CN110335265A - A kind of network witness management method and device based on artificial intelligence image recognition - Google Patents
A kind of network witness management method and device based on artificial intelligence image recognition Download PDFInfo
- Publication number
- CN110335265A CN110335265A CN201910595628.3A CN201910595628A CN110335265A CN 110335265 A CN110335265 A CN 110335265A CN 201910595628 A CN201910595628 A CN 201910595628A CN 110335265 A CN110335265 A CN 110335265A
- Authority
- CN
- China
- Prior art keywords
- image
- test specimen
- feature library
- camera
- bottom plate
- 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
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Library & Information Science (AREA)
- Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Quality & Reliability (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Databases & Information Systems (AREA)
- Computational Linguistics (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to image identification technical field, in particular to a kind of network witness management method and device based on artificial intelligence image recognition.Test specimen is fixed on specific pure color bottom plate by a kind of network witness management method based on artificial intelligence image recognition comprising steps of being sampled in engineering site to test specimen, and the first image of the test specimen is obtained by the first camera；Test specimen is sent to inspection center, and the second image of test specimen in inspection center is obtained by second camera；Calculate the fisrt feature library of the first image and the second feature library of the second image；The characteristic similarity in fisrt feature library and second feature library is calculated, and judges whether the first image corresponds to the same test specimen with the second image according to characteristic similarity.By this way, the network management for realizing witness sampling work not only simplifies witness operation, and ensure that the authenticity of witness sampling result.
Description
Technical field
The present invention relates to image identification technical field, in particular to a kind of network witness based on artificial intelligence image recognition
Management method and device.
Background technique
It is well known that the superiority and inferiority of building materials quality is the fundamental of construction engineering quality, and building material testing is then
It is the important leverage of building site Material quality control.Therefore, witness sampling and inspection be guarantee examine work it is scientific, it is just,
The superiority and inferiority that quality directly affects building materials quality is carried out in accurate important means, witness sampling work.Witness sampling and inspection
System refers on the basis of contractor unit is by regulation self-test, witnesses in the testing inspection personnel of construction unit, supervisor
Under, it is sampled at the scene by construction personnel, send to specified unit and tested.
In the actual operation process, also generally existing many problems have for example, Attesting Officer needs to accompany to scene
The Attesting Officer of related credentials is very few, it is difficult to the witness sampling at the scene of completing and inspection work；Some supervision departments lack
Effectively management, unit in charge of construction or producer practise fraud, and the authenticity for witnessing work is difficult to ensure.
Summary of the invention
For this reason, it may be necessary to a kind of network witness management method based on artificial intelligence image recognition is provided, it is existing to solve
Engineered building materials evidence obtaining needs to spend a large amount of manpowers, and is unable to ensure the problem of sampling result verifies accuracy.Particular technique
Scheme is as follows:
A kind of network witness management method based on artificial intelligence image recognition, comprising steps of in engineering site to test specimen
It is sampled, test specimen is fixed on specific pure color bottom plate, and obtain the first image of the test specimen by the first camera；Examination
Part is sent to inspection center, and the second image of test specimen in inspection center is obtained by second camera；Calculate the first image
Fisrt feature library and second image second feature library；Calculate the spy in the fisrt feature library Yu the second feature library
Similarity is levied, and judges whether the first image corresponds to the same test specimen with the second image according to the characteristic similarity.
Further, first camera and second camera are zoomable cameras, and be set to turn up and down
On dynamic holder.
Further, the pure color bottom plate is arbitrarily and test specimen is at the color of sharp contrast, and spacing is portrayed on surface
Vertical and horizontal grid lines.
Further, the first image is that the first camera zoom scaling makes the test specimen integral into maximum as falling
In picture area, picture edge is the image of pure color bottom plate；Second image is that the second camera zoom scaling makes
For the test specimen integral into maximum as falling in picture area, picture edge is the image of pure color bottom plate.
