CN108596187A - Commodity degree of purity detection method and showcase - Google Patents
Commodity degree of purity detection method and showcase Download PDFInfo
- Publication number
- CN108596187A CN108596187A CN201810291687.7A CN201810291687A CN108596187A CN 108596187 A CN108596187 A CN 108596187A CN 201810291687 A CN201810291687 A CN 201810291687A CN 108596187 A CN108596187 A CN 108596187A
- Authority
- CN
- China
- Prior art keywords
- commodity
- edge
- image
- layer
- degree
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 65
- 238000000034 method Methods 0.000 claims abstract description 19
- 238000012549 training Methods 0.000 claims abstract description 18
- 238000003708 edge detection Methods 0.000 claims abstract description 7
- 238000000605 extraction Methods 0.000 claims abstract description 6
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 4
- 239000011159 matrix material Substances 0.000 claims description 11
- 230000001537 neural effect Effects 0.000 claims description 9
- 210000002569 neuron Anatomy 0.000 claims description 6
- 230000004913 activation Effects 0.000 claims description 3
- 238000013528 artificial neural network Methods 0.000 claims description 3
- 230000004888 barrier function Effects 0.000 claims description 3
- 238000001914 filtration Methods 0.000 claims description 3
- 238000005192 partition Methods 0.000 claims description 3
- 239000000284 extract Substances 0.000 claims description 2
- 238000012544 monitoring process Methods 0.000 abstract description 3
- 238000012545 processing Methods 0.000 abstract description 3
- 238000004364 calculation method Methods 0.000 description 7
- 238000005516 engineering process Methods 0.000 description 4
- 125000002015 acyclic group Chemical group 0.000 description 1
- 238000007792 addition Methods 0.000 description 1
- 235000021443 coca cola Nutrition 0.000 description 1
- 235000020965 cold beverage Nutrition 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000002156 mixing Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 210000004218 nerve net Anatomy 0.000 description 1
- 230000008092 positive effect Effects 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000001629 suppression Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- A—HUMAN NECESSITIES
- A47—FURNITURE; DOMESTIC ARTICLES OR APPLIANCES; COFFEE MILLS; SPICE MILLS; SUCTION CLEANERS IN GENERAL
- A47F—SPECIAL FURNITURE, FITTINGS, OR ACCESSORIES FOR SHOPS, STOREHOUSES, BARS, RESTAURANTS OR THE LIKE; PAYING COUNTERS
- A47F11/00—Arrangements in shop windows, shop floors or show cases
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of commodity degree of purity detection methods, include the following steps：Off-line training commercial detector step；End article detects identification step, including：（1）, acquisition showcase in end article image, be original image；（2）, the extraction original image edge, obtain edge graph；（3）, to edge graph carry out edge detection, obtain n candidate region；（4）, using commercial detector each candidate region is finely detected respectively, obtain m selected regions；（5）, obtain selected region subgraph corresponding in the original image, the size of all subgraphs is adjusted consistent, is standard subgraph；（6）, the standard subgraph is input to convolutional neural networks disaggregated model, export the brand belonging to the commodity in the standard subgraph.The commodity degree of purity detection method of the present invention is realized the unmanned management of refrigerator degree of purity monitoring using intelligent algorithm and image processing techniques, improves Detection accuracy, while saving human cost.
Description
Technical field
The present invention relates to commodity degree of purity detection technique fields, specifically, being to be related to a kind of quotient based on image procossing
Product degree of purity detection method and showcase.
Background technology
The commodity often put in the goods showing cabinet of commodity there are many brand are shown in existing market, supermarket, work as user
It is often irresolute when selection, do not know that this buys the commodity of which kind of brand.In fact, this goods showing cabinet is not allow to put
The commodity of a variety of brands because this goods showing cabinet be brand quotient (for example, Coca-Cola) it is free grant to market,
Supermarket, in should can only put the commodity of brand quotient.But since market, supermarket are excessive, in addition the problems such as personnel expend,
Brand quotient is difficult to real-time live and supervises to ensure the degree of purity of commodity in showcase, thus so that mix in showcase
The phenomenon that placing multi-brand commodity frequently occurs, and compromises the interests of showcase provider.
