JP7705603B2 - Information processing device, information processing method, and program - Google Patents

Description

The present invention relates to an information processing device, an information processing method, and a program.

In the food industry, it has been important to detect and remove any problems with the quality of the food itself or any foreign matter contained in the food. In response to this, a method has been proposed that can detect foreign matter contained in food by, for example, irradiating the food to be inspected with light, performing second-order differential processing on the absorption spectrum obtained, and creating a second-order differential spectroscopic image (see, for example, Patent Document 1).

JP 2004-301690 A

However, the technology described in Patent Document 1 above takes time to perform various calculation processes and acquire spectroscopic images. In addition, the technology described in Patent Document 1 above detects foreign objects by simply applying a predetermined calculation process, and the detection accuracy may be insufficient depending on the type of object to be inspected and the data acquisition conditions. Furthermore, most of the conventional technologies, including the technology described in Patent Document 1 above, are intended to detect foreign objects contained in objects to be inspected, and almost none are intended to detect the quality of the objects to be inspected.

The present invention was made in consideration of these circumstances, and aims to provide a technology that can accurately detect objects that satisfy certain conditions.

In order to achieve the above object, an information processing device according to one aspect of the present invention comprises:
An information processing device used for sorting objects,
An acquisition means for acquiring information regarding the intensity of light in a predetermined wavelength range related to the object to be sorted;
A determination means for determining the nature or quality of the object to be sorted based on a learned result by applying a statistical method to information on the reflection or scattering intensity in a predetermined wavelength range of one or more normal objects and defective objects;
an output means for outputting a result of the determination by the determination means;
Equipped with.

An information processing method and program according to one aspect of the present invention are also provided as an information processing method or program corresponding to an information processing device according to one aspect of the present invention.

The present invention provides a technology that can accurately detect objects that satisfy certain conditions.

1 is a diagram showing a configuration of an information processing system according to an embodiment of the present invention; 2 is a block diagram showing an example of a hardware configuration of a learning device in the information processing system of FIG. 1 . 2 is a block diagram showing an example of a hardware configuration of an inference device in the information processing system of FIG. 1 . 4 is a block diagram showing an example of a functional configuration of the analysis device of FIG. 1, the learning device of FIG. 2, and the inference device of FIG. 3. FIG. 1 shows an example of an image including visual features of typical normal and PTD beans. 5 is a diagram showing an example of a method in which the analysis device of FIG. 4 acquires various types of information. FIG. 13 is a diagram showing an example of an average spectrum generated by smoothing measurement information. FIG. 8 is a diagram showing an example of a result of performing a predetermined statistical process on the average spectrum of FIG. 7 . 9 is a diagram showing an example of a result of performing a predetermined statistical process on the average spectrum of FIG. 7, which is an example different from the example of FIG. 8. FIG. FIG. 9 is a diagram showing an example of a result of performing a predetermined statistical process on the result of the statistical process in FIG. 8 . 11 is a diagram showing an example of a result of performing various statistical processes on raw data, which is an example different from the examples in FIGS. 7 to 10 . FIG. 5 is a flowchart illustrating the flow of a learning process executed by the learning device of FIG. 4 . 5 is a flowchart illustrating the flow of an inference process executed by the inference device of FIG. 4 . FIG. 1 is an example of a plot showing the results of various statistical pre-processing steps.

<Overview>
DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will now be described with reference to the accompanying drawings, in which: Fig. 1 is a diagram showing the configuration of an information processing system (hereinafter referred to as "this system") according to an embodiment of the present invention.

Before describing FIG. 1, a Potato Taste Defect (hereinafter, referred to as "PTD") related to this system will be described.
PTD refers to poor quality coffee beans and various associated issues that are a problem particularly in Rwanda, Burundi, the Republic of the Congo, the Democratic Republic of the Congo, Uganda, etc. Although the exact cause of PTD is not clear, it is thought to be caused by, for example, stress caused by irritation of coffee beans by stink bugs or bacteria contained in the saliva of stink bugs.
Specifically, coffee beans having the characteristics of PTD (hereinafter referred to as "PTD beans") are known to emit a strong unpleasant odor when roasted, unlike ordinary coffee beans (hereinafter referred to as "normal beans"), and to impair the flavor characteristic of coffee. For this reason, in the above-mentioned countries, when shipping coffee beans, it is common to select PTD beans at multiple stages after the harvesting of coffee cherries (fruits). However, currently, this selection of normal beans and PTD beans is often done manually (by visual inspection or touch), which is a very time-consuming task and makes it difficult to completely separate them.
Although the present system can be applied to general-purpose test objects, it is particularly useful for efficiently separating PTD beans from normal beans.

