CN112749731A - Bill quantity identification method and system based on deep neural network - Google Patents
Bill quantity identification method and system based on deep neural network Download PDFInfo
- Publication number
- CN112749731A CN112749731A CN202011456986.5A CN202011456986A CN112749731A CN 112749731 A CN112749731 A CN 112749731A CN 202011456986 A CN202011456986 A CN 202011456986A CN 112749731 A CN112749731 A CN 112749731A
- Authority
- CN
- China
- Prior art keywords
- bill
- image
- mixed
- shot
- dimensional
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/047—Probabilistic or stochastic 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)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Probability & Statistics with Applications (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a bill quantity identification method and a bill quantity identification system based on a deep neural network, which are used for sequentially carrying out graying processing, binarization processing, normalization processing and two-dimensional image mapping one-dimensional data processing on an acquired two-dimensional mixed-shooting bill original image, greatly reducing the data volume while retaining image information, reducing the time waste caused by data transmission and improving the user experience; by designing an OCTC (open code transfer coefficient) model, introducing one-dimensional convolution operation to perform feature extraction and image category calculation on image data, increasing the model receptive field by using a plurality of small-size convolution kernels, reducing the model parameter number while ensuring the experimental effect, realizing the effect of a lightweight model, and facilitating the storage and use of a user; the method provided by the invention can enable a user to receive the judgment result of the number of the bills in the image while uploading the image, improves the user experience of the mixed shooting bill identification system, and simultaneously helps the system to perform subsequent bill target detection and content identification tasks, thereby realizing the real-time work of the system.
Description
Technical Field
The invention relates to the technical field of bill processing, in particular to a bill quantity identification method based on a deep neural network.
Background
The bill quantity identification method is used for the early bill prejudging stage of the mixed shot bill identification system. At present, aiming at bill recognition, bill information in an image is mainly recognized based on technologies such as AlexNet, Fast-RCNN and OCR, the method has good stability and expression effect after long-term use and verification, but due to the fixity of a model, the application condition is limited, and corresponding adjustment cannot be carried out according to the actual condition. In addition, the bill recognition system performs bill analysis by introducing target detection operation, and since bills shot in a real scene have problems of overlapping, inclination, low resolution and the like, the area which can be detected by the model is influenced, so that the model effect is reduced.
Therefore, a method capable of rapidly and accurately identifying the number of bills is needed to assist the detection model in performing monitoring area prejudgment and improve the detection accuracy.
Disclosure of Invention
The invention provides a bill quantity identification method and system based on a deep neural network, and aims to solve the problem of how to quickly identify the bill quantity of an image.
In order to solve the above problem, according to an aspect of the present invention, there is provided a method for identifying a number of tickets based on a deep neural network, the method including:
acquiring mixed shot bill original images corresponding to each group of mixed bills in a plurality of groups of different mixed bills, and preprocessing each acquired mixed shot bill original image to acquire a mixed shot bill processing image corresponding to each mixed shot bill original image;
constructing a One-dimensional Convolutional bill Classification (OCTC) model based on a deep neural network, and training the OCTC model by using each acquired mixed-shooting bill processing image to determine an OCTC optimal model;
acquiring a mixed shot bill original image to be detected, and processing the mixed shot bill original image to be detected to acquire a mixed shot bill processing image to be detected;
and carrying out bill quantity recognition on the mixed shot bill processing image to be detected by utilizing the OCTC optimal model so as to obtain a bill quantity recognition result.
Preferably, the types of the hybrid ticket include: value-added tax invoices, train tickets, quota tickets, business licenses, roll tickets, identification cards, taxi tickets, motor vehicle sales and airline travel tickets.
Preferably, the method for preprocessing the mixed shot bill original image to obtain the mixed shot bill processed image corresponding to the mixed shot bill original image comprises the following steps:
and sequentially carrying out graying processing, binarization processing, normalization processing and two-dimensional image mapping one-dimensional data processing on the mixed shot bill original image so as to obtain a mixed shot bill processing image corresponding to the mixed shot bill original image.
Preferably, the normalization processing and the two-dimensional image mapping one-dimensional data processing include:
wherein, Xi,jFor mixedly photographing original pixel value, mu, of original image of billiFor taking a mixed shot of mean values, sigma, of line pixel values in the original image of the documentiFor the purpose of the corresponding variance, the variance is,pixel values after row normalization; the normalized pixels are accumulated to obtain a mapped one-dimensional row vector, namely one-dimensional mapping data
Preferably, the one-dimensional convolution note classification OCTC model includes: three convolution layers, three maximum pooling layers, two full-connection layers and one softmax layer; wherein, each layer uses five convolution kernels to respectively calculate input data so as to obtain convolution characteristic image data, and the number of the convolution kernels of each layer is [64,128 and 256 ]; the method comprises the steps of adopting a structure of two layers of fully-connected neural networks to realize identification and classification of the number of the mixed-shot bills, enabling the number of neurons of the two fully-connected layers to be [512,128], utilizing a first fully-connected layer to carry out linear integration on feature map data obtained by calculation of a convolutional layer, utilizing a ReLU activation function to carry out nonlinear conversion, and utilizing a second fully-connected layer to carry out calculation of high-dimensional features; and finally, predicting the image category probability through a softmax layer activation function, and returning the bill quantity result in the current image.
According to another aspect of the present invention, there is provided a deep neural network-based ticket quantity recognition system, the system including:
the first data processing unit is used for acquiring mixed shot bill original images corresponding to each group of mixed bills in a plurality of groups of different mixed bills, and preprocessing each acquired mixed shot bill original image to acquire a mixed shot bill processing image corresponding to each mixed shot bill original image;
the optimal model determining unit is used for constructing a one-dimensional convolution note classification OCTC model based on a deep neural network, and training the OCTC model by utilizing each acquired mixed-shooting note processing image to determine an OCTC optimal model;
the second data processing unit is used for acquiring a mixed shot bill original image to be detected and processing the mixed shot bill original image to be detected so as to acquire a mixed shot bill processing image to be detected;
and the bill quantity identification unit is used for identifying the bill quantity of the to-be-detected mixed shot bill processing image by using the OCTC optimal model so as to obtain a bill quantity identification result.
Preferably, the types of the hybrid ticket include: value-added tax invoices, train tickets, quota tickets, business licenses, roll tickets, identification cards, taxi tickets, motor vehicle sales and airline travel tickets.
Preferably, the first data processing unit and the second data processing unit pre-process the mixed shot bill original image to obtain a mixed shot bill processed image corresponding to the mixed shot bill original image, including:
and sequentially carrying out graying processing, binarization processing, normalization processing and two-dimensional image mapping one-dimensional data processing on the mixed shot bill original image so as to obtain a mixed shot bill processing image corresponding to the mixed shot bill original image.
Preferably, the normalization processing and the two-dimensional image mapping one-dimensional data processing include:
wherein, Xi,jFor mixedly photographing original pixel value, mu, of original image of billiFor taking a mixed shot of mean values, sigma, of line pixel values in the original image of the documentiFor the purpose of the corresponding variance, the variance is,pixel values after row normalization; the normalized pixels are accumulated to obtain a mapped one-dimensional row vector, namely one-dimensional mapping data
Preferably, the one-dimensional convolution note classification OCTC model includes: three convolution layers, three maximum pooling layers, two full-connection layers and one softmax layer; wherein, each layer uses five convolution kernels to respectively calculate input data so as to obtain convolution characteristic image data, and the number of the convolution kernels of each layer is [64,128 and 256 ]; the method comprises the steps of adopting a structure of two layers of fully-connected neural networks to realize identification and classification of the number of the mixed-shot bills, enabling the number of neurons of the two fully-connected layers to be [512,128], utilizing a first fully-connected layer to perform linear integration on feature image data obtained by calculation of a convolutional layer, utilizing a ReLU activation function to perform nonlinear conversion, and utilizing a second fully-connected layer to perform high-dimensional feature calculation; and finally, predicting the image category probability through a softmax layer activation function, and returning the bill quantity result in the current image.
The invention provides a bill quantity identification method and system based on a deep neural network, which are used for sequentially carrying out graying processing, binarization processing, normalization processing and two-dimensional image mapping one-dimensional data processing on an acquired two-dimensional mixed-shooting bill original image, greatly reducing the data volume while retaining image information, reducing the time waste caused by data transmission and improving the user experience; by designing an OCTC (open code transfer coefficient) model, introducing one-dimensional convolution operation to perform feature extraction and image category calculation on image data, increasing the model receptive field by using a plurality of small-size convolution kernels, reducing the model parameter number while ensuring the experimental effect, realizing the effect of a lightweight model, and facilitating the storage and use of a user; the method provided by the invention can enable a user to receive the judgment result of the number of the bills in the image while uploading the image, improves the user experience of the mixed shooting bill identification system, and simultaneously helps the system to perform subsequent bill target detection and content identification tasks, thereby realizing the real-time work of the system.
Drawings
A more complete understanding of exemplary embodiments of the present invention may be had by reference to the following drawings in which:
FIG. 1 is a flow diagram of a method 100 for deep neural network based document quantity identification, according to an embodiment of the present invention;
FIG. 2 is a flow chart of a method for rapid bill quantity recognition based on a deep neural network according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a captured raw image of a hybrid ticket according to an embodiment of the present invention;
fig. 4 is a schematic diagram of a grayscale image obtained after performing a graying process according to an embodiment of the present invention;
fig. 5 is a schematic diagram of a black-and-white image obtained by performing binarization processing on a gray-scale image according to an embodiment of the invention;
FIG. 6 is a block diagram of an OCTC model according to an embodiment of the invention;
FIG. 7 is an exemplary diagram of an implementation of expanding a model receptive field by stacking multiple small convolution kernels according to an embodiment of the present invention;
FIG. 8 is a schematic illustration of the result output of a bill identification method according to an embodiment of the present invention;
FIG. 9 is a schematic diagram of a hardware structure of a deep neural network-based bill quantity recognition apparatus according to an embodiment of the present invention;
fig. 10 is a schematic structural diagram of a deep neural network-based bill quantity recognition system 1000 according to an embodiment of the present invention.
Detailed Description
The exemplary embodiments of the present invention will now be described with reference to the accompanying drawings, however, the present invention may be embodied in many different forms and is not limited to the embodiments described herein, which are provided for complete and complete disclosure of the present invention and to fully convey the scope of the present invention to those skilled in the art. The terminology used in the exemplary embodiments illustrated in the accompanying drawings is not intended to be limiting of the invention. In the drawings, the same units/elements are denoted by the same reference numerals.
Unless otherwise defined, terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. Further, it will be understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and will not be interpreted in an idealized or overly formal sense.
Fig. 1 is a flowchart of a method 100 for identifying a number of tickets based on a deep neural network according to an embodiment of the present invention. As shown in fig. 1, the method for identifying the number of tickets based on the deep neural network provided by the embodiment of the invention can enable a user to receive a judgment result of the number of tickets in an image while uploading the image, thereby improving the user experience of a hybrid-shooting ticket identification system, assisting the system in performing subsequent ticket target detection and content identification tasks, and realizing real-time work of the system. In the method 100 for identifying the number of bills based on the deep neural network, which is provided by the embodiment of the invention, starting from step 101, in step 101, a mixed shot bill original image corresponding to each group of mixed bills in a plurality of groups of different mixed bills is obtained, and each obtained mixed shot bill original image is preprocessed to obtain a mixed shot bill processed image corresponding to each mixed shot bill original image.
Preferably, the types of the hybrid ticket include: value-added tax invoices, train tickets, quota tickets, business licenses, roll tickets, identification cards, taxi tickets, motor vehicle sales and airline travel tickets.
Preferably, the method for preprocessing the mixed shot bill original image to obtain the mixed shot bill processed image corresponding to the mixed shot bill original image comprises the following steps:
and sequentially carrying out graying processing, binarization processing, normalization processing and two-dimensional image mapping one-dimensional data processing on the mixed shot bill original image so as to obtain a mixed shot bill processing image corresponding to the mixed shot bill original image.
Preferably, the normalization processing and the two-dimensional image mapping one-dimensional data processing include:
wherein, Xi,jFor mixedly photographing original pixel value, mu, of original image of billiFor taking a mixed shot of mean values, sigma, of line pixel values in the original image of the documentiFor the purpose of the corresponding variance, the variance is,pixel values after row normalization; the normalized pixels are accumulated to obtain a mapped one-dimensional row vector, namely one-dimensional mapping data
In an embodiment of the present invention, as shown in fig. 2, the types of the tickets include value-added tax invoice, train ticket, quota ticket, business license, volume ticket, identification card, taxi ticket, motor vehicle sales and air itinerary. The method for acquiring the original image of the mixed shot bill comprises but is not limited to modes of shooting based on a mobile terminal and/or a PC terminal camera, high-definition camera shooting, photo album uploading, gallery uploading and the like. As shown in fig. 3, the original image of the acquired mixed shooting bill is a high-definition digital image. The color channel of the acquired image is a color or gray image, characters and table information in the invoice image are clearly visible, the invoice area occupies the main area of the acquired image, and the direction of the invoice area is in the positive direction.
The image preprocessing mainly comprises the following steps: the method comprises four steps of color image graying processing, grayscale image binaryzation processing, two-dimensional image normalization processing and two-dimensional image mapping one-dimensional data. Firstly, graying a color image, carrying out graying processing on a scanned image, reserving brightness information of the image, and restoring morphological characteristics to the maximum extent, wherein the processing result is shown in fig. 4; secondly, carrying out binarization processing on the gray level image, wherein the operation is used for carrying out binarization processing on the gray level image, highlighting the outline characteristics on the image, simplifying the image and facilitating the storage and transmission of a machine, and the processing result is shown in figure 5; on the basis, two-dimensional image normalization processing operation is used, the dimension of pixel data is unified, a subsequent deep learning model is convenient to perform feature learning and parameter training, and the image is normalized by using a formula 1; finally, the two-dimensional image is mapped into one-dimensional data through two-dimensional image mapping one-dimensional data processing, the data are further compressed while image information is kept, light weight storage and calculation of the data are achieved, and a mapping function is shown in a formula 2.
Wherein, Xi,jIs the original pixel value, mu, of a two-dimensional imageiBeing the mean, σ, of the pixel values of the lines in the two-dimensional imageiFor the purpose of the corresponding variance, the variance is,are pixel values normalized by row. The normalized pixels are accumulated to obtain a mapped one-dimensional row vector, namely one-dimensional mapping data
The invention can maximally keep the optical and shape characteristics of the bill and unify dimensions by carrying out graying, binarization and normalization processing on the original color bill image, thereby realizing effective data storage of the image. In the mapping operation of image preprocessing, the processed two-dimensional image data is mapped to a one-dimensional space, so that the two-dimensional image is further simplified, rapid and efficient data storage and compression are realized, and the model calculation speed is increased.
In step 102, a one-dimensional convolution bill classification OCTC model is constructed based on the deep neural network, and the OCTC model is trained by using each acquired mixed-shot bill processing image to determine an OCTC optimal model.
Preferably, the one-dimensional convolution note classification OCTC model includes: three convolution layers, three maximum pooling layers, two full-connection layers and one softmax layer; wherein, each layer uses five convolution kernels to respectively calculate input data so as to obtain convolution characteristic image data, and the number of the convolution kernels of each layer is [64,128 and 256 ]; the method comprises the steps of adopting a structure of two layers of fully-connected neural networks to realize identification and classification of the number of the mixed-shot bills, enabling the number of neurons of the two fully-connected layers to be [512,128], utilizing a first fully-connected layer to perform linear integration on feature image data obtained by calculation of a convolutional layer, utilizing a ReLU activation function to perform nonlinear conversion, and utilizing a second fully-connected layer to perform high-dimensional feature calculation; and finally, predicting the image category probability through a softmax layer activation function, and returning the bill quantity result in the current image.
In the bill quantity identification process, one-dimensional convolution is adopted to check one-dimensional data of bills for feature extraction based on an OCTC (open circular transform coefficient) model, feature screening is carried out through maximum pooling operation, and finally the identification of the bill quantity in an image is realized through a classification model, wherein the model structure is shown in FIG. 6. Wherein, the OCTC model consists of three convolutional layers (Conv1/2/3), three maximum pooling layers, two full-link layers and one softmax layer. The method adopts a one-dimensional convolution kernel to perform characteristic calculation on the mapped data, an OCTC model uses five convolution kernels to calculate the input data respectively, the number of the convolution kernels of each layer is [64,128 and 256], and a convolution calculation formula is shown as a formula 3; in addition, the ReLU activation function processing characteristic diagram of formula 5 is introduced, so that the model learning and calculation are facilitated. The invention designs the convolution layer from two dimensions of width and depth respectively so as to be convenient for better characteristic excavation: the width aspect refers to convolution kernels with different scales, five kinds of convolution kernels [2, 3, 4, 5 and 7] are respectively adopted to calculate bill image characteristics under different receptive fields, wherein two kinds of convolution kernels [5 and 7] are replaced by two small convolution kernels of 3- > 3 and 4- > 4, more image details are mined while the receptive fields are guaranteed to be unchanged, and the receptive fields are shown in FIG. 7; in the aspect of depth, three convolutional layers are designed, on one hand, the receptive field is enlarged, on the other hand, the characteristics can be further calculated in a high-dimensional mode, and the optimized extraction of the characteristics is achieved.
On the basis, the OCTC model screens the multi-dimensional feature map by adopting maximum pooling operation, so that the screening fusion of multi-channel features is realized, the calculated amount of the model is reduced, and the rapid calculation of the model is realized. The OCTC model adopts a structure of two layers of fully-connected neural networks to realize the identification and classification of the quantity of the mixed shot bills, the number of neurons in hidden layers of the two layers of networks is [512,128], the first fully-connected layer performs linear integration on the calculated features of the convolutional layers and performs nonlinear conversion by using a ReLU activation function, then the second fully-connected layer performs high-dimensional feature calculation, and finally the result of the quantity of the bills in the current picture is returned by predicting the image category probability by using the softmax activation function shown in formula 6.
hi=g(W·xi:i+n-1+ b) (formula 3)
pi=max{hi} (formula 4)
relu (x) max (0, x) (formula 5)
Wherein, W is a convolution kernel matrix, the size of the convolution kernel is n, linear transformation is firstly carried out on data in the convolution kernel, and then the data is converted by utilizing a nonlinear activation function g, so that a convolution characteristic diagram is obtained; and the maximum pooling operation is used for realizing feature screening by screening the maximum values in different channels at the same position as the current feature map to be output. Equation 5 is the ReLU activation function, x is the current hidden layer result, and the maximum value is taken as the final output result by comparing with 0. Equation 6 is the image class probability prediction, ziAnd (3) the output value of the ith node of the category output layer is represented by the number c of the model output categories, and softmax realizes category probability prediction by introducing an exponential function.
The types of the bill quantity output results of the model are divided into two types, a single type represents that only one bill exists in the mixed shot image, and a multiple type represents that multiple bills exist in the mixed shot image, as shown in fig. 8.
When the number of the bills is identified, the method performs convolution feature extraction, identifies structural features in the bill images, realizes multi-dimensional feature screening through maximum pooling, and can reduce the number of model parameters and realize the identification of the number of the bills while calculating the features of image data; the image features extracted from the convolutional layer are weighted and calculated through the type calculation of the full-connection network, and the number of bills in the image is predicted through the softmax activation function, so that the number identification of the bill images can be realized.
According to the method, the two-dimensional image data is subjected to gray level binarization processing, data cleaning is performed through normalization operation, and the data is finally mapped into the one-dimensional data, so that the data volume is greatly reduced while the image information is kept, the time waste caused by data transmission is reduced, and the user experience is improved. On the basis, an OCTC model is designed, one-dimensional convolution operation is introduced to perform feature extraction and image category calculation on image data, a plurality of small-size convolution kernels are used for increasing the model receptive field, the number of model parameters is reduced while the experimental effect is guaranteed, the effect of a lightweight model is achieved, and the storage and use of a user are facilitated.
In step 103, a hybrid bill original image to be detected is obtained, and the hybrid bill original image to be detected is processed to obtain a hybrid bill processed image to be detected.
In step 104, performing bill quantity recognition on the to-be-tested mixed shot bill processing image by using the OCTC optimal model to obtain a bill quantity recognition result.
In the embodiment of the invention, bill quantity recognition is carried out on the mixed shot bill processing image to be detected according to the determined OCTC optimal model so as to obtain a bill quantity recognition result.
Fig. 9 is a schematic diagram of a hardware structure of a deep neural network-based bill quantity recognition apparatus according to an embodiment of the present invention. As shown in fig. 9, the bill quantity recognition apparatus includes: one or more processors and memory, the apparatus may also include an input device and an output device. The processor, memory, input devices, and output devices may be connected by a bus or otherwise, as shown in fig. 9 to be connected by a bus.
Fig. 10 is a schematic structural diagram of a deep neural network-based bill quantity recognition system 1000 according to an embodiment of the present invention. As shown in fig. 10, the system 1000 for identifying the number of tickets based on the deep neural network according to the embodiment of the present invention includes: a first data processing unit 1001, an optimum model determination unit 1002, a second data processing unit 1003, and a ticket number identification unit 1004.
Preferably, the first data processing unit 1001 is configured to acquire a mixed-shot bill original image corresponding to each group of mixed bills in a plurality of groups of different mixed bills, and perform preprocessing on each acquired mixed-shot bill original image to acquire a mixed-shot bill processed image corresponding to each mixed-shot bill original image.
Preferably, the types of the hybrid ticket include: value-added tax invoices, train tickets, quota tickets, business licenses, roll tickets, identification cards, taxi tickets, motor vehicle sales and airline travel tickets.
Preferably, the optimal model determining unit 1002 is configured to construct a one-dimensional convolution note classification OCTC model based on a deep neural network, and train the OCTC model by using each acquired mixed shot note processing image to determine an optimal OCTC model.
Preferably, the one-dimensional convolution note classification OCTC model includes: three convolution layers, three maximum pooling layers, two full-connection layers and one softmax layer; wherein, each layer uses five convolution kernels to respectively calculate input data so as to obtain convolution characteristic image data, and the number of the convolution kernels of each layer is [64,128 and 256 ]; the method comprises the steps of adopting a structure of two layers of fully-connected neural networks to realize identification and classification of the number of the mixed-shot bills, enabling the number of neurons of the two fully-connected layers to be [512,128], utilizing a first fully-connected layer to perform linear integration on feature image data obtained by calculation of a convolutional layer, utilizing a ReLU activation function to perform nonlinear conversion, and utilizing a second fully-connected layer to perform high-dimensional feature calculation; and finally, predicting the image category probability through a softmax layer activation function, and returning the bill quantity result in the current image.
Preferably, the second data processing unit 1003 is configured to acquire an original image of a hybrid bill to be detected, and process the original image of the hybrid bill to be detected to acquire a processed image of the hybrid bill to be detected.
Preferably, the first data processing unit and the second data processing unit pre-process the mixed shot bill original image to obtain a mixed shot bill processed image corresponding to the mixed shot bill original image, including:
and sequentially carrying out graying processing, binarization processing, normalization processing and two-dimensional image mapping one-dimensional data processing on the mixed shot bill original image so as to obtain a mixed shot bill processing image corresponding to the mixed shot bill original image.
Preferably, the normalization processing and the two-dimensional image mapping one-dimensional data processing include:
wherein, Xi,jFor mixedly photographing original pixel value, mu, of original image of billiFor taking a mixed shot of mean values, sigma, of line pixel values in the original image of the documentiFor the purpose of the corresponding variance, the variance is,pixel values after row normalization; the normalized pixels are accumulated to obtain a mapped one-dimensional row vector, namely one-dimensional mapping data
Preferably, the bill quantity recognition unit 1004 is configured to perform bill quantity recognition on the mixed-shot bill processing image to be tested by using the optimal OCTC model to obtain a bill quantity recognition result.
The system 1000 for identifying the number of tickets based on the deep neural network according to the embodiment of the present invention corresponds to the method 100 for identifying the number of tickets based on the deep neural network according to another embodiment of the present invention, and details thereof are not repeated herein.
The invention has been described with reference to a few embodiments. However, other embodiments of the invention than the one disclosed above are equally possible within the scope of the invention, as would be apparent to a person skilled in the art from the appended patent claims.
Generally, all terms used in the claims are to be interpreted according to their ordinary meaning in the technical field, unless explicitly defined otherwise herein. All references to "a/an/the [ device, component, etc ]" are to be interpreted openly as referring to at least one instance of said device, component, etc., unless explicitly stated otherwise. The steps of any method disclosed herein do not have to be performed in the exact order disclosed, unless explicitly stated.
As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solutions of the present invention and not for limiting the same, and although the present invention is described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: modifications and equivalents may be made to the embodiments of the invention without departing from the spirit and scope of the invention, which is to be covered by the claims.
Claims (10)
1. A bill quantity identification method based on a deep neural network is characterized by comprising the following steps:
acquiring mixed shot bill original images corresponding to each group of mixed bills in a plurality of groups of different mixed bills, and preprocessing each acquired mixed shot bill original image to acquire a mixed shot bill processing image corresponding to each mixed shot bill original image;
constructing a one-dimensional convolution note classification OCTC model based on a deep neural network, and training the OCTC model by utilizing each acquired mixed-shot note processing image to determine an OCTC optimal model;
acquiring a mixed shot bill original image to be detected, and processing the mixed shot bill original image to be detected to acquire a mixed shot bill processing image to be detected;
and carrying out bill quantity recognition on the mixed shot bill processing image to be detected by utilizing the OCTC optimal model so as to obtain a bill quantity recognition result.
2. The method of claim 1, wherein the types of hybrid tickets comprise: value-added tax invoices, train tickets, quota tickets, business licenses, roll tickets, identification cards, taxi tickets, motor vehicle sales and airline travel tickets.
3. The method of claim 1, wherein the method preprocesses the mixed shot bill original image to obtain a mixed shot bill processed image corresponding to the mixed shot bill original image, and comprises:
and sequentially carrying out graying processing, binarization processing, normalization processing and two-dimensional image mapping one-dimensional data processing on the mixed shot bill original image so as to obtain a mixed shot bill processing image corresponding to the mixed shot bill original image.
4. The method of claim 3, wherein the normalization process and the two-dimensional image mapping one-dimensional data process comprise:
wherein, Xi,jFor mixedly photographing original pixel value, mu, of original image of billiFor taking a mixed shot of mean values, sigma, of line pixel values in the original image of the documentiFor the purpose of the corresponding variance, the variance is,pixel values after row normalization; normalizing the pixelAccumulating to obtain one-dimensional row vector after mapping, i.e. one-dimensional mapping data
5. The method of claim 1, wherein the one-dimensional convolutional bill classification OCTC model comprises: three convolution layers, three maximum pooling layers, two full-connection layers and one softmax layer; wherein, each layer uses five convolution kernels to respectively calculate input data so as to obtain convolution characteristic image data, and the number of the convolution kernels of each layer is [64,128 and 256 ]; the method comprises the steps of adopting a structure of two layers of fully-connected neural networks to realize identification and classification of the number of the mixed-shot bills, enabling the number of neurons of the two fully-connected layers to be [512,128], utilizing a first fully-connected layer to perform linear integration on feature image data obtained by calculation of a convolutional layer, utilizing a ReLU activation function to perform nonlinear conversion, and utilizing a second fully-connected layer to perform high-dimensional feature calculation; and finally, predicting the image category probability through a softmax layer activation function, and returning the bill quantity result in the current image.
6. A system for identifying the number of notes based on a deep neural network, the system comprising:
the first data processing unit is used for acquiring mixed shot bill original images corresponding to each group of mixed bills in a plurality of groups of different mixed bills, and preprocessing each acquired mixed shot bill original image to acquire a mixed shot bill processing image corresponding to each mixed shot bill original image;
the optimal model determining unit is used for constructing a one-dimensional convolution note classification OCTC model based on a deep neural network, and training the OCTC model by utilizing each acquired mixed-shooting note processing image to determine an OCTC optimal model;
the second data processing unit is used for acquiring a mixed shot bill original image to be detected and processing the mixed shot bill original image to be detected so as to acquire a mixed shot bill processing image to be detected;
and the bill quantity identification unit is used for identifying the bill quantity of the to-be-detected mixed shot bill processing image by using the OCTC optimal model so as to obtain a bill quantity identification result.
7. The system of claim 6, wherein the types of hybrid tickets comprise: value-added tax invoices, train tickets, quota tickets, business licenses, roll tickets, identification cards, taxi tickets, motor vehicle sales and airline travel tickets.
8. The system of claim 6, wherein the first and second data processing units preprocess the mixedly shot ticket raw image to obtain a mixedly shot ticket processed image corresponding to the mixedly shot ticket raw image, by:
and sequentially carrying out graying processing, binarization processing, normalization processing and two-dimensional image mapping one-dimensional data processing on the mixed shot bill original image so as to obtain a mixed shot bill processing image corresponding to the mixed shot bill original image.
9. The system of claim 8, wherein the normalization process and the two-dimensional image mapping one-dimensional data process comprise:
wherein, Xi,jFor mixedly photographing original pixel value, mu, of original image of billiFor taking a mixed shot of mean values, sigma, of line pixel values in the original image of the documentiFor the purpose of the corresponding variance, the variance is,pixel values after row normalization; the normalized pixels are accumulated to obtain a mapped one-dimensional row vector, namely one-dimensional mapping data
10. The system of claim 6, wherein the one-dimensional convolutional bill classification OCTC model comprises: three convolution layers, three maximum pooling layers, two full-connection layers and one softmax layer; wherein, each layer uses five convolution kernels to respectively calculate input data so as to obtain convolution characteristic image data, and the number of the convolution kernels of each layer is [64,128 and 256 ]; the method comprises the steps of adopting a structure of two layers of fully-connected neural networks to realize identification and classification of the number of the mixed-shot bills, enabling the number of neurons of the two fully-connected layers to be [512,128], utilizing a first fully-connected layer to perform linear integration on feature image data obtained by calculation of a convolutional layer, utilizing a ReLU activation function to perform nonlinear conversion, and utilizing a second fully-connected layer to perform high-dimensional feature calculation; and finally, predicting the image category probability through a softmax layer activation function, and returning the bill quantity result in the current image.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011456986.5A CN112749731A (en) | 2020-12-10 | 2020-12-10 | Bill quantity identification method and system based on deep neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011456986.5A CN112749731A (en) | 2020-12-10 | 2020-12-10 | Bill quantity identification method and system based on deep neural network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN112749731A true CN112749731A (en) | 2021-05-04 |
Family
ID=75648373
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202011456986.5A Pending CN112749731A (en) | 2020-12-10 | 2020-12-10 | Bill quantity identification method and system based on deep neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112749731A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114708608A (en) * | 2022-06-06 | 2022-07-05 | 浙商银行股份有限公司 | Full-automatic characteristic engineering method and device for bank bills |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109145961A (en) * | 2018-07-30 | 2019-01-04 | 上海交通大学 | A kind of mode identification method and system of unstructured Partial Discharge Data |
CN110428414A (en) * | 2019-08-02 | 2019-11-08 | 杭州睿琪软件有限公司 | The method and device of bill quantity in a kind of identification image |
CN111368828A (en) * | 2020-02-27 | 2020-07-03 | 大象慧云信息技术有限公司 | Multi-bill identification method and device |
-
2020
- 2020-12-10 CN CN202011456986.5A patent/CN112749731A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109145961A (en) * | 2018-07-30 | 2019-01-04 | 上海交通大学 | A kind of mode identification method and system of unstructured Partial Discharge Data |
CN110428414A (en) * | 2019-08-02 | 2019-11-08 | 杭州睿琪软件有限公司 | The method and device of bill quantity in a kind of identification image |
CN111368828A (en) * | 2020-02-27 | 2020-07-03 | 大象慧云信息技术有限公司 | Multi-bill identification method and device |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114708608A (en) * | 2022-06-06 | 2022-07-05 | 浙商银行股份有限公司 | Full-automatic characteristic engineering method and device for bank bills |
CN114708608B (en) * | 2022-06-06 | 2022-09-16 | 浙商银行股份有限公司 | Full-automatic characteristic engineering method and device for bank bills |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109584248B (en) | Infrared target instance segmentation method based on feature fusion and dense connection network | |
AU2018102037A4 (en) | A method of recognition of vehicle type based on deep learning | |
Jaiswal et al. | A technique for image splicing detection using hybrid feature set | |
WO2020164282A1 (en) | Yolo-based image target recognition method and apparatus, electronic device, and storage medium | |
CN109684922B (en) | Multi-model finished dish identification method based on convolutional neural network | |
CN108154102B (en) | Road traffic sign identification method | |
CN111160249A (en) | Multi-class target detection method of optical remote sensing image based on cross-scale feature fusion | |
CN112069868A (en) | Unmanned aerial vehicle real-time vehicle detection method based on convolutional neural network | |
US20200134382A1 (en) | Neural network training utilizing specialized loss functions | |
CN108830254B (en) | Fine-grained vehicle type detection and identification method based on data balance strategy and intensive attention network | |
CN111860496A (en) | License plate recognition method, device, equipment and computer readable storage medium | |
CN110222718A (en) | The method and device of image procossing | |
CN108323209A (en) | Information processing method, system, cloud processing device and computer program product | |
CN114913498A (en) | Parallel multi-scale feature aggregation lane line detection method based on key point estimation | |
CN112784754A (en) | Vehicle re-identification method, device, equipment and storage medium | |
CN113688821A (en) | OCR character recognition method based on deep learning | |
CN116630700A (en) | Remote sensing image classification method based on introduction channel-space attention mechanism | |
Thomas et al. | Smart car parking system using convolutional neural network | |
CN114842478A (en) | Text area identification method, device, equipment and storage medium | |
CN112749731A (en) | Bill quantity identification method and system based on deep neural network | |
CN113378609B (en) | Agent proxy signature identification method and device | |
CN111199228B (en) | License plate positioning method and device | |
CN114492634B (en) | Fine granularity equipment picture classification and identification method and system | |
US20220027662A1 (en) | Optical character recognition using specialized confidence functions | |
CN112634162A (en) | Artificial intelligent image processing method and system based on intelligent traffic |
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 |