The degree of purity detection of goods showing cabinet annoyings always brand quotient, to ensure that the interests of brand quotient, the prior art go out
It is as follows a kind of drink degree of purity detection method, detection device and drink showcase, concrete operations have been showed：
The real-time weight data of drink frame where obtaining drink；By the real-time weight data and stored weight data
Library is compared, and judges whether the real-time weight data matches with the weight data in affiliated weight DBMS；If phase
Match, judgement drink is pure；If not matching that, judgement drink is not pure；The weight DBMS is pre-established using following methods：
It is sequentially placed all drinks combination of specified drink on drink frame, obtains the weight number of drink frame when placing the combination of each drink
According to the corresponding all wt data of all drink combinations form the weight DBMS；The specified drink is that possible place
In all drinks of the same brand on the drink frame.The program is by establishing with all combination weight databases pair of brand drink
Weight in showcase is monitored.The weight data relied on is a kind of one-dimensional data, including information content it is very few.Different product
The drink of board is very close in weight, and the weight of the various combination of same drink must with mixing the combination of other brands
Can ponderable intersection, so only relying on a kind of weight this Data Detection precision will not be too high.
Invention content
The present invention judges the low technical problem of precision to solve the degree of purity detection of existing goods showcase based on weight,
A kind of commodity degree of purity detection method is proposed, can be solved the above problems.
In order to solve the above-mentioned technical problem, the present invention is achieved by the following scheme：
A kind of commodity degree of purity detection method, includes the following steps：
Off-line training commercial detector step, including：
The image of standard merchandise is acquired, each band of position is extracted in the band of position of labeled standards commodity in the picture
HOG features, training AdaBoost commercial detectors；
End article detects identification step, including：
(1), the image of end article in showcase is acquired, is original image Isrc；
(2), the original image I is extractedsrcEdge, obtain edge graph Iedge；
(3), to edge graph IedgeEdge detection is carried out, n candidate region R is obtained1,R2,…,Rn, wherein n is positive integer；
(4), using commercial detector respectively to each candidate region R1,R2,…,RnIt is finely detected, it is a selected to obtain m
Region Ri' (i=1,2 ... m), and wherein m is the positive integer less than n；
(5), selected region subgraph corresponding in the original image is obtained, by the size of all subgraphs
Adjustment is consistent, is standard subgraph；
(6), the standard subgraph is input to convolutional neural networks disaggregated model, exports the quotient in the standard subgraph
Brand belonging to product.
Further, further include off-line training depth convolutional Neural sorter network before end article detects identification step
Step, including：
The picture for acquiring the picture of the various samples of this brand and the sample of various other brands, for the sample of this brand
Product are classified, and a class number is respectively set in the sample of this brand of every class, are arranged one to all samples of other brands
The picture and its class number of this brand and other brand all samples are input in neural network and instruct by class number
Practice, depth convolutional Neural sorter network is obtained, for judging the brand belonging to commodity and classification.
Further, the structure of the depth convolutional Neural sorter network is：
With 2 convolutional layers, 2 full articulamentums, wherein：
First layer is convolutional layer, and the activation primitive of first layer is ReLu, and pondization uses maximum value pond, the size in pond to select
Select 3*3, pond stride selections 2；
First layer is convolutional layer, and the quantity of second layer convolution kernel is more than the quantity of first layer convolution kernel, second layer convolution kernel
Size be less than first layer convolution kernel size, second layer convolution step-length be less than first layer convolution step-length, pond size selection
3*3, pond stride selection 2；
Third layer is full articulamentum, neuron number selection 512；
4th layer is full articulamentum, neuron number selection 512；
Layer 6 is output layer, for exporting the class number corresponding to end article.
Further, use Hough transform method to edge graph I in step (3)edgeCarry out edge detection.
Further, in training commercial detector step, the image pattern of similar commodity is acquired, quotient in image pattern is extracted
The HOG features of product, and be input in AdaBoost cascade classifiers and be trained, it obtains training commercial detector.
Further, in step (2), original image I is extractedsrcEdge method be：
(a), to original image IsrcGray-scale map I is calculated according to formula Gray=0.299R+0.587G+0.114BG,
Wherein, R, G, B are original image I respectivelysrcR, the value in tri- channels G, B；
(b), to gray-scale map IGGaussian filtering is carried out, filtered image is IF。
(c), image I is calculatedFGradient matrix M and gradient direction matrix θ；
(d), the shape according to commodity in the picture, gradient direction has specific interval range, by gradient direction square
It is located at the pixel corresponding to the value except interval range in battle array θ to filter out, obtains edge graph Iedge。
Further, the shape according to end article in the picture, using Hough transform to edge graph IedgeCarry out shape
Detection, obtains n candidate region R1,R2,…,Rn。
Further, the shape of end article in the picture is round or ellipse, using Hough transform to edge graph
IedgeEllipses detection is carried out, n candidate region R is obtained1,R2,…,Rn。
Further, the middle method finely detected of step (4) is：Each candidate region is a rectangle region
Domain sets detection window for the candidate region, the detection window is slided in candidate region, and often sliding is primary,
The commercial detector calculates the confidence value for exporting and whether there is end article in a current window, by the confidence value
It is compared with confidence threshold value, judges to whether there is end article in current candidate region.