FIG. 1 is a diagram showing the configuration of this system.
1, the system includes an analysis device 1, a learning device 2, and an inference device 3. The analysis device 1, the learning device 2, and the inference device 3 are connected to each other via a predetermined network N such as the Internet. Note that the network N is not a required component, and for example, NFC (Near Field Communication), Bluetooth (registered trademark), LAN (Local Area Network), etc. may be used.

The objects to be analyzed by the analysis device 1 are, for example, foods, fruits, medicines, beverages, seasonings, etc., and as described above, are typically coffee beans. The analysis device 1 irradiates light of a predetermined wavelength onto the objects to be analyzed, which include normal beans or PTD beans, and determines whether the objects to be analyzed are normal beans or PTD beans based on the obtained reflection spectrum. Specifically, the analysis device 1 is equipped with a spectrometer 11, a light source 12, a control unit 13, and a memory unit 14.

The spectrometer 11 separates the reflected light through the object to be analyzed and measures the intensity of the spectrum. The spectrometer 11 is composed of a general-purpose spectrometer, a CCD camera, an RGB camera, a hyperspectral camera, a multispectral camera, etc.

The light source 12 is, for example, a halogen lamp. The light source 12 irradiates, for example, light in a predetermined wavelength range onto the subject to be analyzed.

The control unit 13 performs various calculations on the information on the intensity of the spectrum obtained by the spectrometer 11 (hereinafter referred to as "spectral information"). The control unit 13 is composed of a CPU (Central Processing Unit) and the like.

The storage unit 14 stores various data. The storage unit 14 is composed of a general-purpose HDD (Hard Disk Drive) or SSD (Solid State Drive), etc.

The learning device 2 executes various calculation processes for learning based on various data and outputs the results of the learning. The learning device 2 is configured with a general-purpose PC (Personal Computer) or the like.

The inference device 3 executes various processes to determine whether the object to be analyzed is a normal bean or a PTD bean based on a program including the learning results output by the learning device 2, and outputs the results. The inference device 3 is configured with a general-purpose PC (Personal Computer) or the like.

<Hardware Configuration>
FIG. 2 is a block diagram showing an example of a hardware configuration of a learning device in the information processing system of FIG.
The learning device 2 is composed of a general-purpose PC (Personal Computer) etc. As shown in Fig. 2, the learning device 2 includes a control unit 21, a ROM (Read Only Memory) 22, a RAM (Random Access Memory) 23, a bus 24, an input/output interface 25, an output unit 26, an input unit 27, a storage unit 28, a communication unit 29, and a drive 30.

The control unit 21 is composed of a microcomputer including a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), and a semiconductor memory, and executes various processes according to a program recorded in a ROM 22 or a program loaded from a storage unit 28 to a RAM 23.
The RAM 23 also stores information necessary for the control unit 21 to execute various processes.

The control unit 21, ROM 22, and RAM 23 are interconnected via a bus 24. An input/output interface 25 is also connected to this bus 24. An output unit 26, an input unit 27, a memory unit 28, a communication unit 29, and a drive 30 are connected to the input/output interface 25.

The output unit 26 is composed of various liquid crystal displays, speakers, etc., and outputs various information as images and sounds.

The input unit 27 is composed of a keyboard, mouse, etc., and is used to input various information.

The storage unit 28 is configured with a hard disk drive (HDD) or a solid state drive (SSD) and stores various data. In this embodiment, for example, various information including various programs and various databases is stored.

The communication unit 29 controls communication with other information processing devices, etc. via a network N including the Internet.

The drive 30 is provided as necessary. Removable media 41, which may be a magnetic disk, optical disk, magneto-optical disk, or semiconductor memory, is appropriately attached to the drive 30. Programs read from the removable media 41 by the drive 30 are installed in the storage unit 28 as necessary. The removable media 41 can also store various data stored in the storage unit 28 in the same way as the storage unit 28.

