | Digital instrument identification method of power inspection robot | |
| CN103279944A (en) | Image division method based on biogeography optimization |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20180403 Termination date: 20210412 |