The present invention proposes a kind of goods showing cabinet, including cabinet, cabinet door simultaneously, and partition board is provided with by cabinet in the cabinet
Internal barrier is respectively arranged with image collecting device, for acquiring at several storage spaces on the top plate of each storage space
The image being disposed below in space, described image harvester are connect with master control borad, and the master control borad will by wireless network
Image is sent to server, carries out commodity degree of purity detection, executes commodity degree of purity detection method noted before.
Compared with prior art, the advantages and positive effects of the present invention are：The commodity degree of purity detection method of the present invention, is adopted
With by slightly to the detection mode of essence, first, the specific shape based on commodity, the method by extracting image border can be rapidly
A large amount of background area is removed, reduces interference factor, while based on oval and circle detection, rapid extraction drink candidate regions
Domain.In candidate region by Adaboost graders carry out commodity be accurately positioned avoid in the big region of picture in its entirety into
Row detection, also can guarantee higher precision while reducing calculation amount.Finally, using depth convolutional network to each of detecting
Commodity carry out brand recognition, and using recognition result, the drink for judging whether there is non-cooperation manufacturer occurs.This programme is calculated using intelligence
Method and image processing techniques realize the unmanned management of refrigerator degree of purity monitoring, improve Detection accuracy, save simultaneously
Human cost.
After the detailed description of embodiment of the present invention is read in conjunction with the figure, the other features and advantages of the invention will become more
Add clear.
Description of the drawings
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technology description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
Obtain other attached drawings according to these attached drawings.
Fig. 1 is a kind of embodiment flow chart of commodity degree of purity detection method proposed by the invention；
Fig. 2 is the sub-process figure of end article detection identification step in Fig. 1；
Fig. 3 is a kind of example structure schematic diagram of showcase proposed by the invention.
Specific implementation mode
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation describes, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
For the not high problem of current commercial display cases or commercial display cabinet monitoring display goods degree of purity accuracy rate, this hair
It is bright to propose by installing camera, the commodity image in periodical or acyclic acquisition cabinet in refrigerator, and it is transmitted to cloud
Server is held, commodity are detected and brand is known to putting in cabinet using image procossing and depth nerve target detection network
Not, the category and quantity of current commodity are obtained, and then the degree of purity of commodity in current refrigerator is analyzed, if there is non-designated
Businessman's commodity, then alert, is transmitted to related supervisor.It will be illustrated below with a specific embodiment.
Embodiment one, the present embodiment propose a kind of commodity degree of purity detection method, as shown in Figure 1 and Figure 2, including it is following
Step：
Off-line training commercial detector step, including：
The image of standard merchandise is acquired, each band of position is extracted in the band of position of labeled standards commodity in the picture
HOG features, training AdaBoost commercial detectors；
The commodity shown under normal circumstances have specific shape, as drink class be generally it is bottled or canned, should
Kind of commodity have a more significant contour feature in the picture, this step by manually collecting the figures of a large amount of ampuliform objects offline
Piece, it is desirable that the type of ampuliform object is no less than 20 classes, per in pictures containing only there are one ampuliform object, and the area of ampuliform object
Account for 90% or more of the whole picture area of pictural surface.By extract picture HOG features, be then input in AdaBoost cascade classifiers into
Row training, obtains an AdaBoost commercial detector.
End article detects identification step, including：
The image of end article, is original image I in S1, acquisition showcasesrc；
S2, the extraction original image IsrcEdge, obtain edge graph Iedge；
S3, to edge graph IedgeEdge detection is carried out, n candidate region R is obtained1,R2,…,Rn, wherein n is positive integer；
This step there may be the region of end article for being obtained roughly from entire image, can remove due to inside showcase
The much noise interference that shelf generates, while calculation amount can be reduced.
S4, using commercial detector respectively to each candidate region R1,R2,…,RnIt is finely detected, obtains m selected areas
Domain Ri' (i=1,2 ... m), and wherein m is the positive integer less than n；This step is used to further carry out high-precision from candidate region
Detection judge, judge whether be truly present end article in the candidate region.