FIG. 3 is a block diagram showing an example of a hardware configuration of an inference device in the information processing system of FIG.
The inference device 3 is configured with a general-purpose PC (Personal Computer) etc. As shown in Fig. 3, the inference device 3 includes a control unit 51, a ROM (Read Only Memory) 52, a RAM (Random Access Memory) 53, a bus 54, an input/output interface 55, an output unit 56, an input unit 57, a storage unit 58, a communication unit 59, and a drive 60.
The drive 60 is fitted with a removable medium 71, such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory.
However, since each hardware configuration of the inference device 3 can be basically similar to each hardware configuration of the learning device 2, a description thereof will be omitted here.

Figure 4 is a block diagram showing an example of the functional configuration of the analysis device in Figure 1, the learning device in Figure 2, and the inference device in Figure 3.

As shown in FIG. 4, the control unit 21 of the learning device 2 is provided with a measurement information acquisition unit 80, a calculation processing unit 82, and a learning processing unit 84.
Furthermore, a learning result DB 300 is provided in one area of the memory unit 28 of the learning device 2 .

The measurement information acquisition unit 80 acquires information regarding the intensity of reflection or scattering in a predetermined wavelength range from one or more normal objects and defective objects.
Specifically, the measurement information acquisition unit 80 acquires information (hereinafter referred to as "measurement information") on the measurement results of the analysis object for learning (hereinafter referred to as the "analyte for learning") transmitted from the analysis device 1 via the communication unit 29. The measurement information is, for example, information on the spectral intensity of reflected light with respect to light irradiated onto the analysis object for learning.
It is preferable that the measurement information obtained here functions as teacher data for learning. In other words, the measurement information may include information on whether the measurement information is obtained from normal beans or from PTD beans.

The calculation processing unit 82 performs various calculation processes on the measurement information. Specifically, the calculation processing unit 82 includes a smoothing processing unit 90, a normalization processing unit 92, a centering processing unit 94, and a wavelength selection unit 96.

The smoothing processor 90 executes a process for smoothing the measurement information at each measurement point. The smoothing process performed by the smoothing processor 90 is intended to, for example, remove noise or detect differences without being affected by local trends.
The calculation processing unit 82 also generates an average spectrum by smoothing the spectral intensity of each piece of measurement information acquired by the measurement information acquiring unit 80 .

The normalization processing unit 92 and the centering processing unit 94 each perform calculations to normalize and center the average spectrum generated by the smoothing processing unit 90. It is known that performing such processing can accentuate statistical differences in various pieces of information. The specific results of the centering processing performed by the centering processing unit 94 will be described later with reference to FIG. 9 etc.

The wavelength selection unit 96 uses various statistical methods to select a wavelength region with high priority that contributes to determining whether a bean is normal or PTD bean.
Specifically, in this embodiment, the wavelength selection unit 96 employs a method called PLS-β, which is a variable selection method related to the partial least squares method (PLS), to select a wavelength region with high priority that contributes to determining whether beans are normal or PTD beans.
Here, the PLS method and the PLS-β method will be briefly explained.
The PLS method is a method for generating a linear regression model using latent variables, similar to the principal component regression method, but unlike the principal component regression method, the PLS method determines the latent variables so as to maximize the covariance between the latent variables and the objective variable. The PLS method is often useful when there is a high correlation between explanatory variables, and is suitable for application to data such as spectral intensity, as in this system.
The PLS-β method is a method of converting a PLS regression model into a linear multiple regression model and selecting important explanatory variables based on the magnitude of the absolute value of the regression coefficient. In this system, the PLS-β method is adopted to select variables (wavelength regions) that contribute greatly to determining whether beans are normal or PTD beans. The specific method and results of wavelength selection by the wavelength selection unit 96 will be described later with reference to FIG. 10 etc.

The learning processing unit 84 performs learning by applying a statistical means to the information acquired by the data acquisition means. Specifically, the learning processing unit 84 executes a learning process for the measurement information in each wavelength region selected by the wavelength selection unit 96, and generates a learning result. The learning processing unit 84 stores the generated learning result in the learning result DB 300, and transmits information related to the learning result (hereinafter referred to as "learning result information") to the inference device 3.
In addition, the present system can execute the learning process by combining any statistical method such as neural networks, deep learning, and various regression models as a learning method.
In addition, the learning result information may include, for example, a so-called learned model, and may include various information regarding various programs, algorithms, formulas, etc. used to determine whether beans are normal or PTD beans.

As shown in FIG. 4, the control unit 51 of the inference device 3 is provided with a learning result acquisition unit 120 , a target information acquisition unit 122 , a calculation processing unit 124 , a determination unit 126 , and an output unit 128 .
Furthermore, a learning result DB 400 is provided in one area of the memory unit 58 of the inference device 3 .