Further, described " to calculate the second feature library in the fisrt feature library and second image of the first image；
The characteristic similarity in the fisrt feature library Yu the second feature library is calculated, and the first figure is judged according to the characteristic similarity
As whether corresponding to the same test specimen with the second image ", further comprise the steps of: the circle pixel RGB of edge one for extracting the first image
Value calculates edge pixel mean value, traverses all pixels rgb value of the first image, calculates and the edge pixel mean value
Otherness is classified as baseplate zone, is otherwise classified as test specimen region when the otherness is less than the threshold value of setting；Extract described
The edge one of two images encloses pixel RGB values, calculates edge pixel mean value, traverses all pixels rgb value of the first image, counts
The otherness with the edge pixel mean value is calculated, when the otherness is less than the threshold value of setting, baseplate zone is classified as, otherwise returns
For test specimen region；The related coefficient for calculating the first test specimen area image and the second test specimen area image, judges the related coefficient
Whether reach preset dependent thresholds, if the related coefficient is not up to preset dependent thresholds, determines the first image and
Corresponding two images are not same test specimens；If the related coefficient reaches preset dependent thresholds, to the first test specimen administrative division map
Picture and the second test specimen area image are normalized, and extract to obtain fisrt feature library and by CNN convolutional neural networks
Two feature databases, using the similarity in fisrt feature library described in k nearest neighbor KNN algorithm comparison and the second feature library, as similar spy
When the percentage that sign number accounts for feature database sum reaches preset similar threshold value, then determine that the first image and the second image are corresponding
Be same test specimen.
In order to solve the above technical problems, additionally providing a kind of network witness management dress based on artificial intelligence image recognition
It sets, specific technical solution is as follows:
A kind of network witness managing device based on artificial intelligence image recognition, comprising: the first camera, the first pure color bottom
Plate, second camera, the second pure color bottom plate and processor；The first pure color bottom plate is used for: what fixation was sampled from engineering site
Test specimen；First camera is used for: obtaining the first image of test specimen on the first pure color bottom plate；The second pure color bottom plate
For: it send to the test specimen of inspection center；The second camera is used for: obtaining the second figure of test specimen on the second pure color bottom plate
Picture；The processor is used for: calculating the fisrt feature library of the first image and the second feature library of second image；It calculates
The characteristic similarity in the fisrt feature library and the second feature library, and according to the characteristic similarity judge the first image with
Whether the second image corresponds to the same test specimen.
Further, further includes: the first holder and the second holder；First camera is zoomable camera, setting
In on the first holder that can be rotated up and down；The second camera is zoomable camera, and be set to turn up and down
On the second dynamic holder.
Further, the first pure color bottom plate or the second pure color bottom plate be arbitrarily and test specimen at sharp contrast color,
And the vertical and horizontal grid lines of spacing is portrayed on surface.
Further, the first image is that the first camera zoom scaling makes the test specimen integral into maximum as falling
In picture area, picture edge is the image of pure color bottom plate；Second image is that the second camera zoom scaling makes
For the test specimen integral into maximum as falling in picture area, picture edge is the image of pure color bottom plate.
Further, the processor is also used to: the edge one for extracting the first image encloses pixel RGB values, calculates side
Edge pixel mean value traverses all pixels rgb value of the first image, calculates the otherness with the edge pixel mean value, when
When the otherness is less than the threshold value of setting, it is classified as baseplate zone, is otherwise classified as test specimen region；Extract the side of second image
Edge one encloses pixel RGB values, calculates edge pixel mean value, traverses all pixels rgb value of the first image, calculates and the side
The otherness of edge pixel mean value is classified as baseplate zone, is otherwise classified as test specimen area when the otherness is less than the threshold value of setting
Domain；The related coefficient for calculating the first test specimen area image and the second test specimen area image, judges whether the related coefficient reaches
Preset dependent thresholds determine the first image and the second image pair if the related coefficient is not up to preset dependent thresholds
What is answered is not same test specimen；If the related coefficient reaches preset dependent thresholds, to the first test specimen area image and second
Test specimen area image is normalized, and extracts to obtain fisrt feature library and second feature library by CNN convolutional neural networks,
Using the similarity in fisrt feature library described in k nearest neighbor KNN algorithm comparison and the second feature library, when similar features number accounts for feature
When the percentage of library sum reaches preset similar threshold value, then determine the first image and the second image it is corresponding be same examination
Part.
The beneficial effects of the present invention are: passing through, networking remotely controls the first camera and second camera shoots engineering and shows
The test specimen image of field and inspection center, pass through the fisrt feature library and second image that calculate the first image second are special
Levy library；The characteristic similarity in the fisrt feature library Yu the second feature library is calculated, and is judged according to the characteristic similarity
Whether the first image corresponds to the same test specimen with the second image, completes long-range witness work；Realize the net of witness sampling work
Network management not only simplifies witness operation, and ensure that the authenticity of witness sampling result by image recognition.
