JP2019008574A - Article determination device, system, method, and program - Google Patents

Abstract

To provide an article determination device, a system, a method, and a program in which a dense and highly accurate article determination including determining the authenticity can be executed.SOLUTION: A transaction management server 14 representing an article determination device comprises a learner 62 for inputting the image feature amount of an article 12 and outputting a multilevel value or a consecutive value indicating the matching degree of the article 12, and a learning processing part 64 for executing the learning processing of the learner 62 using a learning data group D2. The learning processing part 64 assumes the whole or a part of the learning data group D2 as a population, and the same terminates the learning processing when a correct answer ratio curve 80 (84, 86) indicating the relationship between the output level of the learner 62 and the correct answer ratio of output results satisfies the convergence condition.SELECTED DRAWING: Figure 6

Description

  The present invention relates to an article determination apparatus, system, method, and program for determining an article using a learning device.

  Recently, artificial intelligence technology has been dramatically improved, and high-order information processing is becoming feasible. For example, various convergence determination methods for obtaining an optimal solution by learning processing have been proposed.

  Patent Document 1 proposes a method of calculating a total error evaluation function for determining a learning convergence condition using a weighting coefficient including an error evaluation function of each input pattern to be derived and the number of cases for each input pattern. ing.

  Patent Document 2 proposes a method of determining learning convergence based on a correlation coefficient between input data and a weight coefficient between nodes of the input layer and the hidden layer in error back propagation learning.

JP 2011-107885 A International Publication No. 2007/096954

  However, the methods proposed in Patent Documents 1 and 2 are suitable for, for example, prediction of events including rainfall and sea surface temperature. However, the methods are precise and high so as to determine the type or authenticity based on the image information of the article. There is a possibility that it is not suitable for accurate article determination.

  The present invention has been made to solve the above-described problems, and an object thereof is to provide an article determination apparatus, system, method, and program capable of executing precise and highly accurate article determination including authenticity determination.

  The article determination apparatus according to the first aspect of the present invention is a learning device that receives an image feature amount of an article and outputs a multi-value or continuous value indicating the degree of coincidence of the article, and the learning device using a learning data group. A learning processing unit that executes a learning process for the learning unit, wherein the learning processing unit has a whole or a part of the learning data group as a population, and a relationship of a correct answer rate of an output result to an output level of the learning device. A correct answer rate curve is created, and the learning process is terminated when the correct answer rate curve satisfies a convergence condition.

  In this way, the learning process is terminated when the correct answer rate curve having the whole or part of the learning data group satisfies the convergence condition, so that the learning device has a determination characteristic with high statistical regularity (that is, output) The article determination process based on the correct answer rate of the output result with respect to the level can be executed. As a result, precise and highly accurate article determination including authenticity determination can be performed.

  The learning processing unit may end the learning process as satisfying the convergence condition when a change amount of the correct answer rate curve before and after the learning process becomes smaller than a threshold value. By focusing on the convergence of the correct answer rate curve, the learning state can be optimized.

  In addition, the learning processing unit may create the correct answer rate curve in the population composed of only one of learning data related to the real and fake of the article. Thereby, monotonicity (increase or decrease) of the correct answer rate curve is easily obtained, and the statistical regularity is further increased.

  In addition, the learning device outputs a plurality of coincidences according to the classification of the article in batch with respect to the input of the image feature amount, and the learning processing unit outputs the higher ranks output in batch by the learning device. You may create the said correct answer rate curve regarding two or more coincidence degrees. Thereby, the correlation of two or more coincidences can be reflected in the correct answer rate curve.

  In addition, the article determination apparatus outputs the degree of coincidence of the article output from the learning device, the classification result of the article based on the degree of coincidence, and / or information on the commercial transaction, carry-in / out, or customs clearance of the article. An information output unit configured to be able to output the information may be further provided. Thereby, useful information regarding the article to be determined can be provided.

  The article determination system according to the second aspect of the present invention is any of the above-described configurations configured to be able to acquire image information of the article obtained by the portable terminal having a camera capable of imaging the article from the portable terminal. And an article determination device.