The learning result acquisition unit 120 applies statistical means to information regarding the reflection or scattering intensity in a specified wavelength region for one or more normal objects and defective objects to acquire the learning results. Specifically, the learning result acquisition unit 120 acquires the learning result information transmitted from the learning device 2 via the communication unit 59. In addition, the learning result acquisition unit 120 stores the acquired learning result information in the learning result DB 400.

The target information acquisition unit 122 acquires information regarding the reflection or scattering intensity in a predetermined wavelength region of the target object to be selected. Specifically, the target information acquisition unit 122 acquires information (hereinafter referred to as "target information") regarding the spectral intensity of each reflected light for the analysis target object to be selected (hereinafter referred to as "analyte object for measurement") transmitted from the analysis device 1 via the communication unit 59. The target information is, for example, information regarding the spectral intensity of reflected light for light irradiated onto the analysis target object for measurement.

The calculation processing unit 124 performs various statistical processing (normalization processing, centering processing, etc.) on the target information.

The determination unit 126 applies statistical means to information regarding the reflection or scattering intensity in a specified wavelength range for one or more normal objects and defective objects, and determines whether the object to be selected is a normal object or a defective object based on the learning results. Specifically, the determination unit 126 determines whether each measurement analysis object for which target information has been acquired is a normal bean or a PTD bean based on the learning result information acquired by the learning result acquisition unit 120 and the target information acquired by the target information acquisition unit 122.

The output unit 128 performs various processes to output the judgment results made by the judgment unit 126.

FIG. 5 shows an example of an image including visual features of typical normal and PTD beans.
Specifically, in FIG. 5, images of typical normal beans and images of typical PTD beans are displayed. First, as shown in FIG. 5(A), normal beans have no particular discoloration or internal damage and can be recognized by humans as natural beans. In contrast, as shown in FIG. 5(B), PTD beans show holes (caused by the intrusion of stink bugs). In addition to the presence of such unnatural holes, PTD beans may also show green or black discoloration inside the beans, and the presence or absence of such findings can be a criterion for determining whether a bean is normal or PTD. Also, even if there is a hole on the surface of a bean, there are some beans that have no effect inside.
The example in Figure 5 is very easy to understand, but in reality, depending on the size of the hole or the angle of the hole, it may be difficult to distinguish using image recognition or it may take time to visually check. Therefore, if such PTD beans can be efficiently and accurately removed, it is believed that the damage caused by PTD beans can be significantly reduced.

FIG. 6 is a diagram showing an example of a method in which the analysis device of FIG. 4 acquires various types of information.
This system obtains various information (such as the intensity of the reflection spectrum) from a total of six points, three points on the front side and three points on the back side of the coffee bean, as shown in Figure 6. This allows the system to obtain information that has a rough but still spatial resolution, and is believed to enable the system to distinguish between normal beans and PTD beans with higher accuracy.

Next, an example of the results of various statistical processing performed on the acquired measurement information will be described with reference to Figures 7 to 9. One of the purposes of the statistical processing shown in Figures 7 to 9 is to highlight the statistical differences in the measurement information and to confirm that the information acquired by this method reflects the difference in the trends of the spectral intensities of normal beans and PTD beans.
First, Fig. 7 is a diagram showing an example of an average spectrum generated by averaging measurement information, that is, Fig. 7 is a graph of an average spectrum generated by averaging measurement information measured at the six points shown in Fig. 6.

In the example of Figure 7, the graph for normal beans shows a steep rise in the wavelength range from roughly 600 nm to 1100 nm, whereas no steep rise was observed for PTD beans.

Next, Figure 8 is a graph showing the results of differentiation after smoothing and other processing of the average spectrum shown in Figure 7. The example in Figure 8 shows the trend more clearly, with the first derivative data for normal beans showing a decreasing trend in the wavelength range from roughly 600 nm to 1100 nm, whereas the first derivative data for PTD beans shows an increasing trend in the wavelength range from roughly 600 nm to 1100 nm.

Similarly, FIG. 9 is a graph showing the results of normalization and centering on the average spectrum shown in FIG. 7 after smoothing and other processes have been performed.
In the example of Figure 9, the values for normal beans tend to decline to the right, particularly in the wavelength range of approximately 600 nm to 1100 nm, whereas in the same wavelength range for PTD beans, the values do not show a tendency to decline to the right, but rather tend to rise overall from the lower left to the upper right.