Detailed description of the invention
Fig. 1 is a kind of process of the network witness management method based on artificial intelligence image recognition described in specific embodiment
Figure；
Fig. 2 is the flow chart that characteristic similarity is calculated described in specific embodiment；
Fig. 3 is pure color floor diagram described in specific embodiment；
Fig. 4 is a kind of module of the network witness managing device based on artificial intelligence image recognition described in specific embodiment
Schematic diagram.
Description of symbols:
400, the network based on artificial intelligence image recognition witnesses managing device,
401, the first camera,
402, the first pure color bottom plate,
403, second camera,
404, the second pure color bottom plate,
405, processor.
Specific embodiment
Technology contents, construction feature, the objects and the effects for detailed description technical solution, below in conjunction with specific reality
It applies example and attached drawing is cooperated to be explained in detail.
Fig. 1 to Fig. 2 is please referred to, in the present embodiment, a kind of network witness management based on artificial intelligence image recognition
Method can be applicable in a kind of network witness managing device based on artificial intelligence image recognition, described a kind of to be based on artificial intelligence
The network of image recognition witnesses managing device, comprising: the first camera, the first pure color bottom plate, second camera, the second pure color bottom
Plate and processor.Wherein the first camera and the first pure color bottom plate are set to engineering site, second camera and the second pure color bottom
Plate is set to inspection center.First camera and second camera are zoomable cameras, and being respectively arranged at can be up and down
On the first holder and the second holder of rotation, such structure design is to obtain qualified first image and the second figure
Picture.
Below to the specific reality of the network witness management method based on artificial intelligence image recognition a kind of in present embodiment
The mode of applying is illustrated:
Step S101: being sampled test specimen in engineering site, and test specimen is fixed on specific pure color bottom plate, and passes through the
One camera obtains the first image of the test specimen.It specifically can be used such as under type: after the sampling of engineering site test specimen, described in
Test specimen is fixed on pure color bottom plate, and Attesting Officer's remote control the first camera of engineering site aims at the test specimen, is ajusted described
Bottom plate shooting angle, shooting obtain the first image.
It should be noted that referring to Fig. 3, in the present embodiment, the pure color bottom plate is arbitrarily and test specimen is at distinctness
The color of comparison, and the vertical and horizontal grid lines of spacing is portrayed on surface.Described ajust when shooting angle refers to shooting adjusts the examination
Part orientation keeps the edge of the grid lines and picture on the pure color bottom plate equidistant；The first image is first camera
Zoom scaling makes the test specimen integral into maximum as falling in picture area, and picture edge is the image of pure color bottom plate
Step S102: test specimen is sent to inspection center, and the second figure of test specimen in inspection center is obtained by second camera
Picture.Specifically it can be used such as under type: after test specimen is sent to inspection center, the camera shooting of inspection center second described in Attesting Officer's remote control
Head aims at test specimen, ajusts the bottom plate shooting angle, and shooting obtains the second image.
Also need explanation, second image be second camera zoom scaling make the test specimen integral into
Maximum picture is fallen in picture area, and picture edge is the image of pure color bottom plate.
Step S103: the fisrt feature library of the first image and the second feature library of second image are calculated.
Step S104: the characteristic similarity in the fisrt feature library Yu the second feature library is calculated, and according to the spy
Sign similarity judges whether the first image corresponds to the same test specimen with the second image.
The test specimen of the first camera and second camera shooting engineering site and inspection center is remotely controlled by networking
Image, by calculating the fisrt feature library of the first image and the second feature library of second image；Calculate described first
The characteristic similarity of feature database and the second feature library, and the first image and the second image are judged according to the characteristic similarity
The same test specimen whether is corresponded to, long-range witness work is completed；The network management for realizing witness sampling work, not only simplifies
Witness operates, and ensure that the authenticity of witness sampling result by image recognition.
Further, in the present embodiment, archive reservation can be carried out to above-mentioned comparison result, it is ensured that it is subsequent can also be into
Row review.