S5, selected region subgraph corresponding in the original image is obtained, by the size of all subgraphs
Adjustment is consistent, is standard subgraph；Due between each commodity size have differences and the distance of range image harvester and
Angle is different, therefore shared region is also not quite similar each commodity in the picture, after passing through Preliminary detection and essence detection, each time
Size between favored area and each selected region is accordingly also not quite similar, and therefore, calculates for convenience, further includes to selected region
The step of corresponding subgraph is normalized in the original image, that is, the size of all subgraphs is all adjusted
To be onesize, there are many adjustment modes, such as the method that interpolation may be used is realized, but is not limited to interpolation method.
S6, the standard subgraph is input to convolutional neural networks disaggregated model, exports the quotient in the standard subgraph
Brand belonging to product.
The commodity degree of purity detection method of the present embodiment, use by slightly to essence detection mode, first, based on commodity
Specific shape can rapidly remove a large amount of background area by the method for extracting image border, reduce interference factor, together
When based on oval and circle detection, rapid extraction drink candidate region.It is carried out by Adaboost graders in candidate region
Commodity are accurately positioned to avoid and are detected in the big region of picture in its entirety, also can guarantee while reducing calculation amount higher
Precision.Finally, brand recognition is carried out to each commodity detected using depth convolutional network, using recognition result, judgement is
The no drink for having non-cooperation manufacturer occurs.This programme realizes refrigerator degree of purity using intelligent algorithm and image processing techniques and monitors
Unmanned management, improve Detection accuracy, while saving human cost.
Further include off-line training depth convolution before end article detects identification step as a preferred embodiment
Neural sorter network step, including：
The picture for acquiring the picture of the various samples of this brand and the sample of various other brands, for the sample of this brand
Product are classified, and a class number is respectively set in the sample of this brand of every class, are arranged one to all samples of other brands
The picture and its class number of this brand and other brand all samples are input in neural network and instruct by class number
Practice, depth convolutional Neural sorter network is obtained, for judging the brand belonging to commodity and classification.For example, collecting 9 class different manufacturers
Drink picture, ensure at 1000 or more per class drink.Wherein preceding 8 class is the drink of cooperation producer, per class picture setting one
A class number indicates here with 1,2 ..., 8.9th class can be the drink picture of any producer other than this 8 class, and classification is compiled
It number 0 is indicated with number.The training picture for accomplishing fluently number is uniformly zoomed into 227*227 sizes, is input to designed nerve net
It is trained in network, finally obtains a sorter network, the classification of output end article, which is compiled, to be identified for the image of input
Number, and then for judging the brand belonging to drink.
The structure of depth convolutional Neural sorter network is：
With 2 convolutional layers, 2 full articulamentums, wherein：
First layer is convolutional layer, and the activation primitive of first layer is ReLu, and pondization uses maximum value pond, the size in pond to select
Select 3*3, pond stride selections 2；96 convolution kernels may be used in the convolutional layer of first layer, and each convolution kernel size is 7*7, volume
Product step-length is 4.
First layer is convolutional layer, and the quantity of second layer convolution kernel is more than the quantity of first layer convolution kernel, second layer convolution kernel
Size be less than first layer convolution kernel size, second layer convolution step-length be less than first layer convolution step-length, pond size selection
3*3, pond stride selection 2；256 convolution kernels may be used in the convolutional layer of the second layer, and convolution kernel size is 5*5, convolution step
A length of 2.
Third layer is full articulamentum, neuron number selection 512；
4th layer is full articulamentum, neuron number selection 512；
Layer 6 is output layer, for exporting the class number corresponding to end article.For example, the target category mesh of output
Before be set as 9, including the brand of 8 cooperation producers and 1 interference brand (all drink brands of non-cooperation producer).
Using Hough transform method to edge graph I in step S3edgeCarry out edge detection.
In training commercial detector step, the image pattern of similar commodity is acquired, the HOG for extracting commodity in image pattern is special
Sign, and be input in AdaBoost cascade classifiers and be trained, it obtains training commercial detector.
In step S2, original image I is extractedsrcEdge method be：
A, to original image IsrcGray-scale map I is calculated according to formula Gray=0.299R+0.587G+0.114BG,
In, R, G, B are original image I respectivelysrcR, the value in tri- channels G, B；
B, to gray-scale map IGGaussian filtering is carried out, filtered image is IF。
Gauss radius can be set as 3, using two-dimensional Gaussian kernel parameter calculation formula：
C, image I is calculatedFGradient matrix M and gradient direction matrix θ；
Calculation formula is as follows：
P [i, j]=(f [i, j+1]-f [i, j]+f [i+1, j+1]-f [i+1, j])/2
Q [i, j]=(f [i, j]-f [i+1, j]+f [i, j+1]-f [i+1, j+1])/2
θ [i, j]=arctan (Q [i, j]/p [i, j])
Wherein, f [i, j] indicates original image IsrcPixel value at the positions matrix i, j, P [i, j] indicate original image IsrcSquare
Battle array i, the gradient of horizontal direction at the positions j, Q [i, j] indicate original image IsrcThe gradient of vertical direction at the positions matrix i, j.