Figure 10 is a diagram showing an example of the results of further performing a specified statistical process on the results of the statistical process in Figure 9. In other words, Figure 10 is a graph showing an example of the results of further applying the PLS-β method to the results of the centering process in Figure 9.

In the example of Fig. 10, the random numbers used for dividing the learning data and the verification data are changed, and the frequency (number of times, vertical axis) of each wavelength region selected from the results of 2500 trials is displayed as a result. Therefore, in the example of Fig. 10, the wavelength region with a high frequency of wavelengths on the vertical axis can be said to be a wavelength region with a high priority in verifying whether it contributes to the determination of whether the beans are normal or PTD. Specifically, in the example of Fig. 10, it can be assumed that 370 nm, 410 nm, 450 nm, 490 nm, 520 nm, 550 nm, 565 nm, 580 nm, 600 nm, 620 nm, 635 nm, 660 nm, 675 nm, 730 nm, 760 nm, 800 nm, 875 nm, 935 nm, 960 nm, 995 nm, 1020 nm, 1080 nm, 1090 nm, 1100 nm, 1120 nm, 1140 nm, and 1155 nm are wavelengths with a high contribution.
From these data, the inventors of the present application estimated that the wavelength regions with the greatest contributions are, for example, 550 nm to 580 nm, 600 nm to 680 nm, 715 nm to 745 nm, 785 nm to 815 nm, 935 to 995 nm, 1005 nm to 1035 nm, and 1080 nm to 1120 nm.

An example of the case where learning is performed by changing the combination of selected wavelengths is shown in Fig. 11. Fig. 11 shows an example of the results of performing various statistical processes and learning on data obtained by converting only the spectral intensities at selected wavelengths from the raw data, and confirming the accuracy, and is a diagram showing an example different from the examples in Figs. 7 to 10.
In the example of Figure 11, information such as the number of wavelengths selected as a result of each process, the length of the selected wavelengths, the accuracy rate for normal beans, the accuracy rate for abnormal beans (PTD beans), and the accuracy rate for the whole (normal beans and PTD beans) are compiled in a list.
The inventors of the present application further conducted studies including many other combinations of samples and statistical processing, and clarified that by selecting the four wavelength ranges of 550 nm to 580 nm, 600 nm to 680 nm, 715 nm to 745 nm, and 785 nm to 815 nm, it is possible to stably and accurately distinguish between normal beans and PTD beans. It was also clarified that by selecting two wavelengths, for example, 875 nm and 935 nm, it is possible to distinguish between normal beans and PTD beans with sufficient accuracy.

Here, the inventors learned from observations during measurement experiments that PTD beans have parts that are no different from normal beans and parts that have spectral characteristics unique to PTD beans. In this regard, the present system can perform learning processing and wavelength range selection using the results of calculation processing that retains the influence of the spectral characteristics unique to PTD beans even after processing that reduces the spatial resolution, such as averaging processing.
The system then performs a learning process using only data from the corresponding wavelength range of any high spatial resolution data to determine whether a specific wavelength range identified from the results is valid, and examines the accuracy of determining whether the bean is normal or PTD, and considers the conditions under which a highly accurate determination can be made.
By using this method, the present system can reduce the load of learning processing and wavelength region selection, and also ensure a wide range of options for other calculation processing that can be combined and adopted.

Figure 12 is a flowchart explaining the flow of the learning process executed by the learning device of Figure 4.

In step S1, the measurement information acquisition unit 80 of the learning device 2 acquires measurement information for a total of six points, including the outside and inside of the coffee beans, transmitted from the analysis device 1 via the communication unit 29.

In step S2, the smoothing processing unit 90 of the learning device 2 generates an average spectrum by smoothing the spectral intensity of each piece of measurement information acquired by the measurement information acquisition unit 80.

In step S3, the normalization processing unit 92 and the centering processing unit 94 perform calculations to normalize and center the average spectrum generated by the smoothing processing unit 90, respectively.

In step S4, the wavelength selection unit 96 of the learning device 2 applies various statistical methods (e.g., PLS-β) to the results of the calculations performed by the centering processing unit 94 to select a wavelength region with high priority that contributes to determining whether the beans are normal or PTD beans.