  An article determination method according to a third aspect of the present invention is a method using a learning device that receives an image feature quantity of an article and outputs a multi-value or a continuous value indicating the degree of coincidence of the article, and includes a learning data group A correct answer rate curve indicating the relationship of the correct answer rate of the output result with respect to the output level of the learner is created, and the correct answer rate curve satisfies a convergence condition. The learning process ends.

  An article determination program according to a fourth aspect of the present invention is a program that is executed by a computer including a learning device that receives an image feature amount of an article and outputs a multivalue or continuous value indicating the degree of coincidence of the article, The computer creates a correct answer rate curve showing a relationship of a correct answer rate of an output result with respect to an output level of the learning device, wherein the whole or a part of a learning data group is a population, and the correct answer rate curve satisfies a convergence condition. In this case, the learning process for the learning device is terminated.

  According to the article determination device, system, method, and program according to the present invention, precise and highly accurate article determination including authenticity determination can be performed.

1 is an overall configuration diagram of a commercial transaction support system as an article determination system in an embodiment of the present invention. It is a figure which shows the determination result of the articles | goods by a transaction management server. It is a functional block diagram of the article | item determination part shown in FIG. It is a flowchart with which operation | movement description of the learning process part shown in FIG. 3 is provided. It is a figure which shows the classification method of the determination result by a learning device. 6A to 6C are diagrams illustrating examples of creating a correct answer rate curve. It is a figure which illustrates typically the convergence conditions of learning processing. It is a figure which shows the preparation method of the correct answer rate curve in a modification.

  DESCRIPTION OF EMBODIMENTS Hereinafter, an article determination apparatus according to the present invention will be described with reference to the accompanying drawings by giving preferred embodiments in relation to an article determination system, method, and program.

[Description of Commerce Support System 10]
<Overall configuration>
FIG. 1 is an overall configuration diagram of a commercial transaction support system 10 as an article determination system according to an embodiment of the present invention. The business transaction support system 10 is a system that allows a user who has registered for use of a service to exhibit his / her property (hereinafter referred to as an article 12) as an item for sale online.

  This service is, for example, a net flea market, a net auction, or a net pawn shop. In the example shown in the figure, the article 12 is a so-called brand luxury watch, but any type can be used as long as it can be legally traded in addition to bags and accessories (precious metals, jewelry).

  Specifically, the commercial transaction support system 10 includes a transaction management server 14 (article determination device), a storage device 16, and one or a plurality of portable terminals 18.

  The transaction management server 14 is a computer that performs overall control related to the commercial transaction of the article 12. Specifically, the transaction management server 14 includes a communication unit 20 (information output unit), a control unit 22, and a storage unit 24.

  The communication unit 20 is an interface that transmits and receives electrical signals to and from an external device. The control unit 22 includes a processing arithmetic device including a CPU (Central Processing Unit) and an MPU (Micro-Processing Unit). The control unit 22 functions as the database processing unit 26, the article determination unit 28, and the transaction management unit 30 by reading and executing the program stored in the storage unit 24.

  The storage unit 24 is configured of a non-transitory and computer-readable storage medium. Here, the computer-readable storage medium is a portable medium such as a magneto-optical disk, ROM, CD-ROM, or flash memory, or a storage device such as a hard disk built in the computer system. In the example of this figure, the storage unit 24 stores a parameter group 32 described later.

  The storage device 16 is an external storage device capable of constructing a plurality of types of databases related to support processing for commercial transactions, and can exchange data with the transaction management server 14. Specifically, the storage device 16 includes a database relating to learning data (hereinafter referred to as learning information DB 36), a database relating to exhibition information of the article 12 (hereinafter referred to as exhibition information DB 38), and a database relating to commercial transaction information (hereinafter referred to as commercial transaction information DB 40). ) Has been built.

  By the way, the transaction management server 14 is connected to one or a plurality of portable terminals 18 through a network NW (Internet) and a relay device 42 so as to be capable of bidirectional communication. The portable terminal 18 is a device that can be carried by a user (not shown), and is, for example, a smartphone, a tablet, or a wearable terminal.