Referring to Fig. 2, in the present embodiment, further, it is described " calculate the first image fisrt feature library and
The second feature library of second image；Calculate the characteristic similarity in the fisrt feature library Yu the second feature library, and root
Judge whether the first image corresponds to the same test specimen with the second image according to the characteristic similarity ", it further comprises the steps of:
Step S201: the edge one for extracting the first image encloses pixel RGB values, calculates edge pixel mean value, traverses institute
The all pixels rgb value of the first image is stated, the otherness with the edge pixel mean value is calculated, when the otherness is less than setting
Threshold value when, be classified as baseplate zone, be otherwise classified as test specimen region；
Step S202: the edge one for extracting second image encloses pixel RGB values, calculates edge pixel mean value, traverses institute
The all pixels rgb value of the first image is stated, the otherness with the edge pixel mean value is calculated, when the otherness is less than setting
Threshold value when, be classified as baseplate zone, be otherwise classified as test specimen region；
It should be noted that step S201 and step S202 are relationships in no particular order, can carry out simultaneously, it can also be any
One formerly, any one is rear.
Step S203: the related coefficient of the first test specimen area image and the second test specimen area image is calculated.
Does step: S204: the related coefficient reach preset dependent thresholds?
If the related coefficient is not up to preset dependent thresholds, S205 is thened follow the steps: determining the first image and second
Corresponding image is not same test specimen；
If the related coefficient reaches preset dependent thresholds, then follow the steps S206: to the first test specimen area image and
Second test specimen area image is normalized, and extracts to obtain fisrt feature library and the second spy by CNN convolutional neural networks
Library is levied, using the similarity in fisrt feature library described in k nearest neighbor KNN algorithm comparison and the second feature library, when similar features number
When accounting for the percentage of feature database sum and reaching preset similar threshold value, then determines the first image and the second image is corresponding is
Same test specimen.
Above-mentioned steps S201 specifically can be used to step S206 such as under type:
The edge one for extracting the first image encloses pixel RGB values
Wherein subscript n is the total number that edge one encloses pixel, calculates the edge pixel mean value
Traverse all pixels rgb value of the first imageWherein subscript m is
The total number of the first image all pixels calculates the otherness with the edge pixel mean value
, (1≤i≤m), when the othernessLess than setting threshold value when, be classified as baseplate zone, otherwise return
For test specimen region, the first test specimen area image V is obtained1(r, g, b)；
The edge one for extracting second image encloses pixel RGB valuesIts
Middle subscript n is the total number that edge one encloses pixel, calculates the edge pixel mean value
Traverse all pixels rgb value of second imageWherein subscript m is
The total number of the second image all pixels calculates the otherness with the edge pixel mean value
, (1≤i≤m), when the othernessLess than setting threshold value when, be classified as baseplate zone, otherwise return
For test specimen region, the second test specimen area image V is obtained2(r, g, b)；
To the first test specimen area image V1(r, g, b) seeks the mean value M on each channel1(r, g, b), to second examination
Part area image V2(r, g, b) seeks the mean value M on each channel2(r, g, b) obtains covariance Cov (V1, V2)=E [(V1(r, g,
b)-M1(r, g, b)) (V2(r, g, b)-M2(r, g, b))], related coefficient
It when correlation coefficient ρ reaches preset dependent thresholds, is further processed, is otherwise judged to not being same examination
Part；
Be further processed it is as follows, in order to reduce caused by light differential match influence, to the first test specimen area image
It is normalized, obtains the first normalized image
N1(r, g, b)=(V1(r, g, b)-min (V1(r, g, b)))/(max (V1(r, g, b))-min (V1(r, g, b)))
The second test specimen area image is normalized, the second normalized image is obtained
N2(r, g, b)=(V2(r, g, b)-min (V2(r, g, b)))/(max (V2(r, g, b))-min (V2(r, g, b)))
Feature is extracted by CNN convolutional neural networks to the first normalized image, obtains fisrt feature library；To the second normalizing
Change image and extract feature by CNN convolutional neural networks, obtains second feature library；Use K=2 neighbour KNN algorithm comparison first
The similarity of feature database and second feature library is preset when the nearest feature Euclidean distance of KNN is less than divided by secondary nearly feature Euclidean distance
Proportion threshold value when, then be determined as similar features, when the percentage that similar features number accounts for feature database sum reaches preset similar
When threshold value, then determine that the test specimen is same test specimen.