D, the shape according to commodity in the picture, gradient direction has specific interval range, by gradient direction matrix θ
In be located at interval range except value corresponding to pixel filter out, obtain edge graph Iedge.The drink quotient being previously mentioned
For product, regular circular or ellipse is presented in the vertical view of such drink is by gradient direction according to this special rule
0 degree and 90 degree of pixel value filters out, and obtains doubtful elliptical edge.This operation, can be by a large amount of background especially cold drink
The much noise edge that shelf generates is got rid of, and interference is reduced, and carries out region of interesting extraction convenient for follow-up, it is accurate to improve
Rate.Non-maxima suppression is carried out to matrix M simultaneously, local maximum is obtained, to obtain showcase image IsrcEdge two
Value figure Iedge。
According to the shape of end article in the picture, using Hough transform to edge graph IedgeSHAPE DETECTION is carried out, is obtained
N candidate region R1,R2,…,Rn。
For the drink commodity being previously mentioned, regular circular or ellipse is presented in the vertical view of such drink, so,
Using Hough transform to edge graph IedgeEllipses detection is carried out, n candidate region R is obtained1,R2,…,Rn.By utilizing target
The SHAPE DETECTION of commodity can further filter out the interference pixel such as tag in showcase, shelf, further decrease fine inspection
Calculation amount when survey.
The method finely detected in step S4 is：Each candidate region is a rectangular area, is the time
Favored area sets detection window, and the detection window is slided in candidate region, and often sliding is primary, commercial detector meter
The confidence value for exporting and whether there is end article in a current window is calculated, the confidence value and confidence threshold value are carried out
Compare, judges to whether there is end article in current candidate region.Confidence level indicates that there are end articles in current detection window
Credibility.
Since after Preliminary detection and essence detection, the size between each candidate region and each selected region is accordingly
It is not quite similar, therefore, before fine detection, respectively each candidate region setting detection window that size matches therewith,
It is slided using the detection window of the scale and sliding step that match in each candidate region.Often sliding is primary,
Grader in AdaBoost exports the value that whether there is ampuliform drink in a current window according to the confidence threshold value of setting,
0, which indicates that current detection window is interior, is not present ampuliform drink, and 1 indicates exist.In general, confidence threshold value is set as 0.95, set
Reliability indicates that there is currently the credibilities of ampuliform object.
More coarse candidate region is obtained in step s3, which includes bottled object, but is also contained
Background area around bottled drink, position are not very accurate.It is precisely examined with Adaboost graders again on this basis
It surveys, avoids and be detected in full figure region, reduce the calculation amount in step S4, while also ensuring precision.
Embodiment two
The present embodiment proposes a kind of goods showing cabinet, as shown in figure 3, including cabinet 11, cabinet door 12, setting in cabinet 11
There is partition board 13 by barrier in cabinet at several storage spaces, image collector is respectively arranged on the top plate of each storage space
14 are set, for acquiring the image being disposed below in space, image collecting device 14 is connect with master control borad (not shown), main
Image is sent to server by control plate by wireless network, carries out commodity degree of purity detection, according to recorded in embodiment one
Commodity degree of purity detection method is detected, and this will not be repeated here, wherein wide-angle lens reality may be used in image collecting device 14
It is existing, naturally it is also possible to be realized using other cameras with image collecting function.
Certainly, above description is not limitation of the present invention, and the present invention is also not limited to the example above, this technology neck
The variations, modifications, additions or substitutions that the those of ordinary skill in domain is made in the essential scope of the present invention, should also belong to this hair
Bright protection domain.
Claims (10)
1. a kind of commodity degree of purity detection method, which is characterized in that include the following steps：
Off-line training commercial detector step, including：
The image of standard merchandise is acquired, the HOG spies of each band of position are extracted in the band of position of labeled standards commodity in the picture
Sign, training AdaBoost commercial detectors；
End article detects identification step, including：
(1), the image of end article in showcase is acquired, is original image Isrc；
(2), the original image I is extractedsrcEdge, obtain edge graph Iedge；
(3), to edge graph IedgeEdge detection is carried out, n candidate region R is obtained1,R2,…,Rn, wherein n is positive integer；
(4), using commercial detector respectively to each candidate region R1,R2,…,RnIt is finely detected, obtains m selected regions
R′i(i=1,2 ... m), and wherein m is the positive integer less than n；
(5), selected region subgraph corresponding in the original image is obtained, the size of all subgraphs is adjusted
Unanimously, it is standard subgraph；
(6), the standard subgraph is input to convolutional neural networks disaggregated model, exports the commodity institute in the standard subgraph
The brand of category.
