Quality estimation device, estimation method, and program

Abstract

Provided is a technique for preserving the quality of a food item for an extended period of time. A quality estimation device (21) for estimating the quality of a food item belonging to a prescribed type comprises: a spectral image data acquisition unit (31) for acquiring spectral image data pertaining to a plurality of spectra from a single food item belonging to a type by near-infrared spectroscopy; a quality component content estimation unit (32) for estimating, for each of a plurality of components highly correlated with the quality of the single food item, a quality component content indicating the content of each component by analyzing the spectral image data using a machine learning model; and a score calculation unit (33) for calculating a quality score of the single food item on the basis of the content of each component indicated by the quality component content.

Description

Quality estimation device, estimation method, and program

[0001] The present invention relates to a quality estimation device, an estimation method, and a program, and more particularly to a quality estimation device, an estimation method, and a program that enable the provision of technology for improving the quality of food products.

[0002] Conventionally, several types of components contained in vegetables, such as sugar content, amino acids, lipids, etc., have been used as indicators of vegetable quality. For example, the taste, texture, etc. of vegetables can be estimated based on the content of these components.

[0003] Patent Document 1 discloses a technique in which a measurement beam is irradiated onto a sample, and reflected or transmitted light from the sample is received as detection light, and components contained in the sample are analyzed based on the spectrum of the detection light.

[0004] Patent Document 2 discloses that deviation values ​​are calculated for attributes such as sugar content, acidity, size, etc. of agricultural products, and the results are ranked.

[0005] Japanese Patent Publication No. 11-108825 International Publication WO2015 / 182378

[0006] In recent years, there have been concerns about the impact of labor shortages in the logistics sector, and truck transport capacity in the agricultural and fisheries sectors (fresh foods) is expected to decrease by about 30% from the current level, which is expected to lead to delays in shipping fresh foods from production areas in the future.In addition, transportation periods for exports to overseas markets tend to be longer than those for domestic distribution.Under these circumstances, there is a need to create innovative technologies to improve the quality of fresh foods.

[0007] However, the deviation scores and rankings shown in Patent Document 2, for example, are not sufficient for evaluation from the perspective of extending quality. For example, in a situation where a considerable amount of time must pass between the harvest date and consumption of vegetables, it is necessary to manage the quality of fresh food based on multiple different indices. Furthermore, when displaying such indices, it is necessary to display them in a manner that is easy to understand for consumers, sellers, producers, and the like.

[0008] The technology of Patent Document 1 allows for greater flexibility in the selection of indicators for indicating the quality of foods such as vegetables. However, for example, when measuring water content in addition to sugar content and amino acids, the technology of Patent Document 1 requires a separate water content measurement, which can be time-consuming depending on the desired indicator. Furthermore, from the perspective of extending quality, there has been a demand for technology that can output stable analysis results regardless of the type or shape of food, for example, when managing the quality of fresh foods after a considerable number of days have passed.

[0009] One aspect of the present invention has been made in consideration of the above problems, and one of its objectives is to provide a technology for improving the quality of food products.

[0010] In order to solve the above problem, a quality estimation device according to one embodiment of the present invention is a quality estimation device that estimates the quality of food belonging to a predetermined type, and includes: a spectroscopic image data acquisition unit that acquires spectroscopic image data relating to multiple spectroscopic spectra from one food belonging to the type using near-infrared spectroscopy; a quality component content estimation unit that analyzes the spectroscopic image data using a machine learning model to estimate a quality component content indicating the content level of each of multiple components that have a high correlation with the quality of the one food; and a score calculation unit that calculates a quality score of the one food based on the content levels of the multiple components relating to the quality component content level.

[0011] A quality estimation method according to one aspect of the present invention is a method for estimating the quality of food belonging to a predetermined type, and includes the steps of: acquiring spectroscopic image data relating to multiple spectroscopic spectra from one food belonging to the type using near-infrared spectroscopy; analyzing the spectroscopic image data using a machine learning model to estimate a quality component content indicating the content of each of multiple components that have a high correlation with the quality of the one food; and calculating a quality score of the one food based on the content of the multiple components relating to the quality component content.

[0012] According to one aspect of the present invention, a technology for improving the quality of food can be provided.

