CN109377205A - A kind of cafeteria's intelligence settlement system based on depth convolutional network - Google Patents
A kind of cafeteria's intelligence settlement system based on depth convolutional network Download PDFInfo
- Publication number
- CN109377205A CN109377205A CN201811492955.8A CN201811492955A CN109377205A CN 109377205 A CN109377205 A CN 109377205A CN 201811492955 A CN201811492955 A CN 201811492955A CN 109377205 A CN109377205 A CN 109377205A
- Authority
- CN
- China
- Prior art keywords
- convolutional neural
- neural networks
- neural network
- cafeteria
- training
- 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
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 69
- 238000013528 artificial neural network Methods 0.000 claims abstract description 61
- 238000012549 training Methods 0.000 claims abstract description 60
- 235000013311 vegetables Nutrition 0.000 claims abstract description 38
- 238000005457 optimization Methods 0.000 claims abstract description 15
- 230000000694 effects Effects 0.000 claims abstract description 11
- 230000001537 neural effect Effects 0.000 claims description 11
- 238000006243 chemical reaction Methods 0.000 claims description 6
- 235000012054 meals Nutrition 0.000 claims description 5
- 230000008676 import Effects 0.000 claims description 3
- 238000002372 labelling Methods 0.000 claims description 3
- 210000005036 nerve Anatomy 0.000 claims description 2
- 230000001151 other effect Effects 0.000 claims 1
- 238000000034 method Methods 0.000 abstract description 9
- 230000008569 process Effects 0.000 abstract description 6
- 230000000116 mitigating effect Effects 0.000 abstract description 2
- 210000002569 neuron Anatomy 0.000 description 12
- 230000006870 function Effects 0.000 description 6
- 235000013410 fast food Nutrition 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 4
- 238000000605 extraction Methods 0.000 description 4
- 230000008859 change Effects 0.000 description 3
- 238000001514 detection method Methods 0.000 description 3
- 230000006872 improvement Effects 0.000 description 3
- 230000035945 sensitivity Effects 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 239000012141 concentrate Substances 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 239000003550 marker Substances 0.000 description 2
- 238000012797 qualification Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000034303 cell budding Effects 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 210000004209 hair Anatomy 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 230000003340 mental effect Effects 0.000 description 1
- 210000004218 nerve net Anatomy 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000012827 research and development Methods 0.000 description 1
- 230000033764 rhythmic process Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q20/00—Payment architectures, schemes or protocols
- G06Q20/08—Payment architectures
- G06Q20/10—Payment architectures specially adapted for electronic funds transfer [EFT] systems; specially adapted for home banking systems
-
- 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)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- Accounting & Taxation (AREA)
- General Physics & Mathematics (AREA)
- Biophysics (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Finance (AREA)
- Economics (AREA)
- Development Economics (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
Cafeteria's intelligence settlement system based on depth convolutional network that the invention discloses a kind of, the identification and counting machine and vegetable classifier of bowl are generated including the use of convolutional neural networks training, system Automatic-settlement step, this kind invention design is reasonable, system can use the study of neural network and fast parallel function realizes the quick identification to handwritten numeral, the related working efficiency for using field of strong raising, cafeteria's intelligence settlement system based on depth convolutional neural networks by building multilayer depth convolutional neural networks and is trained neural network, by reversely modifying each layer weighting parameter optimization neural network recognition effect in training process, system passes through two classifiers, video image is analyzed, it detects vegetable to be settled accounts and is settled accounts, after the completion of clearing, prompt person sponging on an aristocrat's payment, system improves cafeteria It settles accounts speed, reduce due to calculating loss caused by mistake when staff's clearing, mitigating worker workload, be suitble to be widely popularized.
Description
Technical field
The present invention relates to technical field of image processing, in particular to a kind of cafeteria's intelligence based on depth convolutional network
Settlement system.
Background technique
With the fast development of science and technology, the continuous improvement of rhythm of life, the catering trade structure in China is also carried out constantly
Adjustment, transition and upgrade.In the change in this epoch, fast food restaurant is born and growth attracts attention.Fast food restaurant is quick by this phase,
Convenient, advantages, the catering market of having stood firm steadily such as cuisine is abundant.Now, fast food restaurant is from the establishment budding stage, Fast Growth
Stage has strided forward the stage of ripeness.Although fast food restaurant has been in the stage of ripeness, it is still faced with many problems.