Referring to Fig. 4, in the present embodiment, a kind of network witness managing device based on artificial intelligence image recognition
400 specific embodiment is as follows:
A kind of network witness managing device 400 based on artificial intelligence image recognition, comprising: the first camera 401, first
Pure color bottom plate 402, second camera 403, the second pure color bottom plate 404 and processor 405；The first pure color bottom plate 402 is used for:
The fixed test specimen sampled from engineering site；First camera 401 is used for: obtaining test specimen on the first pure color bottom plate 402
The first image；The second pure color bottom plate 404 is used for: being sent to the test specimen of inspection center；The second camera 403 is used for:
Obtain the second image of test specimen on the second pure color bottom plate 404；The processor 405 is used for: calculating the first image
The second feature library in fisrt feature library and second image；Calculate the feature in the fisrt feature library Yu the second feature library
Similarity, and judge whether the first image corresponds to the same test specimen with the second image according to the characteristic similarity.
Further, further includes: the first holder and the second holder；First camera 401 is zoomable camera, if
It is placed on the first holder that can be rotated up and down；The second camera 403 is zoomable camera, and being set to can be left up and down
It turns right on the second dynamic holder.
Further, the first pure color bottom plate 402 or the second pure color bottom plate 404 are arbitrarily and test specimen is at sharp contrast
Color, and the vertical and horizontal grid lines of spacing is portrayed on surface.
Further, the first image is that 401 zoom of the first camera scaling makes the test specimen integral into maximum
As falling in picture area, picture edge is the image of pure color bottom plate；Second image is 403 zoom of second camera
Scaling makes the test specimen integral into maximum as falling in picture area, and picture edge is the image of pure color bottom plate.
Further, the processor 405 is also used to: the edge one for extracting the first image encloses pixel RGB values, calculates
Edge pixel mean value traverses all pixels rgb value of the first image, calculates the otherness with the edge pixel mean value,
When the otherness is less than the threshold value of setting, it is classified as baseplate zone, is otherwise classified as test specimen region；Extract second image
Edge one encloses pixel RGB values, calculates edge pixel mean value, traverses all pixels rgb value of the first image, calculate with it is described
The otherness of edge pixel mean value is classified as baseplate zone, is otherwise classified as test specimen area when the otherness is less than the threshold value of setting
Domain；The related coefficient for calculating the first test specimen area image and the second test specimen area image, judges whether the related coefficient reaches
Preset dependent thresholds determine the first image and the second image pair if the related coefficient is not up to preset dependent thresholds
What is answered is not same test specimen；If the related coefficient reaches preset dependent thresholds, to the first test specimen area image and second
Test specimen area image is normalized, and extracts to obtain fisrt feature library and second feature library by CNN convolutional neural networks,
Using the similarity in fisrt feature library described in k nearest neighbor KNN algorithm comparison and the second feature library, when similar features number accounts for feature
When the percentage of library sum reaches preset similar threshold value, then determine the first image and the second image it is corresponding be same examination
Part.Specifically it can be used such as under type:
The edge one for extracting the first image encloses pixel RGB values
Wherein subscript n is the total number that edge one encloses pixel, calculates the edge pixel mean value
Traverse all pixels rgb value of the first imageWherein subscript m is
The total number of the first image all pixels calculates the otherness with the edge pixel mean value
, (1≤i≤m), when the othernessLess than setting threshold value when, be classified as baseplate zone, otherwise return
For test specimen region, the first test specimen area image V is obtained1(r, g, b)；
The edge one for extracting second image encloses pixel RGB values
Wherein subscript n is the total number that edge one encloses pixel, calculates the edge pixel mean value
Traverse all pixels rgb value of second imageWherein subscript m is
The total number of the second image all pixels calculates the otherness with the edge pixel mean value
, (1≤i≤m), when the othernessLess than setting threshold value when, be classified as baseplate zone, otherwise return
For test specimen region, the second test specimen area image V is obtained2(r, g, b)；
To the first test specimen area image V1(r, g, b) seeks the mean value M on each channel1(r, g, b), to second examination
Part area image V2(r, g, b) seeks the mean value M on each channel2(r, g, b) obtains covariance Cov (V1, V2)=E [(V1(r, g,
b)-M1(r, g, b)) (V2(r, g, b)-M2(r, g, b))], related coefficient
It when correlation coefficient ρ reaches preset dependent thresholds, is further processed, is otherwise judged to not being same examination
Part；
Be further processed it is as follows, in order to reduce caused by light differential match influence, to the first test specimen area image
It is normalized, obtains the first normalized image
N1(r, g, b)=(V1(r, g, b)-min (V1(r, g, b)))/(max (V1(r, g, b))-min (V1(r, g, b)))
The second test specimen area image is normalized, the second normalized image is obtained
N2(r, g, b)=(V2(r, g, b)-min (V2(r, g, b)))/(max (V2(r, g, b))-min (V2(r, g, b)))
Feature is extracted by CNN convolutional neural networks to the first normalized image, obtains fisrt feature library；To the second normalizing
Change image and extract feature by CNN convolutional neural networks, obtains second feature library；Use K=2 neighbour KNN algorithm comparison first
The similarity of feature database and second feature library is preset when the nearest feature Euclidean distance of KNN is less than divided by secondary nearly feature Euclidean distance
Proportion threshold value when, then be determined as similar features, when the percentage that similar features number accounts for feature database sum reaches preset similar
When threshold value, then determine that the test specimen is same test specimen.