  Specifically, the portable terminal 18 includes a camera 44 that can image the article 12 and a display unit 46 that displays visible information including image information D1 obtained by the camera 44. The camera 44 may be built in the mobile terminal 18 or may be an external module that is detachable via an external terminal.

<Schematic operation>
The commercial transaction support system 10 in the present embodiment is configured as described above. Next, a schematic operation of the commercial transaction support system 10 will be described with reference to FIG.

(1) Inquiry about exhibition First, after preparing the article 12 to be exhibited, the user images the article 12 using the camera 44 of the mobile terminal 18. Thereafter, the mobile terminal 18 transmits inquiry data including the user information and the image information D1 to the transaction management server 14 in accordance with a user operation. Thereby, the transaction management server 14 acquires inquiry data from the portable terminal 18 via the relay device 42, the network NW, and the communication unit 20.

(2) Determination of article 12 Next, the transaction management server 14 (specifically, the article determination unit 28), based on the image information D1 supplied from the user, the type of the article 12 (including the product name) and / or Alternatively, authenticity determination processing is performed. Thereby, the article determination unit 28 outputs the determination result D3 including the corresponding type (product name) or the reliability of determination (for example, the degree of coincidence described later).

  FIG. 2 is a diagram illustrating a determination result of the article 12 by the transaction management server 14. More specifically, an image of a watch that is “real” (genuine) is shown on the left side of the figure, and an image of a watch that is “fake” (non-genuine) is shown on the right side of the figure. As can be understood from the drawing, the fake wristwatch has a shape similar to an extent that cannot be distinguished from the real watch at first glance.

  However, when the logo portion of the dial is enlarged, there are a difference portion 50 relating to the size of the cavity portion and difference portions 51 to 53 relating to the shape of the tip portion. Here, by performing the determination process by the article determination unit 28, it is possible to capture the morphological features of the above-described different portions 50 to 53 and determine the authenticity of the article 12 with high accuracy.

(3) Provision of market price information D4 Next, the transaction management server 14 (specifically, the database processing unit 26), when the type of the article 12 is correctly determined, performs the commercial transaction of the latest market price information D4 corresponding to this type. Obtained from the information DB 40. This market information D4 includes, for example, a transaction price (purchase price / rate), the latest transaction volume / inventory quantity, and future demand forecast.

  Thereafter, the transaction management server 14 transmits provided data including the determination result D3 and the market price information D4 to the mobile terminal 18. Thereby, the portable terminal 18 acquires the provision data from the transaction management server 14 via the communication unit 20, the network NW, and the relay device 42.

(4) Registration of Exhibition Next, the user confirms the contents of the market price information D4 presented on the display unit 46, and performs an operation indicating that he has agreed to the exhibition. Then, the transaction management server 14 (specifically, the transaction management unit 30) collects the user information, the image information D1, and the determination result D3 after receiving a user operation, and generates exhibition data for each transaction unit. .

  Finally, the database processing unit 26 supplies the generated exhibition data to the storage device 16. Thereby, the exhibition information DB 38 is updated to the latest state, and the exhibition registration of the article 12 is completed.

[Detailed Description of Article Determination Unit 28]
<Functional block diagram>
FIG. 3 is a functional block diagram of the article determination unit 28 shown in FIG. The article determination unit 28 includes a feature amount generation unit 60, a learning device 62, and a learning processing unit 64.

  The feature amount generation unit 60 generates one or more image feature amounts to be used for learning processing or determination processing of the learning device 62. For example, in the case of a convolutional neural network (CNN), the image feature amount corresponds to a set of pixel values extracted using convolution kernels having different sizes or shapes.

  The learning device 62 receives the input of one or more image feature values generated by the feature value generation unit 60, and then outputs the determination result D3 of the article 12 according to the calculation rule constructed by the learning process. The learning device 62 is constructed by using known artificial intelligence techniques including, for example, a neural network, machine learning, and deep learning (deep learning). Note that the learning algorithm may employ any method of supervised learning, unsupervised learning, and reinforcement learning.

  In the example of this figure, the learning device 62 is a hierarchical neural network including one input layer 66, one or a plurality of intermediate layers 68 (hidden layers), and one output layer 70. Here, the number of neurons (illustrated by circles), the number of intermediate layers 68, the definition of input parameters, or the definition of output parameters may be changed as appropriate.

  The calculation rule of the learning device 62 is determined by the value of the parameter group 32 that is a set of parameters. The parameter group 32 may include, for example, a coefficient for specifying a neuron response function, a weighting coefficient for synapse connection, the number of intermediate layers 68, and the number of neurons constituting each layer. The parameter group 32 is stored in the storage unit 24 (FIG. 1) and is read out in a timely manner as necessary.

  The learning processing unit 64 executes the learning process of the learning device 62 using a collection of learning data read from the learning information DB 36 (hereinafter, learning data group D2). Specifically, the learning processing unit 64 includes a parameter update unit 72 and a convergence determination unit 74 described later.

  The parameter update unit 72 compares the correct answer (ideal output value) of the learning data with the actual output value from the learning device 62, and updates each value of the parameter group 32 so that the error of the output value becomes small. Specifically, various methods including an error back propagation method may be used as the update algorithm.

<Details of learning process>
Next, the operation of the article determination unit 28 (learning processing for the learning device 62) will be described with reference to the flowchart of FIG.

  In step S1 of FIG. 4, the learning processing unit 64 reads and acquires the learning data group D2 used for the learning process from the learning information DB 36. The learning data group D2 includes one or more sets of real learning data and / or one or more sets of fake learning data.

  Here, “genuine learning data” correlates the learning image indicating the “genuine” article 12 with the label information corresponding to “non-applicable” only for a specific type and “non-applicable” for other types. Learning data. The “fake learning data” is learning data in which a learning image indicating the “fake” article 12 is associated with label information corresponding to “not applicable” for any type.

  In step S <b> 2, the learning processing unit 64 starts learning processing for the learning device 62 using the learning data group D <b> 2 acquired in step S <b> 1.

  In step S <b> 3, the learning processing unit 64 sequentially executes learning processing using learning data constituting the learning data group D <b> 2. Specifically, [1] the feature amount generation unit 60 generates one or more image feature amounts using the learning image included in the learning data supplied from the learning processing unit 64. [2] The learning device 62 inputs the image feature amount supplied from the feature amount generation unit 60 to the input layer 66, and then outputs the degree of coincidence of the article 12 from the output layer 70 through arithmetic processing in the intermediate layer 68. To do. [3] The parameter update unit 72 updates each value of the parameter group 32 using various update methods including the error back propagation method.

  In step S4, the convergence determination unit 74 creates a correct answer rate curve 80 related to the learning process in step S3. Here, the correct answer rate curve 80 is a characteristic curve showing the relationship of the correct answer rate of the output result to the output level of the learning device 62.

  FIG. 5 is a diagram illustrating a classification method of the determination result by the learning device 62. Here, “Negative” means that the article 12 is genuine, and “Positive” means that the article 12 is fake. In this case, the attribute of the determination result is [1] “true negative” (data total Ntn) for determining “genuine” as “applicable”, and [2] “false positive” for determining “genuine” as “not applicable”. (Total number of data Nfp), [3] “True positive” (data total number Ntp) for determining “fake” as “not applicable”, [4] “False negative” for determining “fake” as “applicable” (total number of data) Nfn).

  6A to 6C are diagrams illustrating examples of creating the correct answer rate curve 80. The horizontal axis of the graph indicates “output level X”, and the vertical axis of the graph indicates “correct answer rate Y”. This output level X corresponds to the largest output value (so-called maximum value) among the output values of a plurality of neurons belonging to the output layer 70 (FIG. 3). Each output value is a degree of coincidence (so-called label value) linked to the type of the article 12 and normalized to the range of [0, 1]. “1” is the output level with the highest degree of coincidence, and “0” is the output level with the lowest degree of coincidence.

  On the other hand, the correct answer rate Y indicates the rate at which the learning device 62 outputs a correct result (that is, true negative or true positive) in a specific population. This population is, for example, all or part of the learning data group D2, and [1] a set of only real learning data, [2] a set of only fake learning data, [3] real learning data and fake learning data Any of a mixed set may be used.

  That is, the correct answer rate curve 80 substantially corresponds to a histogram in which the output level X of [0, 1] is divided into a plurality of data sections. In this case, the combination of the width / number of data sections or the attribute / number of the population may be arbitrarily changed.

FIG. 6A shows a correct answer rate curve 80 in a population composed only of genuine learning data. Specifically, the correct answer rate curve 80 shown in this figure is obtained by sequentially connecting K plots (i = 1 to K) shown in the equations (1) and (2).
X = X (i) (1)
Y = Mtn (i) / {Mtn (i) + Mfp (i)} (2)

  Here, Mtn (i) is the number of “true negative (correct answer)” data at the i-th output level X (i). Mfp (i) is the number of data of “false positive (first type error)” at the output level X (i).

  When the output level X = 1, since the classification (that is, corresponding) of the article 12 is roughly determined correctly, the correct answer rate Y = 1 (that is, 100%). On the other hand, as the output level X approaches 0, the number of populations and the number of incorrect answers (false positives) decrease, so that the correct answer rate Y approaches 0. That is, the correct answer rate curve 80 often shows a monotonically increasing function connecting two points (0, 0) and (1, 1).

FIG. 6B shows a correct answer rate curve 80 in a population composed only of fake learning data. Specifically, the correct answer rate curve 80 shown in this figure is obtained by sequentially connecting the K plots (i = 1 to K) shown in the above equation (1) and the following equation (3).
Y = Mtp (i) / {Mtp (i) + Mfn (i)} (3)

  Here, Mtp (i) is the number of “true positive (correct answer)” data at the i-th output level X (i). Mfn (i) is the number of data of “false negative (second type error)” at the output level X (i).

  When the output level X = 0, since the classification (that is, non-applicable) of the article 12 is roughly determined correctly, the correct answer rate Y = 0 (that is, 0%). On the other hand, as the output level X approaches 1, the number of populations and the number of incorrect answers (false negatives) decrease, and the correct answer rate Y approaches 0. That is, the correct answer rate curve often shows a monotone decreasing function connecting two points (1, 0) and (0, 1).

FIG. 6C shows a correct answer rate curve 80 in a population obtained by mixing real learning data and fake learning data. Specifically, the correct answer rate curve 80 shown in this figure is obtained by sequentially connecting the K plots (i = 1 to K) shown in the above equation (1) and the following equation (4).
Y = {Mtn (i) + Mtp (i)}
/ {Mtn (i) + Mfp (i) + Mtp (i) + Mfn (i)} (4)

  For example, when the mixing ratio of the real product and the fake product is 1: 1, the correct answer rate curve 80 has a curve shape substantially equal to the sum of the two correct answer rates Y shown in FIGS. 6A and 6B.

  As described above, the learning processing unit 64 may create the correct answer rate curve 80 in the population composed of only one of the learning data related to the genuine article and the fake article 12. Thereby, the monotonicity (increase or decrease) of the correct answer rate curve 80 is easily obtained, and the statistical regularity is further increased.

  In step S5, the convergence determination unit 74 determines whether or not the correct answer rate curve 80 created in step S4 satisfies the convergence condition. For example, [1] the difference between the correct answer rate curve 80 and a predetermined target curve is smaller than the threshold, and [2] the amount of change in the correct answer rate curve 80 before and after the learning process is the threshold. Smaller than that.

  In FIG. 7, two correct answer rate curves 80 (hereinafter, the former is referred to as “previous curve 81” and the latter is referred to as “current curve 82”) created in the previous time and the current time are superimposed. The previous curve 81 is approximately expressed by a power function with γold as an index. The current curve 82 is approximately expressed by a power function with γnew as an index.

  Here, the amount of change of the correct answer rate curve 80 may be, for example, the maximum difference value (ΔY) of the correct answer rate Y, or the absolute value of the difference between coefficients (here, exponents) that can specify the shape of the function. (| Γnew−γold |) may also be used.

  In this way, the learning processing unit 64 may end the learning process as satisfying the convergence condition when the change amount of the correct answer rate curve 80 before and after the learning process becomes smaller than the threshold value. By focusing on the convergence of the correct answer rate curve 80, the learning state can be optimized.

  When it is determined that the convergence condition is not satisfied (step S5: NO), the process returns to step S3, and thereafter steps S3 to S5 are sequentially repeated. On the other hand, when it is determined that the convergence condition is satisfied (step S5: YES), the process proceeds to step S6.

  In step S <b> 6, the learning processing unit 64 ends the learning process for the learning device 62. By using this learning device 62, as shown in FIG. 2, it is possible to perform precise and highly accurate article determination including authenticity determination.

[Effects of the commercial transaction support system 10]
As described above, the commercial transaction support system 10 (article determination system) can acquire from the mobile terminal 18 the mobile terminal 18 having the camera 44 capable of imaging the article 12 and the image information D1 of the article 12 obtained by the camera 44. The transaction management server 14 comprised by these.

  Then, the transaction management server 14 (article determination apparatus) [1] a learning device 62 that receives the image feature amount of the article 12 and outputs a multi-value or continuous value indicating the degree of coincidence of the article 12; [2] A learning processing unit 64 that performs learning processing on the learning device 62 using the learning data group D2, and [3] the learning processing unit 64 uses all or part of the learning data group D2 as a population. A correct answer rate curve 80 showing the relationship of the correct answer rate of the output result to the output level of the device 62 is created, and the learning process is terminated when the correct answer rate curve 80 satisfies the convergence condition.

  Further, according to the article determination method and the article determination program, the correct answer rate curve 80 showing the relationship of the correct answer rate of the output result with respect to the output level of the learning device 62 with the whole or a part of the learning data group D2 as a population. When the correct answer rate curve 80 satisfies the convergence condition (step S5: YES), the learning process for the learning device 62 is terminated (step S6).

  In this way, the learning process ends when the correct answer rate curve 80 having the whole or a part of the learning data group D2 as a population satisfies the convergence condition. Therefore, the learning device 62 has a determination characteristic (high statistical regularity) That is, the determination process of the article 12 based on the output result correct answer rate) can be performed. As a result, precise and highly accurate article determination including authenticity determination can be performed.

[Modification]
<First example>
In the example of FIGS. 6A to 6C, the correct answer rate curve 80 is created using one output value (only the maximum value) for one set of learning data, but this is not a limitation.

  FIG. 8 is a diagram showing a method of creating correct answer rate curves 84 and 86 in the modification. Here, it is assumed that the output layer 70 of the learning device 62 has 150 neurons, and the first to 150th ranks are ranked in descending order of output values (label values).

  Here, the correct answer rate curve 84 is a set of learning in which the first output value (= 0.95) is “correct” and the second output value (= 0.62) is “incorrect”. This corresponds to one curve created by double counting the number of samples per data.

  On the other hand, in the correct answer rate curve 86, the first output value (= 0.95) is “correct”, the second output value (= 0.62) is “incorrect”, and the third output The value (= 0.38) is “incorrect answer”, which corresponds to three curves created for each rank for one set of learning data.

  As described above, with respect to the learning device 62 that collectively outputs a plurality of matching degrees according to the classification of the article 12 with respect to the input of the image feature amount, the learning processing unit 64 has the higher rank output collectively by the learning device 62 You may create the correct answer rate curves 84 and 86 regarding two or more coincidences. Thereby, the correlation of two or more degrees of coincidence can be reflected in the correct answer rate curves 84 and 86.

<Second example>
Further, the transaction management server 14 sends the degree of coincidence of the article 12 output from the learning device 62, the classification result of the article 12 based on the degree of coincidence, and / or information related to the commercial transaction, carry-in / out, or customs clearance of the article 12 to the outside. An output device (display, speaker, in addition to the communication unit 20) configured to be able to output the signal may be further provided.

  This makes it possible to provide useful information regarding the article 12 to be determined, and can be applied to various use cases. Specifically, it may be used not only for various operations (including screening and second opinion) related to appraisal, inspection, inspection, security, or customs clearance at ports and airports, but also for education of such operations. Alternatively, the present invention can also be applied to automatic inspection of products using the image information D1 (for example, determination of soldering quality on a printed circuit board, determination of age and freshness of animals and plants including crabs).

[Supplement]
In addition, this invention is not limited to embodiment mentioned above, Of course, it can change freely in the range which does not deviate from the main point of this invention. Or you may combine each structure arbitrarily in the range which does not produce technical contradiction.

  For example, in the above-described embodiment, the article determination apparatus is configured by a general-purpose computer (computer). Instead, any hardware configuration including an integrated circuit such as an FPGA (Field-Programmable Gate Array) is employed. May be.

10 ... Commerce support system (goods judgment system)
12 ... Goods 14 ... Transaction management server (goods judgment device)
DESCRIPTION OF SYMBOLS 16 ... Storage apparatus 18 ... Portable terminal 20 ... Communication part (information output part) 22 ... Control part 24 ... Memory | storage part 26 ... Database processing part 28 ... Article determination part 30 ... Transaction management part 32 ... Parameter group 36 ... Learning information DB
38 ... Exhibition information DB 40 ... Commerce transaction information DB
44 ... Camera 46 ... Display unit 60 ... Feature value generation unit 62 ... Learning device 64 ... Learning processing unit 66 ... Input layer 68 ... Intermediate layer 70 ... Output layer 72 ... Parameter update unit 74 ... Convergence determining unit 80, 84, 86 ... Correct answer rate curve D1 ... Image information D2 ... Learning data group D3 ... Judgment result D4 ... Market price information

Claims (8)

A learning device that takes an image feature amount of an article as an input, and outputs a multi-value or continuous value indicating the degree of coincidence of the article;
A learning processing unit that executes learning processing for the learning device using a learning data group;
With
The learning processing unit creates a correct answer rate curve indicating a relation of a correct answer rate of an output result with respect to an output level of the learning device, wherein the whole or a part of the learning data group is a population, and the correct answer rate curve is converged. An article determination apparatus that ends the learning process when a condition is satisfied. The article determination apparatus according to claim 1,
The article determination apparatus according to claim 1, wherein the learning processing unit ends the learning processing as satisfying the convergence condition when a change amount of the correct answer rate curve before and after the learning processing becomes smaller than a threshold value. In the article determination apparatus according to claim 1 or 2,
The article determination apparatus according to claim 1, wherein the learning processing unit creates the correct answer rate curve in the population composed of only one of learning data related to real and fake articles. The article determination apparatus according to any one of claims 1 to 3,
The learning device collectively outputs a plurality of matching degrees according to the classification of the article with respect to the input of the image feature amount,
The said learning process part produces the said correct answer rate curve regarding two or more high-order coincidence output collectively by the said learning device. The article | item determination apparatus characterized by the above-mentioned. In the article determination apparatus according to any one of claims 1 to 4,
The degree of coincidence of the articles output from the learning device, the classification result of the articles based on the degree of coincidence, and / or information related to commercial transactions, loading / unloading or customs clearance of the articles can be output to the outside. The article determination apparatus further comprising an information output unit. A portable terminal having a camera capable of imaging an article;
The article determination device according to any one of claims 1 to 5, wherein image information of the article obtained by the camera is configured to be acquired from the portable terminal.
An article determination system comprising: An article determination method using a learning device that takes an image feature amount of an article as input and outputs a multi-value or continuous value indicating the degree of coincidence of the article,
Create a correct answer rate curve showing the relationship of the correct answer rate of the output result with respect to the output level of the learning device, the whole or a part of the learning data group, and when the correct answer rate curve satisfies the convergence condition, An article determination method characterized by ending learning processing for a learning device. An article determination program to be executed by a computer including a learning device that receives an image feature amount of an article and outputs a multivalue or continuous value indicating the degree of coincidence of the article,
The computer creates a correct answer rate curve showing a relationship of a correct answer rate of an output result with respect to an output level of the learning device, wherein the whole or a part of a learning data group is a population, and the correct answer rate curve satisfies a convergence condition. In this case, the article determination program ends the learning process for the learning device.

Images