2. commodity degree of purity detection method according to claim 1, which is characterized in that detect identification step in end article
Further include off-line training depth convolutional Neural sorter network step before, including：
The picture for acquiring the picture of the various samples of this brand and the sample of various other brands, for this brand sample into
A class number is respectively set in the sample of row classification, this brand of every class, and a classification is arranged to all samples of other brands
Number, the picture and its class number of this brand and other brand all samples are input in neural network and are trained, is obtained
To depth convolutional Neural sorter network, for judging the brand belonging to commodity and classification.
3. commodity degree of purity detection method according to claim 2, which is characterized in that the depth convolutional Neural classification net
The structure of network is：
With 2 convolutional layers, 2 full articulamentums, wherein：
First layer is convolutional layer, and the activation primitive of first layer is ReLu, and pondization uses maximum value pond, the size in pond to select 3*
3, pond stride selection 2；
First layer is convolutional layer, and the quantity of second layer convolution kernel is more than the quantity of first layer convolution kernel, the ruler of second layer convolution kernel
The very little size less than first layer convolution kernel, second layer convolution step-length are less than first layer convolution step-length, and the size in pond selects 3*3,
Pond stride selections 2；
Third layer is full articulamentum, neuron number selection 512；
4th layer is full articulamentum, neuron number selection 512；
Layer 6 is output layer, for exporting the class number corresponding to end article.
4. according to claim 1-3 any one of them commodity degree of purity detection methods, which is characterized in that used in step (3)
Hough transform method is to edge graph IedgeCarry out edge detection.
5. according to claim 1-3 any one of them commodity degree of purity detection methods, which is characterized in that training commercial detector
In step, the image pattern of similar commodity is acquired, extracts the HOG features of commodity in image pattern, and is input to AdaBoost grades
It is trained in connection grader, obtains training commercial detector.
6. according to claim 1-3 any one of them commodity degree of purity detection methods, which is characterized in that in step (2), extraction
Original image IsrcEdge method be：
(a), to original image IsrcGray-scale map I is calculated according to formula Gray=0.299R+0.587G+0.114BG, wherein
R, G, B are original image I respectivelysrcR, the value in tri- channels G, B；
(b), to gray-scale map IGGaussian filtering is carried out, filtered image is IF。
(c), image I is calculatedFGradient matrix M and gradient direction matrix θ；
(d), the shape according to commodity in the picture, gradient direction have specific interval range, will be in gradient direction matrix θ
The pixel corresponding to value except interval range filters out, and obtains edge graph Iedge。
7. commodity degree of purity detection method according to claim 6, which is characterized in that in the picture according to end article
Shape, using Hough transform to edge graph IedgeSHAPE DETECTION is carried out, n candidate region R is obtained1,R2,…,Rn。
8. commodity degree of purity detection method according to claim 6, which is characterized in that the shape of end article in the picture
To be round or oval, using Hough transform to edge graph IedgeEllipses detection is carried out, n candidate region R is obtained1,
R2,…,Rn。
9. according to claim 1-3 any one of them commodity degree of purity detection methods, which is characterized in that carried out in step (4)
The method finely detected is：Each candidate region is a rectangular area, and detection window is set for the candidate region, will
The detection window is slided in candidate region, and often sliding is primary, and the commercial detector calculates one current window of output
The confidence value that whether there is end article in mouthful, the confidence value is compared with confidence threshold value, judges current wait
It whether there is end article in favored area.
