Sake rice quality prediction device and sake rice quality prediction method

Abstract

A sake rice quality prediction device 1 comprises: a section setting unit 1111 that sets a plurality of sections in image data including multi-spectrum information obtained by capturing a sake rice field; and a vegetation index calculation unit 1112 that calculates a vegetation index for each of the sections set by the section setting unit 1111.

Description

Apparatus and method for predicting quality of sake rice

[0001] The present invention relates to an apparatus and a method for predicting quality of sake rice.

[0002] Methods for pre-determining and improving the efficiency of post-harvest processes by estimating the quality of edible rice before harvest have been studied. For example, Patent Document 1 discloses an agricultural work plan formulation support apparatus that efficiently creates an agricultural work plan by linking harvesting work in a field with drying and preparation work in a grain drying preparation facility.

[0003] Japanese Unexamined Patent Publication No. 2023-83973

[0004] According to the technology disclosed in Patent Document 1, it is possible to pre-estimate the moisture content of grains to be harvested using remote sensing data, and formulate a work plan based on the estimated moisture content. However, a method for predicting the quality of sake rice has not been established. Since the quality of sake rice is directly linked to the quality of the produced Japanese sake (sake quality), higher-precision quality evaluation is required, unlike general evaluation methods for agricultural crops.

[0005] The present invention has been made in view of such problems, and an object of the present invention is to provide a sake rice quality prediction apparatus capable of predicting the quality of sake rice in the harvest season with high accuracy.

[0006] The sake rice quality prediction apparatus of the present invention comprises: a section setting unit that sets a plurality of sections in image data including multispectral information obtained by photographing a field of sake rice; and a vegetation index calculation unit that calculates a vegetation index for each of the sections set by the section setting unit.

[0007] The sake rice quality prediction method of the present invention comprises: setting a plurality of sections in image data including multispectral information obtained by photographing a field of sake rice, and calculating a vegetation index for each of the set sections.

[0008] According to the sake rice quality prediction apparatus and method of the present invention, vegetation indices are calculated for a plurality of sections set in the sake rice field, so it is possible to predict the quality of sake rice in the harvest season in finer units with high accuracy.

[0009] By accurately predicting the quality of sake rice during the harvest season, it becomes possible to select the sake rice to be used in the sake brewing process based on the quality of each plot, thereby bringing the quality of the sake produced closer to the desired outcome. Furthermore, since the plots to be used for sake brewing can be determined in advance, preparations before the start of production become easier, improving the productivity of the production process.

[0010] This is a diagram showing the configuration related to the quality prediction device. This is a schematic diagram of the quality prediction device. This is a graph showing the relationship between the vegetation index and the coefficient of determination for Rakufumai. This is a graph showing the relationship between the vegetation index and the coefficient of determination when lodging occurs in Yamada Nishiki. This is a graph showing the change in the MAPE of Rakufumai. This is a graph showing the change in the MAPE of Yamada Nishiki. This is a flowchart showing the processing of the quality prediction device. This is a flowchart showing the process of creating an evaluation map. This is a flowchart showing the process of calculating the evaluation index for each plot. This is a diagram showing an example of the created evaluation map. This is a flowchart showing the plot determination process. This is an explanatory diagram of the processing of the adjustment unit.

[0011] Embodiments of the present invention will be described in detail below with reference to the drawings. In the following description, the same reference numerals are used for identical components, and redundant descriptions are omitted.

[0012] <Creation of Evaluation Map> Figure 1 shows the configuration related to the quality prediction device 1 of this embodiment. The quality prediction device 1 predicts the quality of sake rice at harvest time from images of the sake rice field 10 used for sake production, and determines the plots necessary for production based on the predicted quality. As an example, the sake rice that the quality prediction device 1 predicts includes Yamada Nishiki and Rakufu Mai.

[0013] A multispectrum camera 3 mounted on a drone 2 and a fixed-point camera 4 are used to photograph the field 10. Photography by the multispectrum camera 3 and the fixed-point camera 4 is performed at predetermined intervals (for example, every week). The quality prediction device 1 uses the captured images to predict the quality of the sake rice at harvest time.