[0013] Fig. 1 is a block diagram showing an example of the functional configuration of a quality estimation device according to an embodiment of the present invention. Fig. 2 is a diagram showing a near-infrared spectroscopic image measurement device. Fig. 3 is a diagram explaining an example of spectroscopic image data. Fig. 4 is a diagram showing an example in which the content levels of each component estimated by a quality component content level estimation unit are plotted as a pentagonal radar chart. Fig. 5 is a flowchart explaining an example of the flow of quality estimation processing. Fig. 6 is a diagram showing an example of the configuration of a computer that executes program instructions.

[0014] [First Embodiment] Hereinafter, one embodiment of the present invention will be described in detail.

[0015] (Functional Configuration of Quality Estimation Device) Fig. 1 is a block diagram showing an example of the functional configuration of a quality estimation device according to this embodiment. The quality estimation device 21 shown in the figure estimates the quality of a target food. The target food may be, for example, fresh food such as vegetables, fruits, or meat, and more specifically, fresh produce such as vegetables and fruits. The estimated quality may be, for example, taste quality, tactile quality, or visual quality. Note that other qualities may be included, or none of these qualities may be included.

[0016] As shown in the figure, the quality estimation device 21 includes a spectroscopic image data acquisition unit 31, a quality component content estimation unit 32, a score calculation unit 33, a display control unit 34, a communication unit 35, and a storage unit 40. The quality estimation device 21 is also connected to a network 50 via the communication unit 35, and is configured to be able to communicate with other devices connected to the network 50. In this example, a server 71 is connected to the network 50, and the quality estimation device 21 communicates with the server 71 as necessary.

[0017] The spectroscopic image data acquisition unit 31 acquires spectroscopic image data relating to multiple spectral spectra from one food item belonging to a predetermined type using near-infrared spectroscopy. The quality component content estimation unit 32 analyzes the spectroscopic image data using a machine learning model to estimate a quality component content indicating the content level of each of multiple components that have a high correlation with the quality of the food item. The score calculation unit 33 calculates a quality score for the food item based on the content levels of the multiple components relating to the quality component content levels. The display control unit 34 controls the display of the quality component content levels estimated by the quality component content estimation unit and / or the score calculated by the score calculation unit in a predetermined manner.

[0018] (Spectral Image Data Acquisition Unit) The spectroscopic image data acquisition unit 31 is connected to the near-infrared spectroscopic image measurement device and acquires spectroscopic image data supplied from the near-infrared spectroscopic image measurement device. Fig. 2 is a diagram showing the near-infrared spectroscopic image measurement device 100. In this example, a fruit or vegetable 121 is placed on a sample stage of the near-infrared spectroscopic image measurement device 100. More specifically, a single tomato is placed on the sample stage as the fruit or vegetable 121.

[0019] A photodetector element consisting of, for example, tens of thousands of pixels is placed on top of the near-infrared spectroscopic image measurement device 100. Near-infrared light is irradiated from above the fruit or vegetable 121, and the reflected light, which has been separated into wavelengths by a diffraction grating or the like, is received by the photodetector element. The photodetector element that receives the reflected light outputs a signal indicating the spectrum of the received light corresponding to each pixel.

[0020] In this way, the spectral image data acquisition unit 31 acquires spectral image data including a number of spectral spectra corresponding to the number of pixels of the photodetection element, which is obtained by irradiating near-infrared light onto one piece of fruit or vegetable (in this example, fruit or vegetable 121) and detecting the reflected light using the photodetection element.

[0021] (Spectral Image Data) Fig. 3 is a diagram illustrating an example of spectral image data. Fig. 3 shows an example of spectral image data of a tomato given as the fruit or vegetable 121 in Fig. 2. The diagram on the left of Fig. 3 shows an example in which the spectral image data is displayed as a spectral spectrum, which is a signal output from each pixel of the photodetector element and indicates the frequency (wavelength) and intensity (reflectance) of light received by the pixel. The diagram on the right of Fig. 3 shows an example in which the spectral image data is displayed as a near-infrared spectral image of a tomato, which is the object to be analyzed.

[0022] The left diagram of Figure 3 shows optical spectra corresponding to signals output from two of the pixels of the photodetector element. Here, the two pixels roughly correspond to positions a and b indicated by rectangles in the right diagram of Figure 3. Position a corresponds to the fruit portion of the tomato, and position b corresponds to the stem portion of the tomato. In this way, optical spectra corresponding to signals output from pixels corresponding to positions other than positions a and b in the right diagram of Figure 3 can also be obtained.

[0023] For example, if the pixel arrangement of the photodetector is 137 x 116, then spectral image data having 15,892 (= 137 x 116) different spectral spectra can be obtained from a single tomato (fruit or vegetable 121 in Figure 2). Spectroscopic spectra in the near-infrared region contain chemical information, such as molecular structure, and physical information, such as particle size, about the components of an object. In other words, the components of the object can be calculated by irradiating the object with near-infrared light and detecting changes in reflectance and absorbance.

[0024] Near-infrared light has extremely low energy compared to mid-infrared and far-infrared light, making it possible to measure it non-destructively and without contact. By analyzing spectroscopic image data such as that shown in Figure 3, it is possible to measure a target object containing multiple organic components directly without pretreatment, and perform multi-component analysis.

[0025] (Quality component content estimation unit) The quality component content estimation unit 32 analyzes the spectroscopic image data using a machine learning model to estimate the quality component content, which indicates the content level of each of multiple components that have a high correlation with the quality of a single food product.

[0026] The quality component inclusion level estimation unit 32 uses model parameters 41 stored in the storage unit 40 as model parameters of the machine learning model. That is, the quality component inclusion level estimation unit 32 estimates the quality component inclusion level by performing a calculation using the feature quantities related to the spectroscopic image data acquired by the spectroscopic image data acquisition unit 31 and the model parameters 41. Note that the model parameters may be stored in the server 71, for example, and the quality estimation device 21 may acquire the model parameters via the network 50.

[0027] For example, when estimating the quality of a tomato, the model parameters 41 are learned based on training data that associates the content of quality components with spectroscopic image data obtained from each of a plurality of tomatoes. For a specific type of food, the components that are highly correlated with the quality of that food are assumed to be given in advance.

[0028] Examples of components highly correlated with tomato quality include sugar content, amino acids, pigments, water, and pectin. In this case, the content of these five components is the quality component content. The quality component content, which is made up of the content of the five components, may be estimated as a five-dimensional vector.

[0029] To learn the model parameters 41, for example, several hundred tomatoes are irradiated with near-infrared light and spectroscopic image data is acquired. Then, the sugar content, amino acid content, pigment content, water content, and pectin content of each of the tomatoes are measured in advance. This measurement does not need to be non-destructive or non-contact measurement, and may be performed, for example, using slices of individual tomatoes (destructive analysis may also be performed).

[0030] By associating the thus obtained content levels of quality components of each tomato with the spectroscopic image data, training data is generated. Learning is performed using such training data with a predetermined algorithm to generate model parameters 41. By performing calculations using such model parameters 41, it is possible to estimate the sugar content, amino acid, pigment, moisture, and pectin content levels based on the spectroscopic image data obtained from any one tomato. In other words, a machine learning model is obtained that estimates the sugar content, amino acid, pigment, moisture, and pectin content levels based on the spectroscopic image data obtained from a tomato.

[0031] By using such a machine learning model, the quality component content estimation unit 32 can estimate the quality component content of the tomato that is the estimation target. Here, an example of estimating the quality component content of a tomato has been described as an example, but the quality component content can be similarly estimated for other fruits and vegetables such as eggplant, strawberry, etc. In this way, the quality component content levels are given in advance for multiple foods belonging to a predetermined type, and the quality component content estimation unit 32 estimates the quality component content levels using a machine learning model having model parameters trained based on training data that associates the given quality component content levels with spectroscopic image data obtained from each of the multiple foods.

[0032] Here, an example has been described in which the teacher data is generated by associating the content levels of five components with the spectroscopic image data. However, the teacher data may be generated by associating each of the five components with the spectroscopic image data.

[0033] For example, for each tomato, training data may be generated that associates sugar content with spectroscopic image data, training data may be generated that associates amino acids with spectroscopic image data, training data may be generated that associates pectin with spectroscopic image data, etc. The model parameters 41 generated by learning using such training data are model parameters that correspond to each of the five machine learning models.

[0034] That is, model parameters 41-1 of a machine learning model that estimates the sugar content based on the spectroscopic image data obtained from tomatoes are generated by learning using training data that associates sugar content with spectroscopic image data. Furthermore, model parameters 41-2 of a machine learning model that estimates the amino acid content based on the spectroscopic image data obtained from tomatoes are generated by learning using training data that associates amino acids with spectroscopic image data. Furthermore, model parameters 41-3 of a machine learning model that estimates the pigment content based on the spectroscopic image data obtained from tomatoes are generated by learning using training data that associates pigments with spectroscopic image data. Similarly, model parameters 41-4 of a machine learning model that estimates the water content based on the spectroscopic image data obtained from tomatoes, and model parameters 41-5 of a machine learning model that estimates the pectin content based on the spectroscopic image data obtained from tomatoes are generated.

[0035] The quality component content estimation unit 32 may use such a machine learning model to estimate the content of the quality component of the tomato to be estimated. In this way, by generating model parameters of the machine learning model for each component to be estimated, more accurate estimation of each component becomes possible.

[0036] (Quality Component Content) The sugar content, amino acid content, pigment content, moisture content, and pectin content are examples of the quality component content of tomatoes, and the sugar content and amino acid content are indicators of the taste quality related to the taste of tomatoes. The moisture and pectin content are indicators of the tactile quality related to the texture of tomatoes. Furthermore, the pigment and moisture content are indicators of the visual quality related to the appearance of tomatoes.

[0037] However, the content of other components may be included in the quality component content of tomatoes. Furthermore, a plurality of components may be selected as the components highly correlated with tomato quality without including any of these components. Furthermore, the number of components related to the quality component content is not limited to five, and may be more or less.

[0038] In the above example, the sugar content, amino acid content, pigment content, moisture content, and pectin content are used as the quality component content of a tomato. However, for example, the sugar content, amino acid content, pigment content, moisture content, and pectin content may be used as the quality component content of a tomato and a peach. Alternatively, the sugar content, amino acid content, pigment content, moisture content, and pectin content may be used as the quality component content common to all fruits and vegetables.

[0039] Furthermore, for example, the components related to the quality component content of a tomato and the components related to the quality component content of an eggplant may be different, or the number of those components may be different. Similarly, the components related to the quality component content may be selected such that the first quality component content is the quality component content common to fruit vegetables, the second quality component content is the quality component content common to flower vegetables, the third quality component content is the quality component content common to root vegetables, etc.

[0040] Alternatively, the first quality component content degree, the second quality component content degree, the third quality component content degree, etc. may be selected corresponding to Cruciferous vegetables, Umbelliferous vegetables, Cucurbitaceae vegetables, etc. Furthermore, for example, the first quality component content degree and the second quality component content degree may be selected corresponding to fruits and vegetables that are ripened before eating and other fruits and vegetables.

[0041] Even if the components related to the quality component content are different, training data is generated in the same manner as described above, and model parameters 41 are generated by performing learning using a predetermined algorithm. For example, model parameters 41-11, model parameters 41-12, model parameters 41-13, etc. can be generated corresponding to the first quality component content, the second quality component content, the third quality component content, etc. In this way, according to this embodiment, the degree of freedom in selecting an index indicating the quality of fruits and vegetables can be improved.

[0042] As described above, the content levels of the multiple components indicated by the quality component content level may vary depending on the type of fruit or vegetable to be estimated. Here, the type may be, for example, a type classified by demand part (fruit vegetables, flower vegetables, root vegetables, etc.), a type classified by family (Brassicaceae, Apiaceae, Cucurbitaceae, etc.), or a type classified by whether or not the vegetable is ripened before consumption. Alternatively, the types may be tomatoes, eggplants, broccoli, etc., or the varieties of each vegetable such as tomatoes, eggplants, broccoli, etc.

[0043] In this way, by selecting multiple components whose content levels are indicated by the quality component content level according to the type of fruit or vegetable to be estimated, and performing machine learning using a sufficient amount of training data, stable analysis results can be output regardless of the type, shape, etc. of the fruit or vegetable.

[0044] (Score Calculation Unit) The score calculation unit 33 calculates a quality score for one fruit or vegetable based on the content of multiple components. Here, the quality score is information that indicates, for example, how close the fruit or vegetable is to being in the best condition in terms of taste, appearance, and texture. Generally, the taste, appearance, and texture of fruit or vegetable are in the best condition immediately after harvest.

[0045] For example, if several days have passed since fruit and vegetables were harvested and before they were sold at a retail store, the quality score can indicate how close the fruit and vegetables sold at the retail store are to their original condition immediately after harvest. In other words, the quality score can indicate how much the fruit and vegetables sold at the retail store have deteriorated since their original condition immediately after harvest.

[0046] For example, when sugar content, amino acid content, pigment content, moisture content, and pectin content are used as the quality component content of a tomato, the score calculation unit 33 plots the content of each of the five components as a pentagonal radar chart. Fig. 4 is a diagram showing an example of a pentagonal radar chart plotting the sugar content, amino acid content, pigment content, moisture content, and pectin content of one tomato (e.g., fruit or vegetable 121 in Fig. 2 ) estimated by the quality component content estimation unit 32. Note that the content represents the ratio of the content of each component to the total weight of one tomato (e.g., fruit or vegetable 121 in Fig. 2 ).

[0047] 4, the estimation result of the quality component content of the fruit or vegetable to be estimated is displayed by dotted lines 142 within a pentagonal solid line 141. The score calculation unit 33 calculates a score corresponding to the area of ​​the pentagon indicated by dotted line 142. For example, the score calculation unit 33 calculates the score corresponding to the area of ​​the pentagon indicated by dotted line 142 by setting the area of ​​the pentagon indicated by solid line 141 to 100 and calculating the ratio to the area of ​​the pentagon indicated by dotted line 142.

[0048] The calculation of the quality score by the score calculation unit 33 is not limited to the above-described method. For example, in the above example, the quality component content includes sugar content, amino acid content, pigment content, moisture content, and pectin content. In this case, the quality component content may be a five-dimensional vector, and a score corresponding to the magnitude of the vector may be calculated. Alternatively, measurement results of the fruit's aroma and firmness obtained by another device may be added.

[0049] (Display Control Unit) The display control unit 34 controls the display in a predetermined manner of the quality component content level estimated by the quality component content level estimation unit 32 and / or the quality score calculated by the score calculation unit 33. The display control unit 34 is connected to, for example, a display (not shown), and displays the quality component content level and / or the quality score on the screen of the display.

[0050] The display control unit 34 may display the content levels of each of the multiple components indicated by the quality component content levels as a radar chart corresponding to a polygon with each of the multiple components as a vertex. For example, it may be displayed as a radar chart as shown in Fig. 4. The quality component content level and the quality score may be displayed separately or together.

[0051] Alternatively, the display control unit 34 may display the content levels of each of the multiple components indicated by the quality component content levels as a vector with a number of dimensions corresponding to the multiple components.

[0052] In this way, the indicators for managing the quality of fruits and vegetables can be displayed in a manner that is easy to understand for consumers, sellers, producers, and the like.

[0053] Next, a description will be given of the quality estimation process performed by the quality estimation device 21. Fig. 5 is a flowchart illustrating an example of the flow of the quality estimation process.

[0054] In step S21, near-infrared rays are irradiated onto an object (e.g., fruit or vegetable 121) placed on the sample stage of the near-infrared spectroscopic image measurement device 100. At this time, near-infrared rays are irradiated from above the object, and reflected light is received by the photodetector element of the near-infrared spectroscopic image measurement device 100.

[0055] For example, the irradiation of near-infrared rays may be triggered by the start of the quality estimation process in the quality estimation device 21, and the spectroscopic image data acquisition unit 31 may control the near-infrared spectroscopic image measurement device 100. Alternatively, near-infrared rays may be irradiated by operating an operation unit of the near-infrared spectroscopic image measurement device 100, or near-infrared rays may be irradiated automatically when an object is placed on the sample stage.

[0056] In step S22, the spectroscopic image data acquisition unit 31 acquires spectroscopic image data supplied from the near-infrared spectroscopic image measurement device 100. At this time, spectroscopic image data is acquired, including spectroscopic spectra the number of which corresponds to the number of pixels of the photodetection element, obtained by detecting the reflected light of the near-infrared light irradiated onto the object in step S21 using the photodetection element of the near-infrared spectroscopic image measurement device 100. In other words, spectroscopic image data relating to a plurality of spectroscopic spectra is acquired from one fruit or vegetable belonging to a predetermined type by near-infrared spectroscopy.

[0057] In step S23, the quality component content estimation unit 32 analyzes the spectroscopic image data using a machine learning model, and performs calculations using the feature quantities related to the spectroscopic image data acquired in step S22 and the model parameters 41.

[0058] In step S24, the quality component content estimation unit 32 estimates the quality component content based on the value calculated by the calculation in the process of step S23. That is, the quality component content is estimated, which indicates the content level of each of a plurality of components that have a high correlation with the quality of the object (e.g., fruit or vegetable 121) irradiated with near-infrared rays in step S21.

[0059] In step S25, the score calculation unit 33 calculates a quality score based on the content levels of a plurality of components related to the quality component content levels estimated in the process of step S24.

[0060] For example, when sugar content, amino acid content, pigment content, moisture content, and pectin content are used as the quality component content of fruit or vegetable 121, score calculation unit 33 plots the content of each of the above five components as a pentagonal radar chart. Then, score calculation unit 33 calculates the score corresponding to dotted line 142 by calculating the ratio between the area of ​​the pentagon represented by solid line 141 in the radar chart shown in Figure 4, for example, assuming that the area is 100 and the area of ​​the pentagon represented by dotted line 142.

[0061] Alternatively, the quality component content is treated as a five-dimensional vector, and a score corresponding to the magnitude of the vector is calculated.

[0062] In step S26, the display control unit 34 displays the quality component content degree estimated in the process of step S24 and / or the quality score calculated in the process of step S25 in a predetermined format, for example, on a display (not shown), etc. At this time, the display control unit 34 displays, for example, the content degree of each of the multiple components indicated by the quality component content degree as a radar chart corresponding to a polygon with each of the multiple components as a vertex.

[0063] Alternatively, the display control unit 34 displays the content levels of each of the multiple components indicated by the quality component content levels as a vector with a number of dimensions corresponding to the multiple components.

[0064] As a result of the display in the process of step S26, the quality component content level and the quality score may be displayed separately or together.

[0065] In this way, the quality estimation process is carried out.

[0066] Effect of First Embodiment As described above, the quality estimation device 21 according to this embodiment can display indicators for managing the quality of fruits and vegetables in a manner that is easy to understand for each of consumers, sellers, producers, and the like. Furthermore, because the quality component content level is estimated and the quality score is calculated using a machine learning model, the flexibility in selecting indicators that indicate the quality of fruits and vegetables can be improved. Furthermore, by selecting multiple components whose content levels are indicated by the quality component content level according to the type of fruit or vegetable to be estimated and performing machine learning using a sufficient amount of training data, stable analysis results can be output regardless of the type, shape, etc. of the fruit or vegetable.

[0067] Therefore, according to this embodiment, it is possible to more appropriately control the quality of fruits and vegetables in situations where a considerable amount of time must pass from the harvest date before they are consumed, and it is possible to provide a technology for extending the quality of food products.

[0068] [Embodiment 2] Next, another embodiment of the present invention will be described in detail. In the first embodiment, it is assumed that the quality estimation device 21 is installed at a producer of fruits and vegetables, a shipping location, etc., but the quality estimation device 21 may also be installed in, for example, a fruit and vegetable market, a retail store, etc. In this way, sellers and consumers can also check the quality of fruits and vegetables in a non-destructive manner.

[0069] Furthermore, the model parameters 41 and a program for executing the quality estimation process may be stored in a cloud server, etc. In this way, consumers, sellers, producers, etc. can easily execute the quality estimation process by accessing the cloud server using, for example, a personal computer or a smartphone.

[0070] In addition, the cloud server may be provided with a program that predicts how the quality component content and / or quality score estimated by the quality estimation process will change after a specified number of days have passed.

[0071] By doing this, for example, it becomes possible to simulate in virtual space how much the quality of fruits and vegetables changes after harvest at each stage, such as shipping, transportation, wholesale, retail, etc. As a result, for example, more appropriate distribution of fruits and vegetables can be achieved, and food quality can be improved.

[0072] [Other Embodiments] In the above-described embodiments, the quality estimation device 21 has been described assuming that the quality of fruits and vegetables is estimated, but the subject of quality estimation is not limited to fruits and vegetables. By appropriately selecting components related to the quality component content, the quality of meat, fish, etc. may be estimated, for example. In other words, the quality estimation device 21 may be configured to estimate the quality of fresh food.

[0073] Alternatively, by appropriately selecting components related to the content of quality components, the quality estimation device 21 may estimate the quality of processed meat products, processed dairy products, etc. In other words, the quality estimation device 21 may estimate the quality of various foods, not limited to fresh foods, and calculate a quality score for each food.

[0074] (Example of Software Implementation) The functions of the quality estimation device 21 can be realized by a program that causes a computer to function as the device, and by a program that causes a computer to function as each block of the device.

[0075] Fig. 6 is a block diagram illustrating an example of the physical configuration of a computer 500 used as the quality estimation device 21. As shown in Fig. 6, the computer 500 can be configured by a computer including a bus 510, a processor 501, a main memory 502, an auxiliary memory 503, a communication interface 504, and an input / output interface 505. The processor 501, the main memory 502, the auxiliary memory 503, the communication interface 504, and the input / output interface 505 are connected to one another via the bus 510. An input device 506 and an output device 507 are connected to the input / output interface 505.

[0076] The processor 501 may be, for example, a CPU (Central Processing Unit), a microprocessor, a digital signal processor, a microcontroller, or a combination of these.

[0077] The main memory 502 may be, for example, a semiconductor RAM (random access memory).

[0078] The auxiliary memory 503 may be, for example, a flash memory, a hard disk drive (HDD), a solid state drive (SSD), or a combination of these. The auxiliary memory 503 stores, for example, a program for causing the processor 501 to execute the quality estimation process described above. The processor 501 loads the program stored in the auxiliary memory 503 onto the main memory 502 and executes each instruction included in the loaded program.

[0079] The program may be recorded not temporarily but on one or more recording media that can be read by the computer 500. The recording media may or may not be included in the computer 500. In the latter case, the program may be supplied to the computer 500 via any wired or wireless transmission medium.

[0080] The communication interface 504 is an interface for connecting to a network such as a wired LAN or wireless LAN.

[0081] The input / output interface 505 may be, for example, a USB interface, a short-range communication interface such as infrared or Bluetooth (registered trademark), or a combination of these.

[0082] The input device 506 may be, for example, a keyboard, a mouse, a touchpad, a microphone, or a combination thereof. The output device 507 may be, for example, a display, a printer, a speaker, or a combination thereof.

[0083] [Summary] A quality estimation device according to aspect 1 of the present invention is a quality estimation device that estimates the quality of food belonging to a predetermined type, and includes: a spectroscopic image data acquisition unit that acquires spectroscopic image data relating to a plurality of spectroscopic spectra from one food belonging to the type using near-infrared spectroscopy; a quality component content estimation unit that analyzes the spectroscopic image data using a machine learning model to estimate a quality component content indicating the content level of each of a plurality of components that have a high correlation with the quality of the one food; and a score calculation unit that calculates a quality score of the one food based on the content levels of the multiple components relating to the quality component content levels.

[0084] In the quality estimation device of aspect 2 of the present invention, in the above-mentioned aspect 1, the spectroscopic image data acquisition unit acquires the spectroscopic image data including a number of spectroscopic spectra corresponding to the number of pixels of the photodetection element, which is obtained by irradiating near-infrared light onto the one food item and detecting reflected light using the photodetection element.

[0085] In the quality estimation device according to aspect 3 of the present invention, in the above aspect 1 or 2, the content levels of the multiple components indicated by the quality component content levels vary depending on the type of food to which the food belongs.

[0086] In the quality estimation device of aspect 4 of the present invention, in the above-mentioned aspect 3, the quality component content is given in advance for a plurality of foods belonging to a predetermined type, and the quality component content estimation unit estimates the quality component content using a machine learning model having model parameters learned based on training data that corresponds the given quality component content with the spectroscopic image data obtained from each of the plurality of foods.

[0087] A quality estimation device according to aspect 5 of the present invention is, in the above-mentioned aspect 3 or 4, characterized in that the type is fruits and vegetables, and the content of quality components includes the content of at least one of sugar content, amino acids, pigments, moisture, and pectin.

[0088] A quality estimation device according to aspect 6 of the present invention, in any of aspects 1 to 5 above, further includes a display control unit that controls the display of the quality component content estimated by the quality component content estimation unit and / or the quality score calculated by the score calculation unit in a predetermined manner.

[0089] In the quality estimation device of aspect 7 of the present invention, in the above-mentioned aspect 6, the display control unit controls the display of the content levels of each of the multiple components indicated by the quality component content levels as a radar chart corresponding to a polygon with each of the multiple components as a vertex.

[0090] In the quality estimation device of aspect 8 of the present invention, in the above-mentioned aspect 6, the display control unit controls the display of the content level of each of the multiple components indicated by the quality component content level as a vector with a number of dimensions corresponding to the multiple components.

[0091] A quality estimation method according to aspect 9 of the present invention is a method for estimating the quality of food belonging to a predetermined type, and includes the steps of: acquiring spectroscopic image data relating to a plurality of spectroscopic spectra from one food belonging to the type using near-infrared spectroscopy; analyzing the spectroscopic image data using a machine learning model to estimate a quality component content indicating the content of each of a plurality of components that have a high correlation with the quality of the one food; and calculating a quality score of the one food based on the content of the plurality of components relating to the quality component content.

[0092] A program according to aspect 10 of the present invention causes a computer to function as a quality estimation device that estimates the quality of food belonging to a predetermined type, and includes a spectroscopic image data acquisition unit that acquires spectroscopic image data relating to multiple spectroscopic spectra from one food belonging to the type using near-infrared spectroscopy, a quality component content estimation unit that analyzes the spectroscopic image data using a machine learning model to estimate a quality component content indicating the content level of each of multiple components that have a high correlation with the quality of the one food, and a score calculation unit that calculates a quality score of the one food based on the content levels of the multiple components relating to the quality component content levels.

[0093] 21 Quality estimation device 31 Spectroscopic image data acquisition unit 32 Quality component content estimation unit 33 Score calculation unit 34 Display control unit 35 Communication unit 40 Storage unit 41 Model parameters 50 Network 71 Server

Claims

1. A quality estimation device that estimates the quality of food belonging to a specified type, comprising: a spectroscopic image data acquisition unit that acquires spectroscopic image data relating to multiple spectroscopic spectra from one food belonging to the type using near-infrared spectroscopy; a quality component content estimation unit that analyzes the spectroscopic image data using a machine learning model to estimate a quality component content indicating the content level of each of multiple components that have a high correlation with the quality of the one food; and a score calculation unit that calculates a quality score of the one food based on the content levels of the multiple components relating to the quality component content levels.

2. The quality estimation device according to claim 1, wherein the spectroscopic image data acquisition unit acquires the spectroscopic image data including a number of spectroscopic spectra corresponding to the number of pixels of the photodetection element, the number of spectroscopic spectra being obtained by irradiating near-infrared light onto the one food product and detecting reflected light using the photodetection element.

3. The quality estimation device according to claim 1, wherein the content levels of multiple components indicated by the quality component content level differ depending on the type of food to which the food belongs.

4. The quality estimation device according to claim 3, wherein the quality component content is given in advance for a plurality of foods belonging to a predetermined type, and the quality component content estimation unit estimates the quality component content using a machine learning model having model parameters learned based on training data that associates the given quality component content with the spectroscopic image data obtained from each of the plurality of foods.

5. The quality estimation device according to claim 3, wherein the type is fruits and vegetables, and the content of quality components includes the content of at least one of sugar content, amino acids, pigments, moisture, and pectin.

6. The quality estimation device according to claim 1, further comprising a display control unit that controls the display of the quality component content estimated by the quality component content estimation unit and / or the quality score calculated by the score calculation unit in a predetermined manner.

7. The quality estimation device according to claim 6, wherein the display control unit controls the display of the content levels of each of the multiple components indicated by the quality component content levels as a radar chart corresponding to a polygon with each of the multiple components as a vertex.

8. The quality estimation device according to claim 6, wherein the display control unit controls the display of the content levels of each of a plurality of components indicated by the quality component content levels as a vector with a number of dimensions corresponding to the plurality of components.

9. A method for estimating the quality of food belonging to a predetermined type, comprising the steps of: acquiring spectroscopic image data relating to a plurality of spectroscopic spectra from one food belonging to said type using near-infrared spectroscopy; analyzing the spectroscopic image data using a machine learning model to estimate a quality component content indicating the content level of each of a plurality of components that have a high correlation with the quality of said one food; and calculating a quality score of said one food based on the content levels of said plurality of components relating to said quality component content level.

10. A program that causes a computer to function as a quality estimation device that estimates the quality of food belonging to a specified type, comprising: a spectroscopic image data acquisition unit that acquires spectroscopic image data relating to multiple spectroscopic spectra from one food belonging to the type using near-infrared spectroscopy; a quality component content estimation unit that analyzes the spectroscopic image data using a machine learning model to estimate a quality component content indicating the content level of each of multiple components that have a high correlation with the quality of the one food; and a score calculation unit that calculates a quality score of the one food based on the content levels of the multiple components relating to the quality component content levels.

Images