10. a kind of goods showing cabinet, including cabinet, cabinet door, partition board is provided with by barrier in cabinet at several in the cabinet
Storage space, it is characterised in that：It is respectively arranged with image collecting device on the top plate of each storage space, is located at it for acquiring
Image in underlying space, described image harvester are connect with master control borad, and the master control borad is sent out image by wireless network
It send to server, carries out commodity degree of purity detection, perform claim requires 1-9 any one of them commodity degree of purity detection methods.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810291687.7A CN108596187B (en) | 2018-03-30 | 2018-03-30 | Commodity purity detection method and display cabinet |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810291687.7A CN108596187B (en) | 2018-03-30 | 2018-03-30 | Commodity purity detection method and display cabinet |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108596187A true CN108596187A (en) | 2018-09-28 |
| CN108596187B CN108596187B (en) | 2023-07-04 |
Family
ID=63625285
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810291687.7A Active CN108596187B (en) | 2018-03-30 | 2018-03-30 | Commodity purity detection method and display cabinet |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108596187B (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109409159A (en) * | 2018-10-11 | 2019-03-01 | 上海亿保健康管理有限公司 | A kind of fuzzy two-dimensional code detection method and device |
| CN110705371A (en) * | 2019-09-09 | 2020-01-17 | 上海零眸智能科技有限公司 | Refrigerator structure and commodity arrangement detection method thereof |
| CN111242170A (en) * | 2019-12-31 | 2020-06-05 | 航天信息股份有限公司 | Food inspection and detection item prediction method and device |
| CN111310729A (en) * | 2020-03-16 | 2020-06-19 | 南京掌控网络科技有限公司 | Method for acquiring refrigerator purity based on image recognition technology and asset management system |
| CN111860371A (en) * | 2020-07-24 | 2020-10-30 | 浙江星星冷链集成股份有限公司 | Method for detecting commodity type, quantity and purity and freezer thereof |
| CN113468914A (en) * | 2020-03-30 | 2021-10-01 | 杭州海康威视数字技术股份有限公司 | Method, device and equipment for determining purity of commodities |
| CN113673576A (en) * | 2021-07-26 | 2021-11-19 | 浙江大华技术股份有限公司 | Image detection method, terminal and computer readable storage medium thereof |
| CN113743223A (en) * | 2021-08-04 | 2021-12-03 | Tcl家用电器（合肥）有限公司 | Article storage method and storage device thereof, storage medium and refrigerator |
| CN114689582A (en) * | 2022-03-15 | 2022-07-01 | 郑州凯雪冷链股份有限公司 | Method for detecting purity of commodities of vertical air curtain cabinet |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090127326A1 (en) * | 2007-11-20 | 2009-05-21 | Datalogic Scanning, Inc. | Enhanced virtual scan line processing |
| CN101923645A (en) * | 2009-06-09 | 2010-12-22 | 黑龙江大学 | Iris splitting method suitable for low-quality iris image in complex application context |
| CN106127145A (en) * | 2016-06-21 | 2016-11-16 | 重庆理工大学 | Pupil diameter and tracking |
| CN106644802A (en) * | 2015-10-28 | 2017-05-10 | 青岛海尔智能技术研发有限公司 | Beverage purity detection method, beverage purity detection apparatus and beverage display cabinet |
| CN107392931A (en) * | 2017-08-08 | 2017-11-24 | 南京敏光视觉智能科技有限公司 | Bar tobacco board sorter and method |
| CN107463946A (en) * | 2017-07-12 | 2017-12-12 | 浙江大学 | A kind of type of merchandize detection method of combination template matches and deep learning |
-
2018
- 2018-03-30 CN CN201810291687.7A patent/CN108596187B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090127326A1 (en) * | 2007-11-20 | 2009-05-21 | Datalogic Scanning, Inc. | Enhanced virtual scan line processing |
| CN101923645A (en) * | 2009-06-09 | 2010-12-22 | 黑龙江大学 | Iris splitting method suitable for low-quality iris image in complex application context |
| CN106644802A (en) * | 2015-10-28 | 2017-05-10 | 青岛海尔智能技术研发有限公司 | Beverage purity detection method, beverage purity detection apparatus and beverage display cabinet |
| CN106127145A (en) * | 2016-06-21 | 2016-11-16 | 重庆理工大学 | Pupil diameter and tracking |
| CN107463946A (en) * | 2017-07-12 | 2017-12-12 | 浙江大学 | A kind of type of merchandize detection method of combination template matches and deep learning |
| CN107392931A (en) * | 2017-08-08 | 2017-11-24 | 南京敏光视觉智能科技有限公司 | Bar tobacco board sorter and method |
Non-Patent Citations (1)
| Title |
|---|
| 刘永豪: ""基于深度学习的货架商品检测技术研究"", 《中国优秀博硕士学位论文全文数据库(硕士) 信息科技辑》 * |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109409159A (en) * | 2018-10-11 | 2019-03-01 | 上海亿保健康管理有限公司 | A kind of fuzzy two-dimensional code detection method and device |
| CN110705371B (en) * | 2019-09-09 | 2023-06-06 | 上海零眸智能科技有限公司 | Refrigerator structure and commodity surface arrangement detection method thereof |