In step S5, the learning processing unit 84 of the learning device 2 executes learning processing on the measurement information for each set of wavelength regions selected by the wavelength selection unit 96, and generates a learning result.
Here, the measurement information used by the learning processing unit 84 for learning may be the acquired measurement information as it is, or an average spectrum based on the measurement information may be used. In this embodiment, the learning processing unit 84 executes the learning process by using non-averaged measurement information at the selected wavelength determined based on the averaged measurement information.

In step S6, the learning processing unit 84 of the learning device 2 transmits learning result information regarding the generated learning result to the inference device 3. This ends the learning process executed by the learning device 2.

Next, the flow of the inference process executed by the inference device 3 will be described. Figure 13 is a flowchart explaining the flow of the inference process executed by the inference device of Figure 4.

In step S21, the learning result acquisition unit 120 of the inference device 3 acquires the learning result information transmitted from the learning device 2 via the communication unit 59.

In step S22, the target information acquisition unit 122 of the inference device 3 acquires the target information of the selected wavelength for each analyte to be measured, which is transmitted from the analysis device 1, via the communication unit 59.

In step S23, the calculation processing unit 124 performs various statistical processing (normalization processing, centering processing, etc.) on the target information.

In step S24, the judgment unit 126 of the inference device 3 judges whether each of the measurement analytes for which target information has been acquired is normal beans or PTD beans, based on the learning result information acquired by the learning result acquisition unit 120 and the target information acquired by the target information acquisition unit 122.

In step S25, the output unit 128 of the inference device 3 outputs the judgment result made by the judgment unit 126. This ends the inference process executed by the inference device 3.

Although one embodiment of the present invention has been described above, the present invention is not limited to the above-mentioned embodiment, and modifications and improvements that can achieve the object of the present invention are included in the present invention.

Here, an example of an application of this system is shown with reference to Figure 14. Figure 14 is an example of a plot showing the results of various statistical preprocessing steps.

In the example of Figure 14, a graph is displayed in which each spectrum that has been subjected to computational processing is projected onto a two-dimensional space with important latent variables as its axis. For example, in the example of Figure 14, a relatively large group of normal beans is displayed in the upper left direction of the figure, while a relatively large group of PTD beans is displayed in the lower right direction of the figure. In this way, this system performs various preprocessing operations on raw data, and by using the results as is, it may be possible to determine whether the analyte being measured is normal or PTD beans.

In the above embodiment, the present system has been described as acquiring various pieces of information from a total of six points on the front and back sides of the coffee beans, but this is not limited thereto. The locations and number of points where the present system acquires various pieces of information are arbitrary and are not limited thereto.
Furthermore, the system may employ methods with higher spatial resolution, such as various cameras and imaging technologies, to acquire various information (such as the intensity of the reflected spectrum).

In the above embodiment, the present system has been described as selecting the wavelength region using the PLS-β method, but is not limited thereto. The present system may employ various statistical methods for variable selection, such as PLS -VIP (Variable Importance in Projection), LASSO (Least Absolute Shrinkage and Selection Operator), Stepwise, and GA-PLS (Genetic Algorithm-PLS).

In the above embodiment, the system is described as being able to stably and accurately distinguish between normal beans and PTD beans by selecting four wavelength regions, 550 nm to 580 nm, 615 nm to 645 nm, 715 nm to 745 nm, and 785 nm to 815 nm, but these selected wavelengths are merely examples. The system may, for example, select wavelengths other than the above-mentioned wavelength regions as the selected wavelengths, or may select only one or more of the four wavelength regions, 550 nm to 580 nm, 615 nm to 645 nm, 715 nm to 745 nm, and 785 nm to 815 nm, as the selected wavelengths.

Although not described in the above embodiment, the system may acquire or label additional information, such as spectral information, image region selection, bean quality (normal or PTD beans), moisture content, etc., for an image captured by a camera. In this case, the system may acquire such information as part of the measurement information.

For example, the above-mentioned system configuration is an example and is not limiting. In particular, the analysis device 1, the learning device 2, and the inference device 3 do not necessarily need to function as separate hardware. For example, the functions of the analysis device 1 and the learning device 2 may be provided together, the functions of the analysis device 1 and the inference device 3 may also be provided together, and the functions of the learning device 2 and the inference device 3 may also be provided together.