The first camera 401 is remotely controlled by networking and second camera 403 shoots engineering site and inspection center
Test specimen image, pass through processor 405 calculate the first image fisrt feature library and second image second feature
Library；The characteristic similarity in the fisrt feature library Yu the second feature library is calculated, and judges according to the characteristic similarity
Whether one image corresponds to the same test specimen with the second image, completes long-range witness work；Realize the network of witness sampling work
Change management, not only simplify witness operation, and ensure that the authenticity of witness sampling result by image recognition.
It should be noted that being not intended to limit although the various embodiments described above have been described herein
Scope of patent protection of the invention.Therefore, it based on innovative idea of the invention, change that embodiment described herein is carried out and is repaired
Change, or using equivalent structure or equivalent flow shift made by description of the invention and accompanying drawing content, it directly or indirectly will be with
Upper technical solution is used in other related technical areas, is included within scope of patent protection of the invention.
Claims (10)
1. a kind of network based on artificial intelligence image recognition witnesses management method, which is characterized in that comprising steps of
Test specimen is sampled in engineering site, test specimen is fixed on specific pure color bottom plate, and is obtained by the first camera
First image of the test specimen；
Test specimen is sent to inspection center, and the second image of test specimen in inspection center is obtained by second camera；
Calculate the fisrt feature library of the first image and the second feature library of second image；
The characteristic similarity in the fisrt feature library Yu the second feature library is calculated, and judges according to the characteristic similarity
Whether one image corresponds to the same test specimen with the second image.
2. a kind of network based on artificial intelligence image recognition according to claim 1 witnesses management method, feature exists
In,
First camera and second camera are zoomable cameras, are set on the holder that can be rotated up and down.
3. a kind of network based on artificial intelligence image recognition according to claim 1 witnesses management method, feature exists
In,
The pure color bottom plate is arbitrarily and test specimen is at the color of sharp contrast, and the vertical and horizontal grid lines of spacing is portrayed on surface.
4. a kind of network based on artificial intelligence image recognition according to claim 1 witnesses management method, feature exists
In,
The first image is that the first camera zoom scaling makes the test specimen integral into maximum as falling in picture area,
Picture edge is the image of pure color bottom plate；
Second image is that the second camera zoom scaling makes the test specimen integral into maximum as falling in picture area,
Picture edge is the image of pure color bottom plate.