| CN110705371A (en) * | 2019-09-09 | 2020-01-17 | 上海零眸智能科技有限公司 | Refrigerator structure and commodity arrangement detection method thereof |
| CN111242170A (en) * | 2019-12-31 | 2020-06-05 | 航天信息股份有限公司 | Food inspection and detection item prediction method and device |
| CN111242170B (en) * | 2019-12-31 | 2023-07-25 | 航天信息股份有限公司 | Food inspection and detection project prediction method and device |
| CN111310729A (en) * | 2020-03-16 | 2020-06-19 | 南京掌控网络科技有限公司 | Method for acquiring refrigerator purity based on image recognition technology and asset management system |
| CN111310729B (en) * | 2020-03-16 | 2024-08-02 | 南京掌控网络科技有限公司 | Method for acquiring purity of refrigerator based on image recognition technology and asset management system |
| CN113468914A (en) * | 2020-03-30 | 2021-10-01 | 杭州海康威视数字技术股份有限公司 | Method, device and equipment for determining purity of commodities |
| CN113468914B (en) * | 2020-03-30 | 2023-08-29 | 杭州海康威视数字技术股份有限公司 | Method, device and equipment for determining purity of commodity |
| CN111860371A (en) * | 2020-07-24 | 2020-10-30 | 浙江星星冷链集成股份有限公司 | Method for detecting commodity type, quantity and purity and freezer thereof |
| CN113673576A (en) * | 2021-07-26 | 2021-11-19 | 浙江大华技术股份有限公司 | Image detection method, terminal and computer readable storage medium thereof |
| CN113743223A (en) * | 2021-08-04 | 2021-12-03 | Tcl家用电器（合肥）有限公司 | Article storage method and storage device thereof, storage medium and refrigerator |
| CN114689582A (en) * | 2022-03-15 | 2022-07-01 | 郑州凯雪冷链股份有限公司 | Method for detecting purity of commodities of vertical air curtain cabinet |
| CN114689582B (en) * | 2022-03-15 | 2024-08-20 | 郑州凯雪冷链股份有限公司 | Commodity purity detection method for vertical air curtain cabinet |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108596187B (en) | 2023-07-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108596187A (en) | Commodity degree of purity detection method and showcase | |
| CN105913040B (en) | The real-time pedestrian detecting system of dual camera under the conditions of noctovision | |
| CN108875911A (en) | One kind is parked position detecting method | |
| Clausi et al. | Comparing cooccurrence probabilities and Markov random fields for texture analysis of SAR sea ice imagery | |
| CN109447169A (en) | The training method of image processing method and its model, device and electronic system | |
| US7421125B1 (en) | Image analysis, editing and search techniques | |
| CN102113335B (en) | Image processing apparatus and method | |
| KR101995294B1 (en) | Image analysis apparatus and method | |
| CN107992871A (en) | The automatic accounting method in dining room based on image recognition | |
| CN108345912A (en) | Commodity rapid settlement system based on RGBD information and deep learning | |
| CN106952402A (en) | A kind of data processing method and device | |
| US20140313345A1 (en) | Flying object visual identification system | |
| US20110200249A1 (en) | Surface detection in images based on spatial data | |
| US9092650B2 (en) | Methods and apparatus for automated detection of the presence and type of caps on vials and containers | |
| CN103093212A (en) | Method and device for clipping facial images based on face detection and face tracking | |
| Lu et al. | Saliency modeling from image histograms | |
| CN111080493B (en) | Dish information identification method and device and dish self-service settlement system | |
| CN102084397A (en) | Image processing device, method, and program | |
| CN110189375A (en) | A kind of images steganalysis method based on monocular vision measurement | |
| CN109977251A (en) | A method of building identifies commodity based on RGB histogram feature | |
| CN105224911A (en) | A kind of various visual angles pedestrian detection method and system in real time | |
| Yuan et al. | Road segmentation in aerial images by exploiting road vector data | |
| CN112053371A (en) | Water body extraction method and device in remote sensing image | |
| US11748787B2 (en) | Analysis method and system for the item on the supermarket shelf | |
| CN113468914B (en) | Method, device and equipment for determining purity of commodity |
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 | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20230613 Address after: 266101 Haier Road, Laoshan District, Qingdao, Qingdao, Shandong Province, No. 1 Applicant after: QINGDAO HAIER SMART TECHNOLOGY R&D Co.,Ltd. Applicant after: Haier Smart Home Co., Ltd. Address before: 266101 Haier Road, Laoshan District, Qingdao, Qingdao, Shandong Province, No. 1 Applicant before: QINGDAO HAIER SMART TECHNOLOGY R&D Co.,Ltd. |
|
| TA01 | Transfer of patent application right | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |