WO2019127075A1

WO2019127075A1 - Method for identifying coin year, terminal device, and computer readable storage medium

Info

Publication number: WO2019127075A1
Application number: PCT/CN2017/118891
Authority: WO
Inventors: 陈红磊; 王磊
Original assignee: 中国科学院深圳先进技术研究院
Priority date: 2017-12-27
Filing date: 2017-12-27
Publication date: 2019-07-04

Abstract

The present application is applicable to the technical field of image recognition, and provides a method for identifying a coin year, a terminal device, and a computer readable storage medium. The method comprises: obtaining a target coin image to be identified; inputting the target coin image into a trained coin year recognition model; and determining the year of a coin in the target coin image according to year number information outputted by the coin year recognition model. Because the year of a coin to be identified can be identified by using a pre-trained coin year recognition model, manual classification of coins based on the years of the coins is avoided, so that the coin recovery efficiency is improved and the manual costs are saved.

Description

\¥0 2019/127075 卩(:17 \2017/118891

Method for identifying coin year, terminal device and computer readable storage medium

Technical field

[0001] The present invention belongs to the field of image recognition technologies, and in particular, to a method for identifying a coin year, a terminal device, and a computer readable storage medium.

Background technique

[0002] With the development of the national economy and the improvement of the printing process, the country will issue a new set of coins at regular intervals, and some coins with earlier release years need to be recycled. The traditional coin recycling method is to classify the coins according to the release year of the front printing of the coin, and then select the coins with the earlier release date. This method of manually recycling coins is labor intensive, inefficient, and labor intensive. Summary of invention

technical problem

[0003] The present invention provides a method for identifying a coin year, a terminal device, and a computer readable storage medium, to solve the existing method for manually recycling coins, which has a large workload, low efficiency, and high labor cost. problem.

Problem solution

Technical solution

A first aspect of the invention provides a method of identifying a year of a coin, comprising:

[0005] acquiring an image of a target coin to be identified;

[0006] inputting the target coin image into the trained coin year recognition model;

[0007] determining a year of the coin in the target coin image according to the year number information output by the coin year identification model; wherein the year number information is used to represent the year of the coin.

A second aspect of the invention provides a terminal device comprising means for performing the method as described in the first aspect above.

A third aspect of the present invention provides a terminal device including a memory, a processor, and the \¥0 2019/127075 卩 (: 17 \2017/118891 a computer program in a memory and operable on the processor, wherein the processor executes the computer program to implement the method as described in the first aspect above A step of.

[0010] A fourth aspect of the invention provides a computer readable storage medium storing a computer program, wherein the computer program is executed by a processor to implement the first aspect as described above The steps of the method.

Advantageous effects of the invention

Beneficial effect

[0011] The present invention obtains a target coin image to be identified; inputs the target coin image to a trained coin year recognition model; and determines, in the target coin image, based on year digital information output by the coin year recognition model The year of the coin. Since the year of the coin to be recognized is recognized by the pre-trained coin year recognition model, it is not necessary to manually classify the coin according to the year of the coin, which not only improves the efficiency of coin recovery but also saves labor costs.

Brief description of the drawing

DRAWINGS

1 is a flowchart of an implementation of a method for recognizing a coin year according to an embodiment of the present invention;

[0013] FIG. is a schematic diagram of a coin image provided by an embodiment of the present invention;

[0014] FIG. 215 is a schematic diagram of a lower half of a circular display area in which a coin image is located in a coin image according to an embodiment of the present invention;

[0015] FIG. 2 is a schematic diagram showing the upper half of the circular display area where the coin is located in the coin image provided by the embodiment of the present invention;

2 is a schematic diagram of a coin image in an embodiment of the present invention in which the year number is located in the right half of the circular display area where the coin is located;

[0017] FIG. is a schematic diagram of a coin image in which the year number is located in the left half of the circular display area where the coin is located;

3 is a flowchart of an implementation of a method for recognizing a coin year according to another embodiment of the present invention;

4 is a flowchart of an implementation of 301 in a method for recognizing a coin year according to an embodiment of the present invention;

[0020]

Schematic diagram of the structure of the RCNN model;

[0021] FIG. 6 is a flowchart of an implementation of a method for identifying a coin year according to an embodiment of the present invention; \¥0 2019/127075 卩(:17 \2017/118891

[0022] FIG. 7 is a schematic structural diagram of a terminal device according to an embodiment of the present invention;

8 is a schematic structural diagram of a terminal device according to another embodiment of the present invention;

FIG. 9 is a schematic structural diagram of a terminal device according to still another embodiment of the present invention.

Invention embodiment

Embodiments of the invention

[0025] The technical solutions in the embodiments of the present invention will be clearly and completely described in the following with reference to the accompanying drawings in the embodiments of the present invention. It is obvious that the described embodiments are a part of the embodiments of the present invention, but not all embodiments. . All other embodiments obtained by a person of ordinary skill in the art based on the embodiments of the present invention without departing from the inventive work are all within the scope of the present invention.

The terms "comprising" and "comprising", when used in the claims and the claims Or the presence or addition of a plurality of other features, integers, steps, operations, elements, components, and/or collections thereof.

The terminology used herein is for the purpose of describing particular embodiments and is not intended to limit the invention. The singular forms "a," ",",,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,

[0028] It should also be further understood that the term "and/or" used in the description of the invention and the appended claims means any combination and all possible combinations of one or more of the associated listed items, and These combinations are included.

Referring to FIG. 1, FIG. 1 is a flowchart of an implementation of a method for recognizing a coin year according to an embodiment of the present invention. The execution subject of the method for recognizing the coin year in this embodiment is a terminal device. The terminal device may be a mobile terminal device such as a smart phone or a tablet computer, or may be another terminal device, and is not limited herein. The method of identifying the year of the coin as shown in FIG. 1 may include the following steps:

[0030] 811: Acquire an image of the target coin to be identified.

[0031] In this embodiment, when it is necessary to identify the year of release of the coin, the image of the coin to be recognized by the camera device may be image-collected to obtain a coin image, and the collected coin image is input to the terminal device. The terminal device receives the coin image input by the user if an instruction for identifying the year of the coin is detected, and determines the coin image input by the user as the target coin image to be recognized. \¥0 2019/127075 卩(:17 \2017/118891

[0032] The target coin image may be one sheet or at least two, and is not limited herein.

[0033] The coin to be identified may be a renminbi coin, a euro coin, or other types of hard coins, and is not limited herein.

[0034] 812: Input the target coin image into the trained coin year recognition model.

[0035] After the terminal device acquires the target coin image to be identified, the pre-trained coin year model is invoked to process the target coin image. It should be noted that, in the present embodiment, the input amount of the coin year identification unit is a coin image, and the output amount is the year digital information of the coin in the coin image.

[0036] The year number information includes the name and coordinates of the four-digit year number used to form the coin year on the coin. For example, if the release date of the coin is 1987, the names of the four digits on the coin are "1", "9", "8", "7". In this embodiment, the coordinates of the four-digit year number can be represented by the coordinates of a rectangular area occupied by the four-digit year number in the coin image. The coordinates of the rectangular area can be represented by the coordinates of the two end points of the preset diagonal of the rectangular area.

For example, please refer to FIG. 2 & FIG. 2 for a schematic diagram of a coin image provided by an embodiment of the present invention. As shown in the figure, the rectangular area occupied by the four digits "1", "9", "8", and "7" on the coin in the coin image is

Can pass through the rectangular area

Indicates the coordinates of the year numbers "1", "9", "8", and "7". In an embodiment of the invention, the coordinates of each rectangular region may be represented by the coordinates of the end points of the lower left corner of each rectangular region and the coordinates of the endpoints of the upper right corner. For example, if the coordinates of the rectangular area _{& the} lower left end point are (X)

), the coordinates of the rectangular area & the upper right corner are (X , ₂ ), the shell wrist area &

4 ^)], then the year number "

[0038] In this embodiment, the terminal device invoking the pre-trained coin year model to process the target coin image may be: the terminal device inputs the target coin image to be identified into the coin year recognition model, and acquires the coin year recognition model. The year digital information of the coin in the output target coin image.

[0039] 813: Determine a year of the coin in the target coin image according to the year digital information output by the coin year identification model.

[0040] After acquiring the year number information of the coin in the target coin image output by the coin year identification model, the terminal device determines the year of the coin in the target coin image according to the year number information of the coin in the target coin image. \¥0 2019/127075 卩(:17 \2017/118891

[0041] In a specific embodiment of the present invention, 313 may include the following steps:

And [0042] sorting the four-digit year numbers according to coordinates of four-digit year numbers on coins in the target coin image and a preset sorting strategy to obtain a year of the coins.

[0043] In this embodiment, the terminal device may sort the four-digit year number according to the coordinates of the four-digit year number on the coin in the target coin image and the preset sorting strategy, thereby obtaining the year of the coin.

[0044] In this embodiment, when the arrangement of the four digits on the coin is different, the corresponding preset sorting strategy is different, and the setting is specifically according to actual requirements, and no limitation is made here.

[0045] As shown in the figure, in practical applications, since the four digits on the coin are usually arranged in an arc shape along the edge of the coin, the four-digit year number on the coin is presented along the edge of the coin. The arc shape is sequentially arranged as an example, and the specific process of sorting the four-digit year numbers according to the preset sorting strategy by the terminal device is described in detail. Specifically, in this embodiment, the four-digit year number is sorted according to the coordinates of the four-digit year number on the coin in the target coin image and the preset sorting strategy, to obtain the year of the coin, specifically Includes the following steps:

[0046] processing the target coin image by using a background difference method to obtain a circular display area where the coin in the target coin image is located, and determining coordinates of a center point of the circular display area;

[0047] determining a position of the four-digit year number in the circular display area according to coordinates of a center point of the circular display area and coordinates of four-digit year numbers on the coin;

[0048] if the four digits are located in the lower half of the circular display area, the four digits are sorted according to a first preset order to obtain a year of the coin;

[0049] if the four digits are located in the upper half of the circular display area, the four digits are sorted according to a second preset order to obtain the year of the coin;

[0050] if the four digits are located in the right half of the circular display area, the four digits are sorted according to a third preset order to obtain the year of the coin;

[0051] If the four digits are located in the left half of the circular display area, the four digits are sorted according to a fourth preset order to obtain the year of the coin.

[0052] In this embodiment, after obtaining the coordinates of the four-digit year number on the coin in the target coin image, the terminal device processes the target coin image by the background difference method to extract the coin from the target coin image. Round display area. The terminal device determines based on the original coordinate system corresponding to the target coin image \¥0 2019/127075 卩(:17 \2017/118891 The coordinates of the center point of the circular display area. In practical applications, the terminal device can call

The ^ _§ 1〇1 ]'(3⁄4)8 function in the application is used to find the coordinates of the center point of the circular display area.

[0053] Since the coordinates of the four-digit year number output by the coin year recognition model are represented by the coordinates of the two end points of the preset diagonal of the rectangular area occupied by the four-digit year number, the terminal device identifies the coin year The coordinates of the four-digit year number output by the model are converted, and the coordinates of the center point of the rectangular area occupied by the four-digit year number are obtained, and the coordinates of the center point of the rectangular area occupied by the four-digit year number and the circular display area are The center point coordinates are compared to determine the position of the four-digit year number in the circular display area.

[0054] In an actual application, the first preset order, the second preset order, the third preset order, and the fourth preset order need to be referred to according to the coin year identification model when determining the coordinates of the digital year on the coin. The coordinate system is determined. Assume that the coordinates of the year number output by the coin year recognition model are based on the origin of the upper left corner of the coin image, the vertical side where the coordinate origin is located as the positive direction of the X axis, and the horizontal side where the coordinate origin is located as the positive direction of the X axis. Coordinate system, Bay 1

[0055] As shown in FIG. 215, if the terminal device detects that the X-axis coordinates corresponding to the center points of the four rectangular regions occupied by the four-digit year numbers are greater than the \-axis coordinates corresponding to the center point of the circular display region, the terminal device recognizes The four-digit year numbers are located in the lower half of the circular display area. At this time, the terminal device sorts the four-digit year numbers in the order of the \-axis coordinates corresponding to the center point of the rectangular area from the smallest to the largest, and obtains the year of the coin. In Fig. 215, the year of the coin obtained after sorting is "1987".

[0056] As shown in the figure, if the terminal device detects that the X-axis coordinates corresponding to the center points of the four rectangular regions occupied by the four-digit year number are smaller than the \-axis coordinates corresponding to the center point of the circular display region, the terminal device recognizes that The four-digit year numbers are located in the upper half of the circular display area. At this time, the terminal device sorts the four-digit year numbers in descending order of the \-axis coordinates corresponding to the center point of the rectangular area to obtain the year of the coin. The year of the coin obtained after sorting the figure is "1987".

[0057] As shown in FIG. 2 (1), if the terminal device detects that the X-axis coordinate corresponding to the center point of the four rectangular regions occupied by the four-digit year number is greater than the \-axis coordinate corresponding to the center point of the circular display region, Then, the four-digit year numbers are located in the right half of the circular display area, and the terminal device sorts the four-digit year numbers in descending order of the \-axis coordinates corresponding to the center point of the rectangular area to obtain the coin. Year, in Figure 2 (1, the year of the coin obtained after sorting is "1987". \¥0 2019/127075 卩(:17 \2017/118891

[0058] As shown in the figure, if the terminal device detects that the X-axis coordinate corresponding to the center point of the four rectangular regions occupied by the four-digit year number is smaller than the \-axis coordinate corresponding to the center point of the circular display region, the terminal device recognizes that The four-digit year numbers are located in the left half of the circular display area. At this time, the terminal device sorts the four-digit year numbers according to the \axis coordinate corresponding to the center point of the rectangular area from small to large, and obtains the year of the coin. The year of the coin obtained after sorting is "1987".

[0059] As can be seen from the above, the method for identifying the coin year provided by the embodiment obtains the target coin image to be identified; the target coin image is input to the trained coin year recognition model; The year number information of the recognition model output determines the year of the coin in the target coin image. Since the year of the coin to be recognized is recognized by the pre-trained coin year recognition model, it is not necessary to manually classify the coin according to the year of the coin, which not only improves the efficiency of coin recovery but also saves labor costs.

Referring to FIG. 3, FIG. 3 is a flow chart of an implementation of a method for recognizing a coin year according to another embodiment of the present invention. The execution subject of the method for recognizing the coin year in this embodiment is a terminal device. The terminal device may be a mobile terminal device such as a mobile phone or a tablet computer, or may be another terminal device, and is not limited herein. As shown in FIG. 3, on the basis of the embodiment corresponding to FIG. 1, the embodiment may further include 0 before 311.

1-802, as follows:

[0061] 801: Acquire a training sample set.

[0062] In the present embodiment, the coin year recognition model needs to be trained before the pre-trained coin year recognition model is called to identify the year of the coin.

[0063] When training the coin year recognition model, it is first necessary to acquire a training sample set for training the model. The training sample set includes a plurality of sets of sample data, and each set of sample data is composed of a coin image and year digital information of coins in the coin image. The year number information for the coin includes the name and seat of the four digits on the coin.

[0064] In an embodiment of the present invention, 301 can be implemented by using 3011~3015 as shown in FIG. 4, as follows:

[0065] 8011: Acquire a coin image for training the model.

[0066] In this embodiment, before training the model, a large number of coin images for training the model may be collected by the camera device, and the collected coin images for training the model are uniformly stored in the terminal device. \¥0 2019/127075 卩(:17 \2017/118891 First image folder. When the terminal device receives the model training instruction, it acquires the coin image for training the model from the first image folder.

[0067] It should be noted that all the coin images collected by the imaging device are the same size. For example, all coin images acquired by the camera are of MxN pixels. Wherein, M and N respectively represent the number of pixels included in each row and column of the image, and M and N are positive integers, and M and N can be set according to actual needs, and are not limited herein.

[0068] In this embodiment, after the terminal device acquires all the coin images in the first image folder, all the coin images in the first image folder may be renamed. Specifically, the terminal device can be preset

The algorithm renames all the coin images in the first image folder, and sets the name of the coin image uniformly.

Where X is any number from 0 to 9, for example, the name of the renamed coin image may be 000001.jpg, 000002.jpg, 000003.jpg, and the like. The terminal device can store all the coin images after the rename in the second image folder.

[0069] 8012: determining coordinates of four rectangular regions selected by the user on the coins in each of the coin images

Wherein the rectangular area is used to represent a display area occupied by a year number on the coin.

[0070] In the present embodiment, after the terminal device renames all the coin images for training the model, the name and coordinates of the four-digit year number on the coin in each coin image are determined.

[0071] In practical applications, in determining the coordinates of the four digits of the number on the coin in each coin image, the user is required to assist in marking the four digits of the number on the coin in each coin image. Specifically, the user can frame the four-digit year number on the coin in each coin image through a rectangular frame, and then select four rectangular regions. The four rectangular areas selected by the user in the coin image are used to respectively represent the display area occupied by the four-digit year number on the coin in the coin image.

[0072] It should be noted that since the size of the four digits on the coin is substantially the same, the area of the display area occupied by the four digits is substantially the same, and the four selected by the user in the coin image. The area of the rectangular area is approximately the same.

[0073] In a specific embodiment of the present invention, 3012 may specifically include the following steps:

[0074] acquiring four rectangular regions selected by the user on the coins in each of the coin images;

[0075] determining, according to a preset coordinate calibration strategy, coordinates of two end points of a preset diagonal of each of the rectangular regions, and coordinates coordinates of two end points of a preset diagonal of each of the rectangular regions Determined to be the rectangular area \¥0 2019/127075 卩(:17 \2017/118891 The coordinates of the field.

[0076] In this embodiment, after the user selects four rectangular areas on the coins in the coin image, the terminal device acquires four rectangular areas selected by the user on the coins in each coin image, and Determining the coordinates of the two end points of the preset diagonal of each rectangular area according to the preset coordinate calibration strategy, and determining the coordinates of the two end points of the preset diagonal of each rectangular area as the coordinates of the rectangular area .

[0077] wherein the preset coordinate calibration strategy is used to represent coordinate calibration based on a preset coordinate system. The preset coordinate system can be determined according to actual needs, and there is no restriction here. For example, the preset coordinate system may be the coordinate origin of the upper left corner of the coin image, the vertical direction where the coordinate origin is located as the positive direction of the X axis, and the horizontal edge where the coordinate origin is located as the coordinate system established by the positive direction of the axis.

[0078] The terminal device determines coordinates of two end points of a preset diagonal of each rectangular area in each coin image based on a preset coordinate system. The preset diagonal line can be set according to actual requirements, and is not limited herein. For example, the preset diagonal line may be a diagonal line connecting the lower left corner end point and the upper right corner end point of the rectangular area, and the preset is The two endpoints of the diagonal are the lower left endpoint and the upper right endpoint, respectively.

[0079] after determining the coordinates of the consecutive endpoints of the preset diagonal of each rectangular area in each coin image, the terminal device recognizes the coordinates of the consecutive endpoints of each rectangular area as The coordinates of the corresponding rectangular area.

[0080] For example, as shown in the figure, if a rectangular area &

The coordinates of the rectangle area & the upper right corner are (X 4 ^), and the coordinates of the shell wrist area & is [(X 4 ^) ”.

[0081] 8013: Determine coordinates of four rectangular regions in the coin image as coordinates of four-digit year numbers on coins in the coin image, respectively.

[0082] In the embodiment, after the terminal device determines the coordinates of the four rectangular regions in each coin image, the coordinates of the four rectangular regions are respectively determined as the four-digit year numbers on the coins in the coin image. coordinate of. That is, the coordinates of the year number in the rectangular area are expressed by the coordinates of the rectangular area.

[0083] For example, in conjunction with the figure and step 3012, if a rectangular area &

Then the year number"

[0084] 8014: Determine a name of a four-digit year number on a coin in each of the coin images.

[0085] In the present embodiment, while determining the coordinates of the four-digit year number on the coin in each coin image, the terminal device also determines the name of the four-digit year number on the coin in each coin image. [0086] In practical applications, the name of the four-digit year number on the coin in each coin image may be input by the user. For example, the user inputs the name of the number while selecting each year number through the rectangular frame. The terminal device obtains the name of each year number entered by the user.

[0087] 8015: Associate each of the coin image, the name of the year number on the coin in each of the coin images, and coordinates to obtain a training sample set.

[0088] In this embodiment, after the terminal device acquires the name and coordinates of the four-digit year number on the coin in each coin image, each coin image and the four-digit year on the coin in each coin image The names and coordinates of the numbers are stored in association.

[0089] Specifically, after acquiring the name and coordinates of the four-digit year number on the coin in each coin image, the terminal device may first name the name of each coin image and four digits on the coin in each coin image. The name and coordinates of the year number are stored in a table form in the first text file.

[0090] Please refer to Table 1, Table 1 shows part of the first text file, the part is named 001352.

The name and coordinates of the year number on the coin in the coin image.

Table 1

[] [Table 1]

[0092] In this embodiment, after the terminal device associates the name of each coin image, the name and coordinates of the four digits on the coin in each coin image in a table form in the first text file, According to the name of the coin image, the text information with the same name of the coin image in the first text file is further stored in the same Extensible Markup Language (XML) file, and the names of the coin images in the first text file are different. Text information is stored in different XML files. in this way , you can get multiple XML files. Each XML file stores the name and coordinates of a four-digit year number on a coin in a coin image. For example, the resulting multiple XML files can be: 000001. xml 000002.xml _s 000003 Pass.

[0093] After obtaining the XML file corresponding to each coin image, the terminal device associates the coin image with its corresponding XM L file to obtain a plurality of sets of sample data, and the plurality of sets of sample data constitute a sample set for training the model. .

[0094] After obtaining the sample set, the terminal device can randomly extract 50% of the sample data from the sample set as the training sample data, and the training sample data constitutes the training sample set.

[0095] S02: using the training sample set to pre-build a region-based fast convolutional neural network Faster

The RCNN model is trained to determine the trained Faster RCNN model as a coin year identification model.

[0096] In this embodiment, after acquiring the training sample set, the terminal device uses the training sample set to train the pre-built region-based convolutional neural network (Faster RCNN) model.

[0097] Specifically, the terminal device uses each coin image as an input of a Faster RCNN model, and the name and coordinates of the four digits on the coin in each coin image are output of the Faster RCNN model to the pre-built Faster RCNN model. Train. After training the Faster RCNN model, the terminal device determines the trained Faster RCNN model as the coin year recognition model.

[0098] wherein the coin year identification model is used to identify the year of the coin.

[0099] Please refer to FIG. 5 together. FIG. 5 is a schematic structural diagram of a Faster RCNN model according to an embodiment of the present invention. As shown in FIG. 5, in a specific embodiment of the present invention, the Faster RCNN model specifically includes a feature extraction network and a region extraction network (Region proposal).

Network, RPN) and target identification network. The input end of the RPN and the input end of the target identification network are both connected to the output of the feature extraction network, and the input end of the target identification network is also connected to the output of the PRN.

[0100] The feature extraction network is configured to perform a convolution operation on the pixel values of all the pixel points in the input coin image by the convolution kernel to obtain a feature map, and output the feature map to the RPN and the target recognition network. The dimensions of the feature map are much smaller than the dimensions of the coin image.

[0101] The RPN is configured to determine, according to the feature map, a display area in the coin image that may be a year number, and The display area where the year number is located is output as a candidate area to the target recognition network.

[0102] The target recognition network is configured to select a target region corresponding to the four-digit year number on the coin in the coin image from the candidate region according to the feature map output by the feature extraction network, and identify the year number in the region where the year number is located. And output the name of the four-digit year number and the coordinates of the corresponding target area.

[0103] In practical applications, the feature extraction network may adopt a VGG16 network. The target recognition network may specifically be a Fast Region-based Convolutional Neural Network (Fast RCNN).

[0104] In this embodiment, the target area is a rectangular area, and the coordinates of the target area may be represented by the coordinates of the two end points of the preset diagonal of the target area, for example, the coordinates of the end point of the lower left corner of the target area may be The coordinate representation of the endpoint in the upper right corner.

[0105] In this embodiment, specifically, the RPN is used to gradually slide on the feature map by using a preset sliding window. The sliding step size can be set according to actual needs, and there is no limit here. Each position where the preset sliding window slides on the feature map maps k original image areas of different sizes or areas. For example, assuming that the preset sliding window corresponds to 300 sliding positions on the feature map, the entire feature map corresponds to 300k original image regions. In the present embodiment, the coordinates corresponding to each of the original image regions are known. Since the original image region is a rectangular region, the coordinates of the original image region can be represented by the coordinates of the endpoint positions at which the four corners are located. For each original image area, the RPN calculates the probability value that the original image area is an unrelated area or the area where the year number is located, that is, each original image area corresponds to two probability values. Based on the two probability values corresponding to each original image region, the RPN selects n original image regions with a high probability of the region where the year number is located as the candidate region from all the original image regions, and outputs the coordinates of the candidate region to the target recognition. The internet.

[0106] The target recognition network is configured to calculate a probability value of each of the numbers 0 to 9 for each candidate region, that is, each candidate region corresponds to 10 probability values, which are any numbers of numbers 0-9. Probability value. That is to say, if the number of candidate regions is n, then 10n probability values are finally obtained, and each of the numbers 0-9 corresponds to _n probability values. Considering that the year number on the coin is four digits, the target recognition network selects four probability values with higher probability values from the _n probability values corresponding to each number, that is, each number in 0~9 is separately screened out. Four candidate regions, and finally 40 candidate regions were selected. The target recognition network further determines four candidate regions with the highest probability value from the selected 40 candidate regions, and identifies the four candidate regions with the highest probability value as The four-digit year number on the coin corresponds to the target area. Finally, the target recognition network outputs the coordinates of the four target areas and the names of the corresponding year numbers.

[0107] In a specific embodiment of the present invention, based on the specific structure of the Faster RCNN model, S02 can be implemented by S021~S025 as shown in FIG. 6, which is as follows:

[0108] S021: Initialize parameter values of the feature extraction network according to parameter values of the feature extraction model obtained by pre-training, randomly initialize parameter values of the region extraction network, and adopt the feature extraction after initialization a network and the region extraction network extract candidate regions from each of the coin images

[0109] In this embodiment, the terminal acquires the feature extraction model obtained by the pre-training, and initializes the parameters of the feature extraction network according to the parameter values of the feature extraction model obtained by the pre-training.

[0110] wherein the feature extraction model is a feature extraction model based on ImageNet training. ImageNet is the largest image database used for image recognition in the world.

[0111] In practical applications, the feature extraction network may adopt a VGG16 network. Please refer to Table 2 together. Table 2 shows the parameters of a VGG16 network provided by this embodiment.

[0112] The VGG16 network includes 13 convolution layers and 4 pooling layers. The size and number of convolution kernels in each layer are shown in Table 2.

Table 2

[]

\¥0 2019/127075 卩(:17 \2017/118891

[Table 2]

[0114] In this embodiment, the terminal device may obtain the parameter value of the pre-trained 0016 model from the network, and initialize the parameter value of 0016 by using the parameter value of the pre-trained 0016 model. Among them, the parameter values of 0016 include, but are not limited to, the number of convolution kernels, the size of the convolution kernel, and the values corresponding to the respective elements in the convolution kernel.

[0115] Meanwhile, the terminal device randomly initializes the parameter values of the RPN. Initializing the parameter value of the RPN means initializing the value of each element in each convolution kernel included in the RPN. After initializing the parameter value of the feature extraction network and the parameter value of the RPN in the area extraction network, the terminal device may extract the candidate area from each coin image by using the initialized feature extraction network and the RPN, and The coordinates of the extracted candidate regions are output to the target recognition network.

[0117] S022: Perform the first training on the target recognition network according to the candidate region extracted from each of the coin images and the year digital information of the coins in each of the coin images.

[0118] The terminal device performs the first training on the target recognition network based on the candidate region extracted from each coin image and the year digital information on the coin in each coin image.

[0119] Specifically, the terminal device uses the coordinates of the candidate region extracted from each coin image as the input of the target recognition network, and uses the year digital information on the coin in each coin image as the output of the target recognition network, The first training is performed on the target recognition network.

[0120] After the first training of the target recognition network, an initial parameter value of the target recognition network is obtained. The initial parameter value of the target recognition network refers to the initial parameter value of each element in the convolution kernel in the target recognition network.

[0121] S023: Update parameter values of the area extraction network according to initial parameter values of the target identification network after the first training, and adopt the initialized feature extraction network and the updated area. The extraction network again extracts candidate regions from each of the coin images.

[0122] The terminal device updates the parameter value of the RPN according to the initial parameter value of the target network after the first training, and extracts the candidate region from each coin image again by using the updated region extraction network, and extracts the candidate region again. The coordinates of the candidate regions are output to the target recognition network.

[0123] S024: Perform a second training on the target recognition network according to the candidate region extracted from each of the coin images and the year digital information of the coins in each of the coin images.

[0124] The terminal device performs the second training on the target recognition network after the first training based on the candidate region extracted from each coin image and the year digital information on the coin in each coin image. Specifically, the terminal device will again use the coordinates of the candidate region extracted from each coin image as the input of the target recognition network, and use the year digital information on the coin in each coin image as the output of the target recognition network to target the target. Identify the network for a second training session.

[0125] After the second training of the target recognition network, the final parameter value of the target recognition network is obtained. Wherein, the final parameter value of the target recognition network refers to the final parameter of each element in the convolution kernel in the target recognition network. \¥0 2019/127075 卩(:17 \2017/118891 value.

[0126] 8025: Identify, by the initialized feature extraction network, the updated region extraction network, and the target training network after the second training, the Fa _S t _er RCNN model as a coin year identification. model.

[0127] The 8^3· RCNN model composed of the updated area extraction network of the terminal device and the target recognition network after the second training is determined as a coin year recognition model, and the coin year identification model is stored. When the year of the coin needs to be identified, the terminal calls the coin year identification model.

[0128] It should be noted that, when the terminal receives the coin year identification instruction for the first time, the steps of training the coin year model, that is, 301 to 302, are executed. When the terminal does not receive the coin year identification command for the first time, the steps of training the coin year model are not executed, and 311~313 are directly executed.

[0129] As can be seen from the above, the method for identifying the year of the coin provided by the embodiment is pre-built.

RCNN model, and using training set of images by groups of coins and coin Year of digital information constituted? _&

The type is trained to obtain a coin year recognition model. Because the Fa _S t _er RCNN model detects and locates the target faster and has higher accuracy, the coin year model obtained by training the Fa _S _er RCNN model is used to identify the year of the coin, which not only improves the coin. The efficiency of the year identification, and the accuracy of the coin year identification. And since the positions and postures of the coins in the plurality of coin images included in the training sample set are random, the coin year recognition model trained based on the training sample set can be effective for the years of the coins in the coin images of different specifications. Identification.

[0130] Please refer to FIG. 7. FIG. 7 is a schematic structural diagram of a terminal device according to an embodiment of the present invention. The terminal device 700 can be a mobile terminal device such as a smart phone or a tablet computer, and can also be other terminal devices, and is not limited herein. Each unit included in the terminal device 700 of this embodiment is used to perform the steps in the embodiment corresponding to FIG. 1. For details, refer to the related description in the embodiment corresponding to FIG. 1 and FIG. 1 , and details are not described herein. As shown in FIG. 7, the terminal device 700 of this embodiment includes a first obtaining unit 71, a model calling unit 72, and a year determining unit 73.

[0131] The first acquisition unit 71 is configured to acquire a target coin image to be identified.

[0132] The model invoking unit 72 is configured to input the target coin image to the trained coin year recognition model.

[0133] The year determining unit 73 is configured to determine a year of the coin in the target coin image according to the year number information output by the coin year identification model; wherein the year number information is used to represent the year of the coin \¥0 2019/127075 卩(:17 \2017/118891

[0134] Optionally, in a specific embodiment of the present invention, the year digital information includes a name and a coordinate of a four-digit year number on the coin in the target coin image; the year determining unit 73 is specifically configured to: The coordinates of the four-digit year number on the coin in the target coin image and the preset sorting strategy sort the four-digit year number to obtain the year of the coin.

[0135] Optionally, in a specific embodiment of the present invention, the year determining unit 73 is specifically configured to:

[0136] processing the target coin image by using a background difference method to obtain a circular display area where the coin in the target coin image is located, and determining coordinates of a center point of the circular display area;

[0137] determining a position of the four-digit year number in the circular display area according to coordinates of a center point of the circular display area and coordinates of four-digit year numbers on the coin;

[0138] if the four digits are located in the lower half of the circular display area, the four digits are sorted according to the first preset order to obtain the year of the coin;

[0139] if the four digits are located in the upper half of the circular display area, the four digits are sorted according to a second preset order to obtain a year of the coin;

[0140] if the four digits are located in the right half of the circular display area, the four digits are sorted according to a third preset order to obtain the year of the coin;

[0141] If the four digits are located in the left half of the circular display area, the four digits are sorted according to a fourth preset order to obtain the year of the coin.

[0142] It can be seen that the terminal device provided by the embodiment obtains the target coin image to be identified; the target coin image is input to the trained coin year recognition model; and the model is output according to the coin year identification model. The year number information determines the year of the coin in the target coin image. Since the year of the coin to be identified is recognized by the pre-trained coin year recognition model, it is not necessary to manually classify the coin according to the year of the coin, which not only improves the efficiency of coin recovery but also saves labor costs.

Referring to FIG. 8, FIG. 8 is a schematic structural diagram of a terminal device according to another embodiment of the present invention. Each unit included in the terminal device 700 of this embodiment is used to perform the steps in the embodiment corresponding to FIG. 3. For details, refer to the related description in the embodiment corresponding to FIG. 3 and FIG. 3, and details are not described herein. As shown in FIG. 8, the terminal device 700 in this embodiment further includes a second acquiring unit 74 and a model, with respect to the embodiment corresponding to FIG. \¥0 2019/127075 卩(:17 \2017/118891 Training unit 75.

[0144] The second obtaining unit 74 is configured to acquire a training sample set; wherein each set of sample data in the training sample set is composed of a coin image and year digital information of coins in the coin image, the year of the coin The digital information includes the name and coordinates of the four digits of the year on the coin.

[0145] The model training unit 75 is configured to adopt the training sample set pair to construct a pre-constructed region-based fast convolutional neural network

RCNN model is trained, will be trained

The RCNN model is determined to be a coin year identification model; wherein the coin year identification model is used to identify the year of the coin.

[0146] Optionally, in a specific embodiment of the present invention, the second obtaining unit 74 specifically includes an image obtaining unit 74 1 , a coordinate calibration unit 742 , a digital coordinate determining unit 743 , a digital name determining unit 744 , and a sample set determining unit. 745.

[0147] The image acquisition unit 741 is for acquiring a coin image for training the model.

[0148] The coordinate calibration unit 742 is configured to determine coordinates of four rectangular regions selected by the user on the coin in each of the coin images; wherein the rectangular region is used to represent the year number on the coin Occupied display area.

[0149] The digital coordinate determining unit 743 is for determining the coordinates of the four rectangular regions in the coin image as the coordinates of the four-digit year number on the coin in the coin image, respectively.

[0150] The number name determining unit 744 is for determining the name of the four-digit year number on the coin in each of the coin images.

The sample set determining unit 745 is configured to store the name and the coordinates of the year number on each of the coin images and the coins in each of the coin images to obtain a training sample set.

Optionally, in a specific embodiment of the present invention, the coordinate calibration unit 742 specifically includes a region obtaining unit and a region coordinate determining unit.

[0153] The area obtaining unit is configured to acquire four rectangular areas selected by the user on the coins in each of the coin images.

[0154] The area coordinate determining unit is configured to determine coordinates of two end points of the preset diagonal of each of the rectangular areas according to a preset coordinate calibration strategy, and preset a diagonal of each of the rectangular areas The coordinates of the two endpoints are determined as the coordinates of the rectangular region.

[0155] Optionally, in a specific embodiment of the present invention, the Faster RCNN model includes a feature extraction network \¥0 2019/127075 卩(:17 \2017/118891

And an area extraction network and a target identification network; an input end of the area extraction network and an input end of the target identification network are both connected to an output end of the feature extraction network, and an input end of the target identification network is further The output of the area extraction network is connected. The model training unit 75 specifically includes: a parameter initialization unit 751, a first training unit 752, a parameter update unit 753, a second training unit 754, and a model determination unit 755.

[0156] The parameter initialization unit 751 is configured to initialize a parameter value of the feature extraction network according to a parameter value of the feature extraction model obtained by pre-training, randomly initialize a parameter value of the region extraction network, and adopt an initialized The feature extraction network and the region extraction network extract candidate regions from each of the coin images.

[0157] The first training unit 752 is configured to perform the first training on the target recognition network based on the candidate regions extracted from each of the coin images and the year digital information of the coins in each of the coin images.

[0158] The parameter updating unit 753 is configured to update the parameter value of the area extraction network according to the initial parameter value of the target identification network after the first training, and adopt the initialized feature extraction network and the updated The region extraction network again extracts candidate regions from each of the coin images.

[0159] The second training unit 754 is configured to perform the second training on the target recognition network according to the candidate region extracted from each of the coin images and the year digital information of the coins in each of the coin images. .

[0160] The model determining unit 755 is configured to identify the Fa _S t _er RCNN model composed of the initialized feature extraction network, the updated region extraction network, and the target training network after the second training. Identify the model for the coin year.

[0161] As can be seen from the above, a terminal device provided by this embodiment is pre-built

RCNN model

And using a training sample set consisting of multiple sets of coin images and coin year digital information

The RCNN model is trained to obtain a coin year recognition model.

The target detection and positioning rate is faster and the accuracy is higher. Therefore, the coin year model obtained by training the Fa _S t _er RCNN model is used to identify the year of the coin, which not only improves the efficiency of the coin year recognition, but also improves the efficiency. The accuracy of the coin year identification. And since the positions and postures of the coins in the plurality of coin images included in the training sample set are random, the coin year recognition model trained based on the training sample set can be effective for the years of the coins in the coin images of different specifications. Identification. 9 is a schematic diagram of a terminal device according to still another embodiment of the present invention. As shown in FIG. 9, the terminal device 900 of this embodiment includes: a processor 90, a memory 91, and a computer program 92 stored in the memory 91 and operable on the processor 90. The processor 90, when executing the computer program 92, implements the steps in the various method embodiments described above, such as S11 through S13 shown in FIG. Alternatively, when the processor 90 executes the computer program 92, the functions of the units/units in the above-described respective terminal device embodiments are implemented, for example, the functions of the units 71 to 73 shown in FIG.

[0163] Illustratively, the computer program 92 may be partitioned into one or more units/units, which are stored in the memory ₉₁ and executed by the processor 90. To complete the present invention. The one or more units/units may be a series of computer program instruction segments capable of performing a particular function, the instruction segments being used to describe the execution of the computer program 92 in the terminal device. For example, the computer program 92 can be divided into a first obtaining unit, a model calling unit, and a year determining unit, and the specific functions of each unit are as follows:

[0164] The first acquisition unit is configured to acquire a target coin image to be identified.

[0165] The model calling unit is configured to input the target coin image to the trained coin year recognition model.

[0166] The year determining unit is configured to determine a year of the coin in the target coin image according to the year number information output by the coin year identification model; wherein the year number information is used to represent the year of the coin.

[0167] The terminal device 900 may be a computing device such as a desktop computer, a notebook, a palmtop computer, and a cloud server. The terminal device 900 can include, but is not limited to, a processor 90, a memory 91. It will be understood by those skilled in the art that FIG. 9 is merely an example of a terminal device, and does not constitute a limitation on the terminal device 900, and may include more or less components than those illustrated, or combine some components, or different components. For example, the terminal device may further include an input/output device, a network access device, a bus, and the like.

The processor 90 may be a central processing unit (CPU), or may be another general-purpose processor, a digital signal processor (DSP), or an application specific integrated circuit (Application Specific Integrated Circuit, ASIC), ready-to-use programmable gate array

(Field-Programmable Gate Array, FPGA) or other programmable logic device, discrete gate or transistor logic device, discrete hardware components, etc. The general purpose processor may be a microprocessor or the processor or any conventional processor or the like.

[0169] The memory 91 may be an internal storage unit of the terminal device, such as a hard disk or a terminal of the terminal device. Save. The memory 91 may also be an external storage device of the terminal device, for example, a plug-in hard disk provided on the terminal device, and an intelligent memory card (Smart Media)

Card, SMC), Secure Digital (SD) card, Flash Card, etc. Further, the memory 91 may also include both an internal storage unit of the terminal device and an external storage device. The memory 91 is used to store the computer program and other programs and data required by the terminal device. The memory 91 can also be used to temporarily store data that has been output or is about to be output.

[0170] It will be clearly understood by those skilled in the art that, for convenience and brevity of description, only the division of each functional unit and unit described above is exemplified. In practical applications, the above functions may be assigned differently according to needs. The functional unit and the unit are completed, that is, the internal structure of the terminal device is divided into different functional units or units to complete all or part of the functions described above. Each functional unit and unit in the embodiment may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit, and the integrated unit may be implemented by hardware. Formal implementation can also be implemented in the form of software functional units. In addition, the specific names of the functional units and units are only for the purpose of distinguishing from each other, and are not intended to limit the scope of protection of the present application. For the specific working process of the unit and the unit in the above system, refer to the corresponding process in the foregoing method embodiment, and details are not described herein again.

[0171] In the above embodiments, the descriptions of the various embodiments are all focused, and the parts that are not detailed or described in the specific embodiments may be referred to the related descriptions of other embodiments.

[0172] Those of ordinary skill in the art will appreciate that the elements and algorithm steps of the various examples described in connection with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the solution. A person skilled in the art can use different methods for implementing the described functions for each particular application, but such implementation should not be considered to be beyond the scope of the present invention.

[0173] In the embodiments provided by the present invention, it should be understood that the disclosed terminal devices/systems and methods may be implemented in other manners. For example, the terminal device/system embodiment described above is only illustrative. For example, the division of the unit or unit is only a logical function division. In actual implementation, there may be another division manner, for example, multiple units. Or components can be combined or can be integrated into another Systems, or some features can be ignored, or not executed. Alternatively, the mutual coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interface, terminal device or unit, and may be in electrical, mechanical or other form.

[0174] The units described as separate components may or may not be physically separated, and the components displayed as the unit may or may not be physical units, that is, may be located in one place, or may be distributed to multiple networks. On the unit. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of the embodiment.

[0175] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above integrated unit can be implemented in the form of hardware or in the form of a software functional unit.

[0176] The integrated unit/unit, if implemented in the form of a software functional unit and sold or used as a standalone product, may be stored in a computer readable storage medium. Based on such understanding, the present invention implements all or part of the processes in the foregoing embodiments, and may also be completed by a computer program to instruct related hardware. The computer program may be stored in a computer readable storage medium. The steps of the various method embodiments described above may be implemented when the program is executed by the processor. The computer program includes computer program code, and the computer program code may be in the form of a source code, an object code, an executable file, or some intermediate form. The computer readable medium may include: any entity or terminal device capable of carrying the computer program code, a recording medium, a USB flash drive, a removable hard disk, a magnetic disk, an optical disk, a computer memory, a read only memory (Read-Only Memory, ROM) ), Random Access Memory (RAM), electrical carrier signals, telecommunications signals, and software distribution media. It should be noted that the content contained in the computer readable medium may be appropriately increased or decreased according to the requirements of legislation and patent practice in a jurisdiction, for example, in some jurisdictions, according to legislation and patent practice, computer readable media It does not include electrical carrier signals and telecommunication signals.

The above-described embodiments are only for explaining the technical solutions of the present invention, and are not intended to be limiting thereof; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art will understand that The technical solutions described in the foregoing embodiments are modified, or some of the technical features are equivalently replaced; and the modifications or substitutions do not deviate from the essence of the corresponding technical solutions. \¥0 2019/127075 ?€1^2017/118891 The spirit and scope of the embodiments of the present invention are all included in the scope of protection of the present invention.

Claims

\¥0 2019/127075 卩(:17 \2017/118891 Claims

[Claim 1] A method for identifying a year of a coin, comprising:

Obtaining a target coin image to be identified;

Inputting the target coin image into the trained coin year recognition model;

Determining a year of the coin in the target coin image based on the year number information output by the coin year identification model; wherein the year number information is used to characterize the year of the coin.

[Claim 2] The method according to claim 1, wherein the year number information includes a name and a coordinate of a four-digit year number on a coin in the target coin image; The year number information of the recognition model output determines the year of the coin in the target coin image, including:

The four-digit year number is sorted according to the coordinates of the four-digit year number on the coin in the target coin image and the preset sorting strategy to obtain the year of the coin.

[Claim 3] The method according to claim 2, wherein the four-digit year number is performed according to coordinates of a four-digit year number on a coin in the target coin image and a preset sorting strategy Sorting, obtaining the year of the coin, comprising: processing the target coin image by using a background difference method, obtaining a circular display area where the coin in the target coin image is located, determining a center point of the circular display area coordinate of;

Determining a position of the four-digit year number in the circular display area according to coordinates of a center point of the circular display area and coordinates of four-digit year numbers on the coin; if the four-digit year number All located in the lower half of the circular display area, the four year digits are sorted according to a first preset order to obtain the year of the coin; if the four digits are located in the circle Displaying the upper half of the area, then sorting the four digits in the second preset order to obtain the year of the coin; if the four digits are located in the right half of the circular display area And sorting the four-digit year number according to a third preset order to obtain a year of the coin; if the four-digit year number is located in the left half of the circular display area, according to the fourth pre- The four year digits are sorted in order to obtain the year of the coin. \¥0 2019/127075 卩(:17 \2017/118891

[Claim 4] The method according to claim 1, wherein before the acquiring the target coin image to be identified, the method further includes:

Obtaining a training sample set; wherein each set of sample data in the training sample set is composed of a coin image and year digital information of coins in the coin image, and the year digital information of the coin includes four digits on the coin The name and coordinates of the year number; using the training sample set to train the pre-built region-based fast convolutional neural network & 81 61 RCNN model, which will be trained

The RCNN model is determined as a coin year identification model; wherein the coin year recognition model is used to identify the year of the coin

[Claim 5] The method according to claim 4, wherein the acquiring a training sample set comprises:

Obtaining a coin image for training the model;

Determining a coordinate of four rectangular regions selected by a user on a coin in each of the coin images; wherein the rectangular region is used to represent a display region occupied by a year number on the coin;

Determining the coordinates of the four rectangular regions in the coin image as the coordinates of the four-digit year number on the coin in the coin image;

Determining a name of a four-digit year number on a coin in each of the coin images; associating each coin image, a name and a coordinate of a year number on a coin in each of the coin images, to obtain training Sample set.

[Claim 6] The method according to claim 5, wherein the determining coordinates of four rectangular regions selected by the user on the coins in each of the coin images comprises: acquiring a user at each The four rectangular regions selected by the upper frame of the coin in the coin image; determining the coordinates of the two end points of the preset diagonal of each of the rectangular regions according to a preset coordinate calibration strategy, The coordinates of the two end points of the preset diagonal of the rectangular area are determined as the coordinates of the rectangular area.

[Claim 7] The method according to any one of claims 4 to 6, wherein the ? _& 3·

The RCNN model includes a feature extraction network, a region extraction network, and a target recognition network; \¥0 2019/127075 卩(:17 \2017/118891 The input end of the area extraction network and the input end of the target identification network are both connected to the output end of the feature extraction network, and the input end of the target recognition network is further Connected to an output of the area extraction network;

Pre-constructed region-based fast convolutional neural network using the training sample set

The RCNN model is determined as a coin year identification model, including:

Initializing the parameter values of the feature extraction network according to the parameter values of the feature extraction model obtained by the pre-training, randomly initializing the parameter values of the region extraction network, and adopting the initialized feature extraction network and the region Extracting a network to extract candidate regions from each of the coin images;

Performing a first training on the target recognition network based on candidate regions extracted from each of the coin images and year digital information of coins in each of the coin images; according to the target after the first training Identifying an initial parameter value of the network, updating a parameter value of the area extraction network, and extracting a candidate area from each of the coin images by using the initialized feature extraction network and the updated area extraction network; Performing a second training on the target recognition network based on the candidate region extracted again from each of the coin images and the year digital information of the coins in each of the coin images; the feature extraction network to be initialized And the updated area extraction network and the target recognition network composition after the second training

The type is identified as a coin year recognition model.

[Claim 8] A terminal device, comprising: means for performing the method according to any one of claims 1-7.

[Claim 9] A terminal device comprising a memory, a processor, and a computer program stored in the memory and operable on the processor, wherein the processor is implemented when the computer program is executed The steps of the method of any of claims 1-7.

[Claim 10] A computer readable storage medium storing a computer program, wherein the computer program is executed by a processor to implement the method of any one of claims 1 to 7. The steps of the method.