5. a kind of network based on artificial intelligence image recognition according to claim 1 witnesses management method, feature exists
In described " to calculate the second feature library in the fisrt feature library and second image of the first image；It is special to calculate described first
The characteristic similarity in library and the second feature library is levied, and judges that the first image is with the second image according to the characteristic similarity
The no same test specimen of correspondence ", further comprises the steps of:
The edge one for extracting the first image encloses pixel RGB values, calculates edge pixel mean value, traverses the institute of the first image
There are pixel RGB values, calculate the otherness with the edge pixel mean value, when the otherness is less than the threshold value of setting, is classified as
Otherwise baseplate zone is classified as test specimen region；
The edge one for extracting second image encloses pixel RGB values, calculates edge pixel mean value, traverses the institute of the first image
There are pixel RGB values, calculate the otherness with the edge pixel mean value, when the otherness is less than the threshold value of setting, is classified as
Otherwise baseplate zone is classified as test specimen region；
The related coefficient for calculating the first test specimen area image and the second test specimen area image, judges whether the related coefficient reaches
Preset dependent thresholds determine the first image and the second image pair if the related coefficient is not up to preset dependent thresholds
What is answered is not same test specimen；
If the related coefficient reaches preset dependent thresholds, to the first test specimen area image and the second test specimen area image into
Row normalized is extracted to obtain fisrt feature library and second feature library by CNN convolutional neural networks, be calculated using k nearest neighbor KNN
The similarity in method the fisrt feature library and the second feature library, when similar features number accounts for the percentage of feature database sum
When reaching preset similar threshold value, then determine the first image and the second image it is corresponding be same test specimen.
6. a kind of network based on artificial intelligence image recognition witnesses managing device characterized by comprising the first camera,
First pure color bottom plate, second camera, the second pure color bottom plate and processor；
The first pure color bottom plate is used for: the fixed test specimen sampled from engineering site；
First camera is used for: obtaining the first image of test specimen on the first pure color bottom plate；
The second pure color bottom plate is used for: being sent to the test specimen of inspection center；
The second camera is used for: obtaining the second image of test specimen on the second pure color bottom plate；
The processor is used for: calculating the fisrt feature library of the first image and the second feature library of second image；Meter
The characteristic similarity in the fisrt feature library Yu the second feature library is calculated, and the first image is judged according to the characteristic similarity
The same test specimen whether is corresponded to the second image.
7. a kind of network based on artificial intelligence image recognition according to claim 6 witnesses managing device, feature exists
In, further includes: the first holder and the second holder；
First camera is zoomable camera, is set on the first holder that can be rotated up and down；
The second camera is zoomable camera, is set on the second holder that can be rotated up and down.
8. a kind of network based on artificial intelligence image recognition according to claim 6 witnesses managing device, feature exists
In,
The first pure color bottom plate or the second pure color bottom plate be arbitrarily and test specimen at sharp contrast color, and between surface has been portrayed
Away from vertical and horizontal grid lines.
9. a kind of network based on artificial intelligence image recognition according to claim 6 witnesses managing device, feature exists
In,
The first image is that the first camera zoom scaling makes the test specimen integral into maximum as falling in picture area,
Picture edge is the image of pure color bottom plate；
Second image is that the second camera zoom scaling makes the test specimen integral into maximum as falling in picture area,
Picture edge is the image of pure color bottom plate.
10. a kind of network based on artificial intelligence image recognition according to claim 6 witnesses managing device, feature exists
In,
The processor is also used to: the edge one for extracting the first image encloses pixel RGB values, calculates edge pixel mean value, time
The all pixels rgb value of the first image is gone through, the otherness with the edge pixel mean value is calculated, when the otherness is less than
When the threshold value of setting, it is classified as baseplate zone, is otherwise classified as test specimen region；The edge one for extracting second image encloses pixel RGB
Value calculates edge pixel mean value, traverses all pixels rgb value of the first image, calculates and the edge pixel mean value
Otherness is classified as baseplate zone, is otherwise classified as test specimen region when the otherness is less than the threshold value of setting；Calculate the first examination
The related coefficient of part area image and the second test specimen area image, judges whether the related coefficient reaches preset related threshold
Value, if the related coefficient is not up to preset dependent thresholds, determine the first image and the second image it is corresponding be not same