Furthermore, for example, the above-described series of processes can be executed by hardware or software.
That is, the functional configuration in Fig. 4 is merely an example and is not limited to the above. It is sufficient that the information processing system is provided with a function capable of executing the above-mentioned series of processes as a whole, and there is no particular limitation on the type of functional block to be used to realize this function. In addition, the location of the functional block is not limited to the example in Fig. 4 and may be arbitrary.
Furthermore, one functional block may be configured as a single piece of hardware, a single piece of software, or a combination of both.

When the series of processes is executed by software, the program constituting the software is installed into a computer or the like from a network or a recording medium.
The computer may be a computer built into dedicated hardware. The computer may also be a computer capable of executing various functions by installing various programs, such as a server, a general-purpose smartphone, or a general-purpose PC (Personal Computer).

The recording medium containing such a program may be configured not only as a removable medium (not shown) that is distributed separately from the device body in order to provide the program to the user, but also as a recording medium that is provided to the user in a state where it is already installed in the device body.

In this specification, the steps of describing a program to be recorded on a recording medium include not only processes that are performed chronologically according to the order, but also processes that are not necessarily performed chronologically but are executed in parallel or individually.
That is, some of the steps in FIG. 12 and FIG. 13 may be modified or omitted as appropriate.

For example, in this specification, the term "system" refers to an overall device that is composed of multiple devices, multiple means, etc.

In other words, it is sufficient for an information processing device to which the present invention is applied to have the following configuration, and various embodiments can be adopted.
That is, the information processing device to which the present invention is applied is
An information processing device used for sorting objects,
An acquisition unit (e.g., an object information acquisition unit 122) that acquires information about the light intensity in a predetermined wavelength range related to the object to be sorted;
A determination means (e.g., a determination unit 126) that applies a statistical method to information regarding the reflection or scattering intensity in a predetermined wavelength range in one or more normal objects and defective objects, and determines the nature or quality of the objects to be selected based on the learned results;
An output unit (e.g., an output unit 128) for outputting a result of the determination by the determination unit;
Any information processing device having the above will suffice.

The determination means may also apply statistical means to information relating to the intensity of reflection or scattering in the wavelength range from 310 nm to 1180 nm, and determine whether the object to be selected is a normal object or a defective object based on the learned results.

The wavelength range may also be calculated by applying statistical methods to information regarding the intensity of reflection or scattering in two or more regions of the object to be selected.

The statistical method may also include a process for reducing the spatial resolution of information relating to the intensity of reflection or scattering in two or more regions of the object to be selected.

The determination means may apply statistical means to information regarding the intensity of reflection or scattering in wavelength ranges including at least one of 550 nm to 580 nm, 600 nm to 680 nm, 715 nm to 745 nm, 785 nm to 815 nm, 935 to 995 nm, 1005 nm to 1035 nm, and 1080 nm to 1120 nm, and determine whether the object to be selected is a normal object or a defective object based on the learned results.

In addition, as a second aspect of the present invention, there is provided an information processing device to which the present invention is applied,
An information processing device for sorting objects,
A data acquisition means (e.g., a measurement information acquisition unit 80) for acquiring information regarding the intensity of reflection or scattering in a predetermined wavelength range in one or more normal objects and defective objects;
A learning means (e.g., a learning processing unit 84) that performs learning by applying a statistical method to the information acquired by the data acquisition means;
Any information processing device having the above will suffice.

As a third aspect of the present invention, there is provided an information processing method to which the present invention is applied, comprising:
1. An information processing method executed by a computer for use in sorting objects, comprising:
An acquisition means for acquiring information regarding the intensity of light in a predetermined wavelength range related to the object to be sorted;
A determination means for determining the nature or quality of the object to be sorted based on a learned result by applying a statistical method to information on the reflection or scattering intensity in a predetermined wavelength range of one or more normal objects and defective objects;
an output means for outputting a result of the determination by the determination means;
Any information processing method including the above will suffice.

In addition, as a fourth aspect of the present invention, a program to which the present invention is applied is
The computer used to select the objects is
An acquisition means for acquiring information regarding the intensity of light in a predetermined wavelength range related to the object to be sorted;
A determination means for determining the nature or quality of the object to be sorted based on a learned result by applying a statistical method to information on the reflection or scattering intensity in a predetermined wavelength range of one or more normal objects and defective objects;
an output means for outputting a result of the determination by the determination means;
Any program that executes a control process including the above is sufficient.

REFERENCE SIGNS LIST 1 Analysis device 2 Learning device 21 Control unit 80 Measurement information acquisition unit 82 Arithmetic processing unit 90 Smoothing processing unit 92 Normalization processing unit 94 Centering processing unit 96 Wavelength selection unit 84 Learning processing unit 300 Learning result DB
3... Inference device 51... Control unit 120... Learning result acquisition unit 122... Target information acquisition unit 124... Arithmetic processing unit 126... Determination unit 128... Output unit 400... Learning result DB

Claims (8)

An information processing device used for sorting objects,
An acquisition means for acquiring information regarding the intensity of light in a predetermined wavelength range related to the object to be sorted;
a determination means for applying a statistical method to learning information regarding the reflection or scattering intensity in a predetermined wavelength range in one or more normal objects and defective objects, and determining the nature or quality of the objects to be selected based on the learning results;
an output means for outputting a result of the determination by the determination means;
Equipped with
the determination means generates an average spectrum by smoothing the spectral intensity of each piece of learning information acquired at locations including at least the front and back sides of the coffee bean of interest, applies a predetermined statistical method to the result of a calculation process that normalizes the average spectrum, and selects a wavelength range with a high priority that contributes to determining whether the bean is normal or PTD, and determines whether the coffee bean of interest is normal or PTD based on an algorithm generated by executing a learning process on the learning information in the selected set of wavelength ranges.
Information processing device. The selection of high priority wavelength regions that contribute to the determination of normal beans or PTD beans was based on the difference in spectral intensity from 600 nm to 1100 nm;
The information processing device according to claim 1 . the determining means applies a statistical means to the learning information regarding the intensity of reflection or scattering in the wavelength region of 310 nm to 1180 nm, and determines whether the object to be sorted is a normal object or a defective object based on the learning result.
The information processing device according to claim 1 . The wavelength range is calculated by performing a statistical method on the learning information regarding the reflection or scattering intensity in two or more regions of the object to be selected;
The information processing device according to claim 3 . The statistical method includes a process of reducing the spatial resolution of the learning information regarding the reflection or scattering intensity in two or more regions of the object to be selected.
The information processing device according to claim 4 . An information processing device for sorting objects,
A data acquisition means for acquiring learning information regarding the intensity of reflection or scattering in a predetermined wavelength range in one or more normal objects and defective objects;
a learning means for performing learning by applying a statistical means to the learning information acquired by the data acquisition means;
Equipped with
the learning means generates an average spectrum by smoothing the spectral intensity of each of the learning information acquired at locations including at least the front and back sides of the coffee beans of the object, and applies a predetermined statistical method to the result of a calculation process in which the average spectrum is normalized, thereby selecting a wavelength region with a high priority that contributes to determining whether the beans are normal or PTD beans, and executes a learning process on the learning information for the selected set of wavelength regions;
Information processing device. 1. An information processing method executed by a computer for use in sorting objects, comprising:
acquiring information about the light intensity in a predetermined wavelength range related to the object to be sorted ;
a determining step of applying a statistical method to learning information regarding the reflection or scattering intensity in a predetermined wavelength range in one or more normal objects and defective objects, and determining the nature or quality of the object to be sorted based on the learning result;
an output step of outputting a result of the determination by the determination step ;
Including,
The determination step generates an average spectrum by smoothing the spectral intensity of each piece of learning information acquired at a location including at least the front and back sides of the coffee bean of interest, applies a predetermined statistical method to the result of a calculation process that normalizes the average spectrum, and determines whether the coffee bean of interest is a normal bean or a PTD bean based on an algorithm generated by executing a learning process on the learning information in the selected set of wavelength regions.
Information processing methods. The computer used to select the objects is
An acquisition means for acquiring information regarding the intensity of light in a predetermined wavelength range related to the object to be sorted;
a determination means for applying a statistical method to learning information regarding the reflection or scattering intensity in a predetermined wavelength range in one or more normal objects and defective objects, and determining the nature or quality of the objects to be selected based on the learning results;
an output means for outputting a result of the determination by the determination means;
Executing a control process including
the determination means generates an average spectrum by smoothing the spectral intensity of each piece of learning information acquired at locations including at least the front and back sides of the coffee bean of interest, applies a predetermined statistical method to the result of a calculation process that normalizes the average spectrum, and selects a wavelength range with a high priority that contributes to determining whether the bean is normal or PTD, and determines whether the coffee bean of interest is normal or PTD based on an algorithm generated by executing a learning process on the learning information in the selected set of wavelength ranges.
program.