1. vegetable pattern is more, cashier's price is misremembered:
As people's living standard constantly improves, vegetable is multifarious simultaneously meeting, the workload meeting of cashier
It greatly increases, vegetable price is be easy to cause to misremember；
2. the meal time concentrates, flow of the people is big:
Fast food is mainly gathered in business garden, office building, and office worker's meal time concentrates, and causes specific time flow of the people
Greatly, it is lined up congestion, is lost part traveller；
3. traditional ideoelectotype human resources are big:
Waiter buys dish by part, is placed on corresponding from constituency, and customer freely selects dish；
4. speed of settling accounts is slow, error-prone: artificial cash register is relied on, is valuated one by one by cashier's mental arithmetic or number tableware, speed
Slowly, error-prone, it be easy to cause and stands in a long queue, causes dispute.
Summary of the invention
Cafeteria's intelligence settlement system based on depth convolutional network that the main purpose of the present invention is to provide a kind of, can
Effectively to solve the problems in background technique.
To achieve the above object, the technical scheme adopted by the invention is as follows:
A kind of cafeteria's intelligence settlement system based on depth convolutional neural networks, is instructed including the use of convolutional neural networks
Practice the identification and counting machine and vegetable classifier, system Automatic-settlement step for generating bowl.
Further, convolutional neural networks training and generate bowl identification and counting machine the following steps are included:
Step 1 makes training image sample: a variety of bowls of sample image data is chosen from cafeteria's tableware picture library；
Step 2 sample image regularization: carrying out the operation of image size conversion for original sample image according to step 1, will
Its resolution ratio is unified to be converted into floating point vector data to 416*416, and by image data；
The initialization of step 3 neural network: a multilayer depth convolutional neural networks are constructed according to step 2, by initial
Setting weighting parameter initializes each layer of neural network；
Step 4 neural network input: according to step 3 by the floating point vector data and figure of marked image pattern
The flag data of picture is input to neural network input layer, completes the input layer of the depth convolutional neural networks of initialization；
Step 5 trains neural network: according to step 4 by the data of neural network input layer, by multilayer convolutional Neural
The calculating of network obtains the output vector of the last layer；
Step 6 Neural Network Optimization: according to step 5 by constantly reversely modifying each layer weighting parameter, reach optimization
The ideal output vector of output vector and marking class is compared, obtains error amount, utilize error by the effect of neural network
Value reversely corrects the weighting parameter of each layer neural network automatically, completes one group of training；
Step 7 neural metwork training is completed: generating a depth convolution for being identified and being counted to bowl according to step 6
Neural network.
Further, convolutional neural networks training and generate vegetable classifier the following steps are included:
Step steady training image sample: choosing multiple groups sample image data from cafeteria's vegetable picture library, each
Class vegetable is one group of sample；
Step 2 sample image regularization: carrying out the operation of image size conversion for original sample image according to step 1, will
Its resolution ratio is unified to be converted into floating point vector data to 416*416, and by image data；
The initialization of step 3 neural network: a multilayer depth convolutional neural networks are constructed according to step 2, by initial
Setting weighting parameter initializes each layer of neural network；
Step 4 neural network input: according to step 3 by the floating point vector data and sample of marked image pattern
Input of this flag information as neural network, is input to the input layer of neural network.
Step 5 trains neural network: according to step 4 by the data of neural network input layer, i.e. training sample data, passing through
The calculating for crossing multilayer convolutional neural networks obtains the output vector of the last layer；
Step 6 Neural Network Optimization: network is optimized by reversely modifying each layer weighting parameter by nerve net according to step 5
Network recognition effect.The ideal output vector of output vector and marking class is compared, obtains error amount, it is anti-using error amount
To the weighting parameter for correcting each layer neural network automatically, one group of training is completed；
Step 7 neural metwork training is completed: cafeteria's difference vegetable can be known by generating one according to step 6
Other depth convolutional neural networks classifier.