Test specimen；If the related coefficient reaches preset dependent thresholds, to the first test specimen area image and the second test specimen area image
It is normalized, extracts to obtain fisrt feature library and second feature library by CNN convolutional neural networks, use k nearest neighbor KNN
The similarity in fisrt feature library described in algorithm comparison and the second feature library, when similar features number accounts for the percentage of feature database sum
When than reaching preset similar threshold value, then determine the first image and the second image it is corresponding be same test specimen.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910595628.3A CN110335265A (en) | 2019-07-03 | 2019-07-03 | A kind of network witness management method and device based on artificial intelligence image recognition |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910595628.3A CN110335265A (en) | 2019-07-03 | 2019-07-03 | A kind of network witness management method and device based on artificial intelligence image recognition |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110335265A true CN110335265A (en) | 2019-10-15 |
Family
ID=68144778
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910595628.3A Pending CN110335265A (en) | 2019-07-03 | 2019-07-03 | A kind of network witness management method and device based on artificial intelligence image recognition |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110335265A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115936540A (en) * | 2023-02-16 | 2023-04-07 | 南京方园建设工程材料检测中心有限公司 | Digital witness sampling and detecting system for building engineering |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105374036A (en) * | 2015-11-04 | 2016-03-02 | 玉溪师范学院 | Picture processing method and device |
| CN106960214A (en) * | 2017-02-17 | 2017-07-18 | 北京维弦科技有限责任公司 | Object identification method based on image |
| CN108304880A (en) * | 2018-02-05 | 2018-07-20 | 福建汇川物联网技术科技股份有限公司 | A kind of building site sample witness sampling method based on image tagged |
| CN108446660A (en) * | 2018-03-29 | 2018-08-24 | 百度在线网络技术（北京）有限公司 | The method and apparatus of facial image for identification |
-
2019
- 2019-07-03 CN CN201910595628.3A patent/CN110335265A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105374036A (en) * | 2015-11-04 | 2016-03-02 | 玉溪师范学院 | Picture processing method and device |
| CN106960214A (en) * | 2017-02-17 | 2017-07-18 | 北京维弦科技有限责任公司 | Object identification method based on image |
| CN108304880A (en) * | 2018-02-05 | 2018-07-20 | 福建汇川物联网技术科技股份有限公司 | A kind of building site sample witness sampling method based on image tagged |
| CN108446660A (en) * | 2018-03-29 | 2018-08-24 | 百度在线网络技术（北京）有限公司 | The method and apparatus of facial image for identification |
Non-Patent Citations (2)
| Title |
|---|
| MAYSSAA AL NAJJAR 等著: "《传感器平台的视频监控-算法与结构》", 30 June 2018 * |
| 王文峰 等著: "《MATLAB计算机视觉与机器认知》", 31 August 2017 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115936540A (en) * | 2023-02-16 | 2023-04-07 | 南京方园建设工程材料检测中心有限公司 | Digital witness sampling and detecting system for building engineering |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107507173B (en) | No-reference definition evaluation method and system for full-slice image | |
| CN103954542B (en) | Based on the PM2.5 concentration detectors that non-reference picture definition is evaluated | |
| CN109696788A (en) | A kind of fast automatic focusing method based on display panel | |
| CN106022345B (en) | A kind of high voltage isolator state identification method based on Hough forest | |
| CN103530599A (en) | Method and system for distinguishing real face and picture face | |
| CN107424142A (en) | A kind of weld joint recognition method based on saliency detection | |
| CN109711267A (en) | A kind of pedestrian identifies again, pedestrian movement's orbit generation method and device | |
| CN108537787A (en) | A kind of quality judging method of facial image | |
| WO2022206161A1 (en) | Feature point recognition-based block movement real-time detection method | |
| CN109299634A (en) | Spot detection method, system, equipment and storage medium | |
| CN109492647A (en) | A kind of power grid robot barrier object recognition methods | |
| CN113688817A (en) | Instrument identification method and system for automatic inspection | |
| CN110969202A (en) | Portrait collection environment verification method and system based on color component and perceptual hash algorithm | |
| CN105654452A (en) | Background clutter measurement method based on edge structures | |
| CN110335265A (en) | A kind of network witness management method and device based on artificial intelligence image recognition | |
| CN114462646B (en) | Pole number plate identification method and system based on contact network safety inspection | |
| CN113705298A (en) | Image acquisition method and device, computer equipment and storage medium | |
| CN116614705A (en) | Coal face camera regulation and control system based on multi-mode video feature analysis | |
| CN103607558A (en) | Video monitoring system, target matching method and apparatus thereof | |
| Julian et al. | Precise forecasting of sky images using spatial warping | |
| CN109711313B (en) | Real-time video monitoring method for identifying sewage poured into river channel | |
| CN108961209A (en) | Pedestrian image quality evaluating method, electronic equipment and computer-readable medium | |
| CN115147646A (en) | Garden engineering small-target pest detection method based on super-resolution reconstruction and data enhancement | |
| CN103679116B (en) | The method and device of detection scene internal object quantity | |
| JP2016220129A (en) | Camera calibration method and device |
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 |