Further, the Automatic-settlement step includes the following steps:
Pair step 1 obtains menu name: the relevant information provided in conjunction with above two deep neural network identifier, i.e.,
The identification of bowl and count information and the recognition data and information to vegetable obtain menu name；
Step 2 obtains vegetable price: the vegetable price number of the menu name obtained according to step 1 and cafeteria's offer
According to table, related vegetable price is obtained；
Step 3 consumption clearing: the related vegetable price provided according to step 2 settles accounts client's consumption, prompts visitor
Family is paid dues, and paying dues successfully can normally have meal.
Further, the number of plies of the multilayer depth convolutional neural networks is 26 layers, the depth convolutional neural networks packet
24 convolutional layers and 2 full articulamentums are included, multi-level convolutional neural networks have preferably classifying quality and identification robustness
Raising, network is by reversely modifying each layer weighting parameter optimization neural network recognition effect.It is trained simultaneously in convolutional neural networks
Before generating criteria classification device, there are one training effect is verified, if effect is undesirable, by more image patterns
Vector data input depth convolutional neural networks are trained.By a large amount of training image sample, constantly automatically update excellent
Change each layer weighting parameter of each neural network to complete to train if effect is ideal.
Further, the system comprises convolutional neural networks training module, intelligent identification module, intelligent settlement module,
Camera head monitor module；The camera face settles accounts service plate, and camera is electrically connected with convolutional neural networks training module, and will
The transmission of video images of shooting carries out vegetable detection into convolutional neural networks training module, by convolutional neural networks training module
Identification；And it is managed by the main control module finishing man-machine interaction in settlement module, and count consumption price network input layer data dimension
It is high, the network number of plies is more, network passes through and reversely modifies each layer weighting parameter optimization neural network during training convolutional network
Discrimination, in convolutional neural networks, the input of convolutional layer is that the pixel value of all samples combines.A combination thereof form is as follows:
X=[x0 x1 ... xN-1]
It is wherein the pixel value vector of the sample.If the pixel of sample is bigger, during neural metwork training more
It is easy to extract feature vector, but calculation amount also will increase simultaneously,
Convolutional neural networks need the operation completed are as follows:
In formula
Referred to as " weighted vector " of neuron；Referred to as " power function " of neuron；Referred to as " the net input " of neuron；Claim
For " output " of neuron；Referred to as " threshold value " of neuron.The process of network training is exactly to pass through constantly to optimize weighted vector
And threshold value, have reached the true mark classification for making the output of neural network close to input sample.
The convolutional calculation process of one convolutional layer are as follows: deconvoluted the image inputted with a trainable filter fx
(being input picture at C1 layers, convolutional layer input later is then the convolution characteristic pattern of preceding layer), passes through an activation primitive (one
As use Sigmoid function), then plus a biasing bx, obtain convolutional layer Cx.Concrete operation such as following formula, Mj is in formula
The value of input feature vector figure:
The process of pond layer includes: that m pixel (m is to be manually set) summation of every neighborhood becomes a pixel, is then led to
Scalar Wx+1 weighting is crossed, biasing bx+1 is further added by, Feature Mapping figure is then generated by activation primitive Sigmoid.It is flat from one
The mapping of face to next plane can be regarded as making convolution algorithm, and S layers are considered as fuzzy filter, play quadratic character
The effect of extraction.Spatial resolution between hidden layer and hidden layer is successively decreased, and number of planes contained by every layer is incremented by, and can be used for examining in this way
Survey more characteristic informations.For sub-sampling layer, there is N number of input feature vector figure, just has N number of output characteristic pattern, only each
The size of characteristic pattern has obtained corresponding change, concrete operation such as following formula, and down () indicates down-sampling function in formula.
Backpropagation optimizes weighting parameter: the error reversely returned can regard the sensitivity of the base of each neuron as,
Shown in it is defined as follows:
Wherein the error on whole training sets is the summation of the error of each training sample, such as total c class, altogether N number of instruction
Practice sample.
Shown herein as the kth dimension of the corresponding label of n-th of sample.Indicate k-th of the corresponding network output of n-th of sample
Output.Backpropagation is exactly with this following relational expression:
HereIndicate each element multiplication.The sensitivity of the neuron of output layer is different:
Finally, carrying out right value update with delta (i.e. δ) rule to each neuron.It is specifically exactly to be given to one
Fixed neuron obtains its input, is then zoomed in and out with the delta of this neuron (i.e. δ).With the form table of vector
It states and is exactly, for l layers, error is the input of this layer (equal to upper for the derivative of each weight (group is combined into matrix) of the layer
One layer of output) multiplication cross with the sensitivity (form that the δ of this layer of each neuron is combined into a vector) of this layer.Then
To partial derivative be exactly that the weight of neuron of this layer has updated multiplied by a negative learning rate:
Further, the convolutional neural networks training module include Image Acquisition labeling module, image import modul and
Image Acquisition labeling module: GPU computing module chooses multiple groups sample image data, and labeled bracketing, figure from image pattern library
As import modul: one group of sample image data grouping successively being imported memory according to hardware actual disposition, and is input to multilayer mind
Input layer through network, GPU computing module: calculating the operations such as convolution, the pond of each layer of convolutional neural networks, obtains next layer
Input feature vector indicate data (feature vector), the reality output of the output character representation data of middle layer and output layer is special
Sign indicates that the feature vector element in data compared with the element in ideal output feature vector, obtains intermediate each layer error respectively
And output layer error；The reality output vector for calculating network, by the element in the element and object vector in output vector
It is compared, calculates output error；Error is also calculated for the hidden unit of middle layer；Calculate each weight adjustment amount and
The adjustment amount of threshold value；Adjust weight and adjustment threshold value；After undergoing n times iteration, whether judge index meets required precision, such as
Fruit is unsatisfactory for, then continues to repeat the above process；Each layer weighting parameter is updated by constantly training, e-learning is become better and better, from
Ideal function is more and more closer, when meeting the error amount being previously set, terminates convolutional neural networks training, and by each layer weight etc.
File is written in parameter, generates classifier file.
Compared with prior art, the invention has the following beneficial effects: this kind invention designs rationally, easy to use, is based on
The video image intelligent identification technology of depth convolutional neural networks becomes one of the research and development focus of field of machine vision gradually, no
Disconnected is employed in daily life, so-called " digital image recognition technology ", is exactly to be realized using deep learning algorithm
The key feature that extracts from video image stream indicates information, and is provided for user using this key feature information valuable
Service, chief value are embodied in substitution human eye and differentiate the business scopes such as things, reduce human input, improve working efficiency, than
Such as, number identification during, the identification for handwritten form is always a problem, and its postcode identification,
Banking etc. has relatively broad application, but its font form changes greatly, and causes to improve the discrimination to it
Become a problem, accurately identifies there is bigger difficulty, and after using the nerual network technique in artificial intelligence, it is
System can use the study of neural network and fast parallel function realizes the quick identification to handwritten numeral, strong raising phase
The working efficiency in the field of using is closed, cafeteria's intelligence settlement system based on depth convolutional neural networks is deep by building multilayer
Degree convolutional neural networks are simultaneously trained neural network, by reversely modifying each layer weighting parameter optimization nerve in training process
Network Recognition effect, system analyze video image by two classifiers, bowl classification counter and vegetable classifier,
It detects vegetable to be settled accounts and is settled accounts, after the completion of clearing, prompt person sponging on an aristocrat's payment, system improves the clearing speed of cafeteria
Degree is reduced as calculating dining room monetary losses caused by mistake when staff's clearing, mitigating restaurant operation person works amount, is fitted
Conjunction is widely popularized.
Figure of description
Fig. 1 is bowl identifier depth convolutional neural networks training step of the present invention；
Fig. 2 is vegetable identifier depth convolutional neural networks training step of the present invention；
Fig. 3 is cafeteria's intelligence settlement system work flow diagram of the present invention；
Fig. 4 is the number of iterations sample size of the invention.
Specific embodiment
Technical means, creative features, achievable purpose and effectiveness to realize this practical invention are easy to understand, below
In conjunction with specific embodiment, the present invention is further explained.
Embodiment 1:
Referring to Fig. 1, a kind of bowl identification and counter based on depth convolutional neural networks, including depth convolutional Neural net
Network training step and bowl identification calculate step；
The convolutional neural networks training mainly includes the next steps；
Step 1 obtains training sample: the training sample of the bowl of different patterns is obtained from image library.Every group of sample image
Diversity is required, samples sources carry out classification marker in different sample spaces, all samples.
Step 2 sample image regularization: original sample image is subjected to the operation of image size conversion, by its resolution ratio pressure
It is reduced to 416*416, and converts floating point vector data for image data.
The initialization of step 3 neural network: 26 layer depth convolutional neural networks of building one, this network is respectively by 24
Convolutional layer, 2 full articulamentums are constituted, and each layer of neural network is initialized with initial setting weighting parameter.
The input of step 4 neural network: the sample that will be marked is input in neural network and is largely trained, learns life
At the characteristic model of standard qualification.
Step 5 neural metwork training: initialization is completed into the floating point vector data input of marked image pattern
The input layers of depth convolutional neural networks obtain the output vector of the last layer by the calculating of 26 layers of convolutional neural networks.
Step 6 Neural Network Optimization: the ideal output vector of output vector and marking class is compared, error is obtained
Value.It reversely corrects the weighting parameter of each layer neural network automatically using error amount, completes one group of training.
Step 7 repetition training network and each two step of layer weighting parameter is reversely modified, by all image pattern vector numbers
It is trained according to input depth convolutional neural networks.By a large amount of training image sample, each mind of optimization is constantly automatically updated
Through each layer weighting parameter of network.
After the completion of step 8 training, the classifier of a depth convolutional neural networks is generated.
Embodiment 2:
Referring to Fig. 2, a kind of vegetable identifier based on depth convolutional neural networks, including depth convolutional neural networks instruction
Practice step and bowl identification calculates step；
The convolutional neural networks training mainly includes the next steps；
Step 1 obtains training sample: the training sample of the bowl of different patterns is obtained from image library.Every group of sample image
Diversity is required, samples sources carry out classification marker in different sample spaces, all samples.
Step 2 sample image regularization: original sample image is subjected to the operation of image size conversion, by its resolution ratio pressure
It is reduced to 416*416, and converts floating point vector data for image data.
The initialization of step 3 neural network: 26 layer depth convolutional neural networks of building one, this network is respectively by 24
Convolutional layer, 2 full articulamentums are constituted, and each layer of neural network is initialized with initial setting weighting parameter.
The input of step 4 neural network: the sample that will be marked is input in neural network and is largely trained, learns life
At the characteristic model of standard qualification.
Step 5 neural metwork training: initialization is completed into the floating point vector data input of marked image pattern
The input layers of depth convolutional neural networks obtain the output vector of the last layer by the calculating of 26 layers of convolutional neural networks.
Step 6 Neural Network Optimization: the ideal output vector of output vector and marking class is compared, error is obtained
Value.It reversely corrects the weighting parameter of each layer neural network automatically using error amount, completes one group of training.
Step 7 repetition training network and each two step of layer weighting parameter is reversely modified, by all image pattern vector numbers
It is trained according to input depth convolutional neural networks.By a large amount of training image sample, each mind of optimization is constantly automatically updated
Through each layer weighting parameter of network.
After the completion of step 8 training, the classifier of a depth convolutional neural networks is generated.
Embodiment 3
Referring to Fig. 3, cafeteria's intelligence settlement system.The settlement system the following steps are included:
Step 1 obtains service plate picture to be settled accounts: obtaining service plate picture to be settled accounts by camera, and picture is reached mind
Through network.
The identification of step 2 bowl is counted with number: using picture to be settled accounts as the input of bowl identification network, carrying out the identification of bowl
And calculating.That is, identifying the counting of the position of bowl and statistics bowl in picture.
The identification of step 3 vegetable: using picture to be settled accounts as the input of vegetable identification network, the identification of vegetable is carried out, simultaneously
Obtain coordinate information of the vegetable in picture.
Step 4 obtains menu name: related to two kinds of deep neural network identifiers offers of step 3 in conjunction with step 2
Information obtains menu name that is, to the identification of bowl and count information and to the recognition data and information of vegetable.
Step 5 obtains vegetable price: the vegetable price data table of the menu name of acquisition and cafeteria's offer obtains
Related vegetable price；
Step 6 consumption clearing: the related vegetable price of offer settles accounts client's consumption.Prompt client pays dues, and pays dues
Success can normally have meal.
It can be obtained by embodiment 1-3, intelligently clearing are for the cafeteria provided by the invention based on depth convolutional neural networks
System, is compared with the element in object vector by the element in output vector, calculates output error；And according to overall error
Function judges that further adjusting weight obtains good classifier come the relationship with error threshold, has that generalization ability is strong, robust
The advantages that property is strong, Detection accuracy is high.Since the feature extraction of convolutional neural networks is learnt to mention automatically by training data
It takes, so when using convolutional neural networks, substantially increases image characteristics extraction diversity, feature complexity is closer to true
It is situation, has extremely important meaning to subsequent detection accuracy of identification.And Automatic Feature Extraction movement is by training, big
It measures and is learnt in the sample data of label, all learn to arrive by all characteristic informations as far as possible, to reach high accurate
Rate.
The above display describes basic principles and main features and advantages of the present invention of the invention.The technology people of the industry
Member is it should be appreciated that the present invention is not limited to the above embodiments, and the above embodiments and description only describe this hairs
Bright principle, without departing from the spirit and scope of the present invention, various changes and improvements may be made to the invention, these variations
It all fall within the protetion scope of the claimed invention with improvement.The claimed scope of the invention is by appended claims and its waits
Effect object defines.
Claims (7)
1. a kind of cafeteria's intelligence settlement system based on depth convolutional neural networks, which is characterized in that including the use of convolution
The identification and counting machine and vegetable classifier, system Automatic-settlement step of neural metwork training generation bowl.
2. a kind of cafeteria's intelligence settlement system based on depth convolutional neural networks according to claim 1, feature
It is；Convolutional neural networks training and generate bowl identification and counting machine the following steps are included:
Step 1 makes training image sample: a variety of bowls of sample image data is chosen from cafeteria's tableware picture library；
Step 2 sample image regularization: original sample image is carried out by the operation of image size conversion according to step 1, by image
Resolution ratio is unified to be converted into floating point vector data to 416*416, and by image data；
The initialization of step 3 neural network: a multilayer depth convolutional neural networks are constructed according to step 2, by initially setting
Weighting parameter initializes each layer of neural network；
Step 4 neural network input: according to step 3 by the floating point vector data of marked image pattern and image
Flag data is input to neural network input layer, completes the input layer of the depth convolutional neural networks of initialization；
Step 5 trains neural network: according to step 4 by the data of neural network input layer, by multilayer convolutional neural networks
Calculating, obtain the output vector of the last layer；
Step 6 Neural Network Optimization: according to step 5 by constantly reversely modifying each layer weighting parameter, reach optimization nerve
The ideal output vector of output vector and marking class is compared, obtains error amount by the effect of network, anti-using error amount
To the weighting parameter for correcting each layer neural network automatically, one group of training is completed；
Step 7 neural metwork training is completed: generating a depth convolutional Neural for being identified and being counted to bowl according to step 6
Network.
3. a kind of cafeteria's intelligence settlement system based on depth convolutional neural networks according to claim 1, feature
It is；Convolutional neural networks training and generate vegetable classifier the following steps are included:
Step steady training image sample: multiple groups sample image data, every one kind dish are chosen from cafeteria's vegetable picture library
Product are one group of sample；
Step 2 sample image regularization: original sample image is carried out by the operation of image size conversion according to step 1, by image
Resolution ratio is unified to be converted into floating point vector data to 416*416, and by image data；
The initialization of step 3 neural network: a multilayer depth convolutional neural networks are constructed according to step 2, by initially setting
Weighting parameter initializes each layer of neural network；
Step 4 neural network input: according to step 3 by the floating point vector data and sample mark of marked image pattern
Input of the will information as neural network, is input to the input layer of neural network；
Step 5 trains neural network: according to step 4 by the data of neural network input layer, i.e. training sample data, through excessive
The calculating of layer convolutional neural networks, obtains the output vector of the last layer；
Step 6 Neural Network Optimization: network is known by reversely modifying each layer weighting parameter optimization neural network according to step 5
The ideal output vector of output vector and marking class is compared, obtains error amount by other effect, reversed certainly using error amount
The weighting parameter of dynamic each layer neural network of amendment, completes one group of training；
Step 7 neural metwork training is completed: can be identified to cafeteria's difference vegetable according to step 6 generation one
Depth convolutional neural networks classifier.
4. a kind of cafeteria's intelligence settlement system based on depth convolutional neural networks according to claim 1, feature
It is；The Automatic-settlement step includes the following steps:
Step 1 obtains menu name: the relevant information provided in conjunction with above two deep neural network identifier, i.e., to bowl
Identification and count information and the recognition data and information to vegetable, obtain menu name；
Step 2 obtains vegetable price: the vegetable price data of the menu name obtained according to step 1 and cafeteria's offer
Table obtains related vegetable price；
Step 3 consumption clearing: the related vegetable price provided according to step 2 settles accounts client's consumption, and client is prompted to hand over
Take, paying dues successfully can normally have meal.
5. a kind of cafeteria's intelligence settlement system based on depth convolutional neural networks according to claim 1, feature
It is, the number of plies of the multilayer depth convolutional neural networks is 26 layers, and the depth convolutional neural networks include 24 convolutional layers
With 2 full articulamentums.
6. a kind of cafeteria's intelligence settlement system based on depth convolutional neural networks according to claim 1, feature
It is, the system comprises convolutional neural networks training module, intelligent identification module, intelligent settlement modules, camera head monitor mould
Block；The camera face settles accounts service plate, and camera is electrically connected with convolutional neural networks training module.
7. a kind of cafeteria's intelligence settlement system based on depth convolutional neural networks according to claim 6, feature
It is, the convolutional neural networks training module includes Image Acquisition labeling module, image import modul and GPU computing module.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811492955.8A CN109377205A (en) | 2018-12-06 | 2018-12-06 | A kind of cafeteria's intelligence settlement system based on depth convolutional network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811492955.8A CN109377205A (en) | 2018-12-06 | 2018-12-06 | A kind of cafeteria's intelligence settlement system based on depth convolutional network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109377205A true CN109377205A (en) | 2019-02-22 |
Family
ID=65376672
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811492955.8A Pending CN109377205A (en) | 2018-12-06 | 2018-12-06 | A kind of cafeteria's intelligence settlement system based on depth convolutional network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109377205A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110059551A (en) * | 2019-03-12 | 2019-07-26 | 五邑大学 | A kind of automatic checkout system of food based on image recognition |
| CN110399804A (en) * | 2019-07-01 | 2019-11-01 | 浙江师范大学 | A kind of food inspection recognition methods based on deep learning |
| CN112149577A (en) * | 2020-09-24 | 2020-12-29 | 信雅达系统工程股份有限公司 | Intelligent settlement system based on neural network image recognition |
| CN112699822A (en) * | 2021-01-05 | 2021-04-23 | 浪潮云信息技术股份公司 | Restaurant dish identification method based on deep convolutional neural network |
| CN113128296A (en) * | 2019-12-31 | 2021-07-16 | 重庆傲雄在线信息技术有限公司 | Electronic handwriting signature fuzzy labeling recognition system |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103971471A (en) * | 2014-05-08 | 2014-08-06 | 上海海事大学 | Automatic pricing method and system for dinner plates based on color discrimination |
| CN105741139A (en) * | 2016-01-31 | 2016-07-06 | 上海海角网络科技有限公司 | Automatic charging method for self-service restaurant and implementation apparatus thereof |
| CN206292857U (en) * | 2016-12-29 | 2017-06-30 | 浙江正元智慧科技股份有限公司 | Wisdom dinner table based on machine vision |
| CN107122730A (en) * | 2017-04-24 | 2017-09-01 | 乐金伟 | Free dining room automatic price method |
| CN108509912A (en) * | 2018-04-03 | 2018-09-07 | 深圳市智绘科技有限公司 | Multipath network video stream licence plate recognition method and system |
| CN108831530A (en) * | 2018-05-02 | 2018-11-16 | 杭州机慧科技有限公司 | Vegetable nutrient calculation method based on convolutional neural networks |
-
2018
- 2018-12-06 CN CN201811492955.8A patent/CN109377205A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103971471A (en) * | 2014-05-08 | 2014-08-06 | 上海海事大学 | Automatic pricing method and system for dinner plates based on color discrimination |
| CN105741139A (en) * | 2016-01-31 | 2016-07-06 | 上海海角网络科技有限公司 | Automatic charging method for self-service restaurant and implementation apparatus thereof |
| CN206292857U (en) * | 2016-12-29 | 2017-06-30 | 浙江正元智慧科技股份有限公司 | Wisdom dinner table based on machine vision |
| CN107122730A (en) * | 2017-04-24 | 2017-09-01 | 乐金伟 | Free dining room automatic price method |
| CN108509912A (en) * | 2018-04-03 | 2018-09-07 | 深圳市智绘科技有限公司 | Multipath network video stream licence plate recognition method and system |
| CN108831530A (en) * | 2018-05-02 | 2018-11-16 | 杭州机慧科技有限公司 | Vegetable nutrient calculation method based on convolutional neural networks |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110059551A (en) * | 2019-03-12 | 2019-07-26 | 五邑大学 | A kind of automatic checkout system of food based on image recognition |
| CN110399804A (en) * | 2019-07-01 | 2019-11-01 | 浙江师范大学 | A kind of food inspection recognition methods based on deep learning |
| CN113128296A (en) * | 2019-12-31 | 2021-07-16 | 重庆傲雄在线信息技术有限公司 | Electronic handwriting signature fuzzy labeling recognition system |
| CN112149577A (en) * | 2020-09-24 | 2020-12-29 | 信雅达系统工程股份有限公司 | Intelligent settlement system based on neural network image recognition |
| CN112699822A (en) * | 2021-01-05 | 2021-04-23 | 浪潮云信息技术股份公司 | Restaurant dish identification method based on deep convolutional neural network |
| CN112699822B (en) * | 2021-01-05 | 2023-05-30 | 浪潮云信息技术股份公司 | Restaurant dish identification method based on deep convolutional neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109377205A (en) | A kind of cafeteria's intelligence settlement system based on depth convolutional network | |
| Wu et al. | Convolution neural network based transfer learning for classification of flowers | |
| Yu et al. | Mixed pooling for convolutional neural networks | |
| Yuan et al. | Gated CNN: Integrating multi-scale feature layers for object detection | |
| Hussain et al. | Automatic fruit recognition based on dcnn for commercial source trace system | |
| CN109886121A (en) | A kind of face key independent positioning method blocking robust | |
| CN109359538A (en) | Training method, gesture identification method, device and the equipment of convolutional neural networks | |
| CN108304826A (en) | Facial expression recognizing method based on convolutional neural networks | |
| CN109614996A (en) | The recognition methods merged based on the weakly visible light for generating confrontation network with infrared image | |
| CN105894045A (en) | Vehicle type recognition method with deep network model based on spatial pyramid pooling | |
| CN109166100A (en) | Multi-task learning method for cell count based on convolutional neural networks | |
| CN104361313B (en) | A kind of gesture identification method merged based on Multiple Kernel Learning heterogeneous characteristic | |
| CN110148120A (en) | A kind of disease intelligent identification Method and system based on CNN and transfer learning | |
| CN106778506A (en) | A kind of expression recognition method for merging depth image and multi-channel feature | |
| CN107122798A (en) | Chin-up count detection method and device based on depth convolutional network | |
| Nandini et al. | Dropout technique for image classification based on extreme learning machine | |
| CN106991386A (en) | A kind of gesture identification method based on depth residual error network | |
| CN109583419A (en) | A kind of emotional prediction system based on depth convolutional network | |
| CN109272107A (en) | A method of improving the number of parameters of deep layer convolutional neural networks | |
| CN104809469A (en) | Indoor scene image classification method facing service robot | |
| CN110222696A (en) | A method of plant leaf blade disease identification is carried out using GPCNNs and ELM | |
| CN110059586A (en) | A kind of Iris Location segmenting system based on empty residual error attention structure | |
| CN106372597B (en) | CNN Vehicle Detection method based on adaptive contextual information | |
| CN107491729B (en) | Handwritten digit recognition method based on cosine similarity activated convolutional neural network | |
| CN110334715A (en) | A kind of SAR target identification method paying attention to network based on residual error |
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: 20191223 Address after: 518000 third floor, r1-a modern building, No.23, gaoxinnan 4th Road, South District, Nanshan District, Shenzhen City, Guangdong Province Applicant after: Shenzhen Shengli Wantong Technology Co., Ltd Address before: 518055 Guangdong city of Shenzhen province Nanshan District Xili Street Science Park North Qi Min Road No. 1 Betel building 3 floor Applicant before: Shenzhen rice washing technology Co., Ltd. |
|
| TA01 | Transfer of patent application right |