[0014] The multispectrum camera 3 is mounted on the drone 2, and while the drone 2 flies over the field 10, it acquires multispectrum images of the field 10. In addition to images captured using normal visible light, the multispectrum camera 3 generates recorded images in specific wavelength bands. The specific wavelength bands acquired include, for example, blue (450 nm ± 16 nm), green (560 nm ± 16 nm), red (650 nm ± 16 nm), red edge (730 nm ± 16 nm), and near-infrared (NIR) (840 nm ± 26 nm).

[0015] The fixed-point camera 4 is installed at a designated location in the field 10 and is used to monitor the condition of the field 10. Preferably, the fixed-point camera 4 is installed at a high elevation, allowing observation of the condition of the entire field 10. The fixed-point camera 4 can observe not only the weather conditions around the field 10 but also the general condition of the rice ears (such as whether or not they are lodged). Furthermore, it is possible to observe various conditions of the field 10 and the rice ears.

[0016] The quality prediction device 1 uses multispectral images of the field 10 taken from above by the multispectral camera 3 and images taken by the fixed-point camera 4 to divide the field 10 into multiple sections and predict the quality of the sake rice at harvest time for each section. The quality prediction device 1 further selects the sections to be used for sake production based on the predicted quality. The quality prediction device 1 may be a standalone computer or may be configured as a virtual processing program on the cloud.

[0017] Figure 2 is a schematic diagram of the quality prediction device 1. The quality prediction device 1 comprises a control unit 11, a database 12, and a communication unit 13. The quality prediction device 1 is configured to communicate with a multispectral camera 3 and a fixed-point camera 4 via the communication unit 13, and acquires captured images.

[0018] The control unit 11 includes an evaluation map generation unit 111 and a plot determination unit 112. The evaluation map generation unit 111 includes a plot setting unit 1111, a vegetation index calculation unit 1112, and an evaluation index calculation unit 1113. The plot determination unit 112 includes a class setting unit 1121, a plot selection unit 1122, and an adjustment unit 1123. The database 12 includes an acquired image database 121, an evaluation map database 122, a selected plot database 123, and an evaluation result database 124.

[0019] Images acquired by the multispectrum camera 3 and the fixed-point camera 4 are recorded in the acquired image database 121. Then, in the processing of the evaluation map generation unit 111, the section setting unit 1111 sets multiple sections within the captured image, the vegetation index calculation unit 1112 calculates the vegetation index for each section, and the evaluation index calculation unit 1113 calculates an evaluation index based on that vegetation index.

[0020] In this way, an evaluation map is created to predict the quality of the field 10 at harvest time using the images recorded in the acquired image database 121, and the created evaluation map is recorded in the evaluation map database 122.

[0021] In the processing in the section determination unit 112, the class setting unit 1121 classifies each section using a threshold value for the evaluation index in the evaluation map created by the evaluation map generation unit 111. The desired quality of sake to be produced is then defined, and the section selection unit 1122 determines the class necessary for producing sake of the desired quality and selects the section to be used for production.

[0022] The adjustment unit 1123 is responsible for optimizing the processing within the partition determination unit 112. For example, the partition determination unit 112 adjusts the threshold used for class division by the class setting unit 1121 and adjusts the internal parameters in partition selection by the partition selection unit 1122.

[0023] The selection results from the section selection unit 1122 are recorded in the selected section database 123. In addition, the quality of the sake rice used in sake production, the sake production process, and the post-production evaluation results are recorded in the evaluation results database 124. The adjustment unit 1123 adjusts the parameters based on the evaluation results recorded in the evaluation results database 124.

[0024] Before providing a detailed explanation of the evaluation map generation unit 111 and the plot determination unit 112, we will first explain the vegetation index used in the evaluation of sake rice. The vegetation index is known as an indicator that quantifies the health and distribution of plants using satellite images and drone images. The vegetation index is mainly calculated using the reflectance of visible light (red, green, blue) and near-infrared light (NIR) and is widely used in agriculture, environmental monitoring, and forest management.

[0025] The quality of sake produced is influenced by the protein content of the sake rice used. The NDVI (Normalized Difference Vegetation Index) and GNDVI (Green Normalized Difference Vegetation Index) are used to estimate the protein content of edible rice. The inventors of this application have confirmed that, similar to edible rice, the NDVI and GNDVI can also be used to estimate the protein content of sake rice.

[0026] NDVI and GNDVI are calculated based on the following formulas. However, NIR represents the reflectance of near-infrared light, Red represents the reflectance of red light, and Green represents the reflectance of green light.

[0027] Furthermore, the accuracy of estimates using NDVI and GNDVI is measured using MAPE (Mean Absolute Percentage Error). MAPE is calculated by summing the absolute values ​​of the error rates for each data point and converting them to a percentage (multiplying by 100), as shown in the following formula. MAPE is an indicator that expresses the error between the predicted value and the actual value as a percentage; the closer to 0%, the higher the accuracy, and the closer to 100%, the lower the accuracy. However, y ｉ ^ represents the predicted value, y ｉ represents the measured value, and n represents the total number of data points. The measured value is an evaluation index for the growth state, and is determined, for example, by the stem diameter and amino acid content of harvested rice ears.

[0028] The configuration of the quality prediction device 1 is not limited to the illustrated example. The control unit 11 is generally a CPU (Central Processing Unit) or GPU (Graphics Processing Unit) that controls the entire system, but it may also include a memory unit used for the operation of the control unit 11. The memory unit consists of flash memory, ROM (Read Only Memory), RAM (Random Access Memory), and hard disk, and can store programs and various data. The database 12 is not limited to being stored inside the quality prediction device 1 or in a memory unit connected to the quality prediction device 1, but may also be stored in network storage. The quality prediction device 1 may further include an information display device such as a touch panel, a display unit that displays information according to the data, and an input unit that accepts input from devices such as touch panels.

[0029] The quality prediction device 1 may be configured to perform predetermined processing by executing a program stored in its memory unit, or it may be a cloud-based virtual processing program executed on a server on a network. Alternatively, the quality prediction device 1 may be configured to perform predetermined processing by executing a program stored in a specific area.

[0030] Figure 3 shows the temporal changes in the relationship between NDVI, GNDVI, and the coefficient of determination during the cultivation of Rakufumai rice. A coefficient of determination closer to 1 indicates a high correlation between the predicted vegetation index and the evaluation results of the harvested sake rice, while a coefficient of determination closer to 0 indicates a low correlation. The coefficient of determination can be calculated using, for example, biomass, which has a high correlation with the total weight of vegetation, or chlorophyll concentration, which has a high correlation with the health of the leaves.

[0031] According to this figure, both NDVI and GNDVI can be used to evaluate sake rice, and the correlation is higher as the harvest time approaches. However, as the rice yellowing progresses as the harvest time approaches, the coefficient of determination decreases. In the results shown, the coefficient of determination is highest 28 days after heading (7 days before harvest).

[0032] Figure 4 shows the temporal changes in the relationship between NDVI, GNDVI, and the coefficient of determination during the cultivation of Yamada Nishiki rice. Here, Yamada Nishiki, which is used as sake rice, has a long height of approximately 130 cm (combined culm length and ear length), making it susceptible to lodging in the wind, whereas Rakufumai is said to be more resistant to lodging, with a height of only about 90 cm.

[0033] In this example, a typhoon approached field 10 approximately 21 days after heading, causing some of the rice ears to locate. After lodging, a decrease in the coefficient of determination is observed for both NDVI and GNDVI, indicating that it is inappropriate to use NDVI and GNDVI after lodging for quality evaluation.

[0034] Figure 5 shows the time-dependent changes in the MAPE (Mape) evaluation index using NDVI and GNDVI in field 10 during the cultivation of Rakufumai rice. In the Rakufumai estimation accuracy shown, after heading, MAPE gradually decreases (estimation accuracy increases), and 28 days after heading (7 days before harvest) MAPE is at its lowest point (estimation accuracy is highest).

[0035] Figure 6 shows the time-dependent changes in the MAPE (Mape Equivalent) in the evaluation index using NDVI and GNDVI during the cultivation of Yamada Nishiki rice. After heading, MAPE gradually decreases (estimation accuracy increases), and reaches its lowest point (estimation accuracy is highest) 14 days after heading.

[0036] Furthermore, up to 14 days after heading, GNDVI has a lower MAPE (higher accuracy) compared to NDVI. This is thought to be because NDVI becomes saturated around 14 days after heading. Therefore, it can be seen that a more accurate evaluation is possible by changing the ratio of NDVI to GNDVI used in the final evaluation according to the number of days elapsed after heading.

[0037] Furthermore, a MAPE score below 10% is considered highly accurate. As shown in Figures 5 and 6, the NDVI and GNDVI for Rakufumai and Yamada Nishiki were 3% and 6% respectively at the heading stage, which is well below the 10% or lower standard for high accuracy. Therefore, it is appropriate to evaluate sake rice using the NDVI and GNDVI at the heading stage.

[0038] Using the characteristics of NDVI and GNDVI shown in Figures 3 to 6 above, the quality prediction device 1 calculates an evaluation index for sake rice. Then, using the calculated evaluation index, the quality prediction device 1 determines which plots within the field 10 will be used for producing sake that achieves the desired sake quality.

[0039] Figure 7 is a flowchart showing the process performed by the quality prediction device 1. First, in step S1, the control unit 11 of the quality prediction device 1 calculates an evaluation index for each section within the field 10 and creates an evaluation map. Next, in step S2, the section determination unit 112 uses the evaluation map created in step S1 to determine the section to be used for sake production.

[0040] Figure 8 is a flowchart detailing the evaluation map creation process in step S1 of Figure 7. This evaluation map creation process is performed by the evaluation map generation unit 111. The evaluation map is created using images captured by the multispectral camera 3 and the fixed-point camera 4.

[0041] The evaluation map creation process is performed by setting a plurality of sections in the photographed farm field 10 and calculating an evaluation index for each of these sections. The evaluation index is calculated for each photographing date, and in principle, the evaluation index is updated using newer photographing data. An evaluation map is generated using the evaluation indices calculated for each section in this manner.

[0042] In step S11, the section setting unit 1111 acquires image data captured by the multispectrum camera 3 and the fixed-point camera 4. The captured image data is recorded in advance in an acquired image database 121, and the section setting unit 1111 acquires the image data by referring to the acquired image database 121.

[0043] In step S12, the section setting unit 1111 generates an orthomosaic image. An orthomosaic image is a single undistorted map image created by stitching together a plurality of aerial images. The multispectrum camera 3 divides the entire farm field 10 into a plurality of parts and photographs the same from above. The section setting unit 1111 corrects the influence of camera distortion and terrain on these captured images, synthesizes the images, and generates an orthomosaic image. In the orthomosaic image, the intensity distribution (spectrum) of each wavelength band at each point in the image is recorded.

[0044] In step S13, the section setting unit 1111 analyzes the orthomosaic image generated in step S12, and sets sections where rice ears are actually planted. The set sections may be each individual rice paddy, or may be a section in which a plurality of rice paddies are grouped together. It is preferable that a pallet for collecting harvested crops in one harvesting operation includes only harvested crops from one section. That is, since harvested crops from one rice paddy are collected in one harvesting operation, if a plurality of sections are set in one rice paddy, it becomes difficult to separate the harvested crops for each section. For this reason, setting a plurality of sections in one rice paddy is not preferable.

[0045] In step S14, the vegetation index calculation unit 1112 calculates a vegetation index for each section, and the evaluation index calculation unit 1113 calculates an evaluation index for each section using the calculated vegetation index.

[0046] Figure 9 is a flowchart showing the calculation process for the evaluation index shown in step S14 of Figure 8. Note that the vegetation index calculation unit 1112 primarily performs steps S141 and S142, while the evaluation index calculation unit 1113 performs steps S143 to S147. However, the division of labor is not limited to this example.

[0047] In step S141, the vegetation index calculation unit 1112 determines whether or not rice ears have headed in the plot by analyzing images captured by the fixed-point camera 4. This determination process is performed only for a predetermined period after rice planting and may be omitted if it exceeds the average period until heading. If it is determined that heading has occurred (S141: YES), then step S142 is executed. If it is determined that heading has not occurred (S141: NO), the vegetation index calculation process is terminated, and the vegetation index for that plot is considered nonexistent (for example, set to zero).

[0048] In step S142, the vegetation index calculation unit 1112 calculates the NDVI and GNDVI values ​​for each plot using the images acquired by the multispectrum camera 3. These values ​​are calculated using the above-described equations (1) and (2).

[0049] In step S143, the evaluation index calculation unit 1113 calculates the evaluation index by changing the ratio of NDVI and GNDVI values ​​for each plot according to the number of days elapsed since heading. As shown in Figures 5 and 6, the MAPES of NDVI and GNDVI differ depending on the number of days elapsed. According to the change in MAPES of Rakufumai shown in Figure 5, there is a discrepancy after 21 days post-heading, and GNDVI is more accurate. Therefore, by gradually increasing the ratio of GNDVI after 21 days post-heading, the estimation accuracy using the evaluation index can be improved. According to the change in MAPES of Yamada Nishiki shown in Figure 6, GNDVI is more accurate until 21 days post-heading. Therefore, by setting a higher ratio of GNDVI until 21 days post-heading, the estimation accuracy using the evaluation index can be improved.

[0050] In step S144, the evaluation index calculation unit 1113 determines whether the rice ears being measured are a predetermined period (e.g., one week) before the planned harvest time. A decrease in the coefficient of determination is observed once a certain timing is exceeded that is 28 days after heading (7 days before the harvest time).

[0051] If the rice ears to be measured are before a predetermined period prior to the planned harvest time (S144: YES), the evaluation index calculation unit 1113 stops the evaluation index calculation process and determines the evaluation index calculated previously as that for that section. If the rice ears are after a predetermined period prior to the harvest time (S144: NO), the evaluation index calculation unit 1113 continues the evaluation index calculation process.

[0052] In step S145, the evaluation index calculation unit 1113 determines whether the rice ears in the plot are in a non-lost state by analyzing the images captured by the fixed-point camera 4. If it is determined that the rice ears are in a non-lost state (S145: YES), the evaluation index calculation unit 1113 then executes the process in step S146. If it is determined that the rice ears are not in a non-lost state (lost) (S145: NO), the evaluation index calculation unit 1113 stops the evaluation index calculation process and decides the evaluation index calculated up to that point as the evaluation index for that plot. As described above, when comparing Yamada Nishiki and Rakufu Mai, Yamada Nishiki has a longer culm length and is more prone to lodging. Therefore, for Yamada Nishiki, the criteria used to determine whether or not it is in a non-lost state may be changed, taking into consideration that it is more prone to lodging than Rakufu Mai.

[0053] In step S146, the evaluation index calculation unit 1113 updates the evaluation index for each section to the latest calculated value. If the calculation of the evaluation index is interrupted in steps S144 and S145, the previously calculated evaluation index is maintained.

[0054] In step S147, the evaluation index calculation unit 1113 creates an evaluation map by associating the evaluation index calculated in step S146 with the respective section.

[0055] In this way, an evaluation index is calculated for each section in the orthomosaic image. This evaluation index can be classified using a predetermined threshold, and each class can be displayed with hatching. In the example above, the evaluation index was calculated using NDVI, which uses red light, and GNDVI, which uses green light, but this is not the only option. When using vegetation indices of multiple different wavelengths of light, the correlation with quality may differ. In such cases, when calculating the evaluation index using vegetation indices of those wavelengths of light, the weighting of each vegetation index should be determined according to the degree of correlation.

[0056] Figure 10 is an example of an evaluation map showing evaluation indices classified using predetermined thresholds. According to this figure, the evaluation indices are visualized through classification using thresholds. Since the protein content can be estimated from the evaluation indices, it is possible to consider which plot of rice harvested for sake production should be used to produce sake of the desired quality.

[0057] <Determination of plots> The following describes how the plot determination unit 112 determines the plots of sake rice to be used in sake production using the evaluation map generated by the evaluation map generation unit 111.

[0058] Generally, the sake brewing process is as follows: First, in the preparation stage, the sake rice is polished, washed, soaked, and steamed, and then koji mold is added to make rice koji. Next, in the fermentation stage, the rice koji, steamed rice, water, and yeast are mixed to make mash, and saccharification and alcohol fermentation proceed. Finally, in the pressing and finishing stages, the fermented mash is pressed to separate the sake from the sake lees, and then the sake is pasteurized, aged, and bottled to complete the sake.

[0059] To improve the quality of sake, it is desirable to select appropriate rice polishing and koji based on the protein content of the sake rice. Table 1 shows the characteristics of sake and sake quality design according to the protein content of the sake rice used as raw material. In this table, protein content is classified into three stages: "high / medium / low," but this is just one example, and classification into two or four or more stages is also possible.

[0060]

[0061] The section determination unit 112 determines the protein content necessary to produce sake of the desired quality based on these characteristics and selects an appropriate section according to the planned production volume. Furthermore, since the quality of sake is often expressed in language, artificial intelligence technology using a large-scale language model (LLM) with advanced language analysis capabilities can also be used for its analysis.

[0062] Figure 11 is a flowchart of the partition determination process shown in step S2 of Figure 7. This partition determination process is performed by the partition determination unit 112. In these processes, steps S21 and S22 are mainly performed by the class setting unit 1121, and steps S23 to S25 are performed by the partition selection unit 1122. However, the division of processing is not limited to this example. In this partition determination process, it is assumed that the protein content of the sake rice is classified into three classes, as shown in Table 1.

[0063] In step S21, the class setting unit 1121 classifies the evaluation index of the evaluation map generated by the evaluation map generation unit 111 into three levels using two thresholds. This sets up classes according to the protein content as shown in Table 1.

[0064] In step S22, the class setting unit 1121 predicts the yield of sake rice for each plot based on the area, etc.

[0065] In step S23, the section selection unit 1122 receives information from the user regarding the sake to be produced, including the desired sake quality and planned production volume.

[0066] In step S24, the section selection unit 1122 refers to the general characteristics of sake produced using pre-stored sake rice for each class and determines one or more classes and their ratios necessary to achieve the desired sake quality. For example, the section selection unit 1122 determines one or more classes to use depending on the similarity between the desired sake quality and the sake quality using a single class shown in Table 1.

[0067] In step S25, the plot selection unit 1122 uses the yield for each plot predicted in step S22 to determine the plots of the class selected in step S24. In this way, the plots necessary to produce the planned amount of sake of the desired quality are determined.

[0068] Furthermore, the sake produced is not limited to sake made from a single class of rice; it may be made by blending multiple classes of rice. When blending different classes of rice, the blending ratio of each class may be determined not only based on the characteristics of each class of rice shown in Table 1, but also based on the sake quality obtained by blending rice with different protein content, as shown in Table 2, in order to produce sake of the desired quality.

[0069]

[0070] Table 2 shows descriptions of different types of sake produced with varying blending ratios. This table shows the blending ratios, flavor profiles, and details for achieving five types of sake: balanced, umami-emphasized, lightness-focused, contrast-focused, and shelf-life-focused.

[0071] In step S24, the plot selection unit 1122, when determining the ratio of sake rice of each class necessary to achieve the desired sake quality, considers not only the properties when using a single class of sake rice as shown in Table 1, but also the properties when blending different classes of sake rice as shown in Table 2. Based on this information, it then determines the plot of field 10 to be used to produce sake of the desired quality.

[0072] Table 3 is used to manage each section of field 10. In this table, each section is assigned a management number, and the field location and producer are recorded. In addition, the class set in step S24 and the yield predicted in step S22 are recorded for each section.

[0073]

[0074] Table 4 is a table for managing the manufacturing process. This table shows the manufacturing lot for each desired sake quality determined by the section determination unit 112, and indicates the blending ratio for each class to achieve that sake quality, as well as the management number of the selected section.

[0075]

[0076] The management data for each section in Table 3, and the management data for the manufacturing process in Table 4, are recorded in the selected section database 123.

[0077] <Parameter Adjustment of Section Determination Unit 112> Using the evaluation map generated by the evaluation map generation unit 111, the section determination unit 112 selects the sections to be used for sake production. In order to improve the quality of the sake produced, it is necessary to optimize the section selection process, and for this purpose, the adjustment unit 1123 provided in the section determination unit 112 is used. As shown below, the adjustment unit 1123 optimizes the operation of the section determination unit 112 by utilizing the rice harvesting period, the production process, and the evaluation results after production.

[0078] Table 5 shows an example of evaluation results before and during harvest. The plots and evaluation index classes in the table are the same as those shown in Table 3, and were determined based on the evaluation map created by the evaluation map generation unit 111. In the harvest evaluation, in addition to the results of the visual evaluation of each plot immediately before harvest, the grade evaluation and measured yield are included as a result of the legally mandated grade inspection conducted after harvest.

[0079]

[0080] Table 6 shows an example of the sake manufacturing process and the evaluation results after manufacturing (at completion). The manufacturing lots in the table are the same as those shown in Table 4, and utilize the sake rice from the plot determined in the plot determination process by the plot determination unit 112.

[0081] The items listed in Table 6 for manufacturing process evaluation, post-production analysis results, and post-production sensory test results are examples only and are not limited to these. The manufacturing process evaluation includes some of the evaluation items for fermentation power in the fermentation process using Aspergillus oryzae. The post-production analysis results include some of the analysis items for the produced sake. The post-production sensory test results include some of the sensory evaluation of the produced sake.

[0082] The evaluation items listed in Tables 5 and 6 are recorded in the evaluation results database 124. Note that the evaluation items listed in Tables 5 and 6 are examples only and are not limited to these. For example, in the manufacturing process evaluation, α-amylase and moisture content are shown as examples of quality evaluation results for koji (rice, koji mold), but glucoamylase, acid protease, acid carboxypeptidase, etc., may also be included. Other factors that may be used include manufacturing conditions (temperature progression, humidity, aroma), sensory evaluation (aroma, sweetness, fermentation, etc.), and enzyme activity analysis results. In post-manufacturing analysis results, in addition to alcohol content and Baumé degree, total acidity, amino acids, and glucose concentration may be used. In post-manufacturing sensory evaluation results, in addition to aftertaste and sweetness / dryness, other sensory items may be used.

[0083] Figure 12 is a diagram illustrating how to adjust the parameters used in the internal processing of the section determination unit 112 using the evaluation results described above. The section determination unit 112 receives the evaluation map created by the evaluation map generation unit 111 and the information on the desired sake quality, and determines the sections necessary for sake production.

[0084] In this process, the plot determination unit 112 classifies each plot by applying a class threshold to the evaluation index shown in the evaluation map. The classification results are recorded in the evaluation result database 124 along with the harvest evaluation as pre-harvest evaluation results, as shown in Table 5.

[0085] The plots determined by the plot determination unit 112 are recorded in the selected plot database 123. Furthermore, once sake is produced, evaluations are performed during and after the production process, and the evaluation results shown in Table 6 are recorded in the evaluation results database 124.

[0086] The adjustment unit 1123 is located inside the section determination unit 112. The adjustment unit 1123 takes the target sake quality as input and quantitatively determines the degree of deviation from the target sake quality using the post-production evaluation (analytical test and sensory test) results recorded in the evaluation result database 124 as the correct value. Since sensory evaluation is expressed in language, the degree of deviation may be calculated using LLM (Large-Scale Language Model) or the like. The degree of deviation may also be expressed in vector form with multiple elements.

[0087] The adjustment unit 1123 determines whether the classification of the sake rice was appropriate based on the degree of deviation, using the class setting process (S21) in the class setting unit 1121. If it is determined that the classification was inappropriate, the threshold used for classification is adjusted using the harvest time, manufacturing process, and post-manufacturing evaluation results. This threshold adjustment can also be performed using machine learning.

[0088] Furthermore, the adjustment unit 1123 can optimize the processing mechanism used in the section selection process (S25) by the section selection unit 1122. In particular, since the evaluation of the fermentation process using koji mold in the manufacturing process has a significant impact on the final quality of the sake, it is also possible to strengthen the weighting of the evaluation results of the fermentation process so that the adjustment unit 1123 can make an appropriate selection.

[0089] In this way, the sake rice quality prediction device 1 uses an evaluation map generation unit 111 to create an evaluation map that predicts the quality of the sake rice to be harvested. Then, the plot determination unit 112 determines which plots to combine to produce sake based on the evaluation map. The adjustment unit 1123 of the plot determination unit 112 can adjust the parameters and processes necessary for more accurate plot selection, taking into account the harvest period, the manufacturing process, and the evaluation results after manufacturing.

[0090] The sake rice quality prediction device 1 of this embodiment includes a plot setting unit 1111 for setting multiple plots in image data including multispectral information of a sake rice field 10, and a vegetation index calculation unit 1112 for calculating a vegetation index for each plot set by the plot setting unit 1111. In this way, since the vegetation index is calculated individually for multiple plots set in the sake rice field 10, it becomes possible to predict the quality of sake rice at harvest time with high accuracy.

[0091] Furthermore, improved accuracy in predicting the quality of sake rice during harvest season allows for the selection of sake rice to be used in the sake brewing process based on the quality of each plot, bringing the resulting sake closer to the desired quality. In addition, since the plots to be used for sake brewing can be determined in advance, preparations before the start of production become easier, leading to improved productivity in the brewing process.

[0092] According to the sake rice quality prediction device 1 of this embodiment, the vegetation index calculation unit 1112 calculates a normalized vegetation index (e.g., NDVI) using a first wavelength light (e.g., red light) and a normalized vegetation index (e.g., GNDVI) using a second wavelength light (e.g., green light). The quality prediction device 1 further includes an evaluation index calculation unit 1113 that uses the two vegetation indices (e.g., NDVI and GNDVI) calculated by the vegetation index calculation unit 1112 to calculate an evaluation index indicating the quality of sake rice at harvest time for each plot.

[0093] Thus, by providing an evaluation index calculation unit 1113 that calculates an evaluation index using two vegetation indices, even if there are areas that cannot be properly evaluated using one vegetation index due to shooting conditions or other factors, an appropriate evaluation can be made by using the other vegetation index. As a result, the occurrence of inappropriate evaluation results can be suppressed.

[0094] According to the sake rice quality prediction device 1 of this embodiment, the evaluation index calculation unit 1113 calculates an evaluation index by combining two vegetation indices (for example, NDVI and GNDVI) in a ratio corresponding to the number of days elapsed since heading. As shown in Figures 5 and 6, the relationship between NDVI and GNDVI and the evaluation of sake rice at harvest time differs depending on the number of days elapsed. For example, according to the time-dependent changes in the MAPE of Rakufumai shown in Figure 5, GNDVI is more accurate after 21 days have elapsed since heading. Therefore, by gradually increasing the ratio of GNDVI after 21 days have elapsed since heading, the estimation accuracy of the evaluation index can be improved.

[0095] According to the sake rice quality prediction device 1 of this embodiment, when the sake rice is Yamada Nishiki, the longer the number of days elapsed since heading, the higher the ratio of GNDVI in the processing of the evaluation index calculation unit 1113. As shown in Figure 6, which shows the change in the MAPE of Yamada Nishiki over time, GNDVI is more accurate until 21 days have passed since heading. Therefore, by setting a higher ratio of GNDVI until 21 days have passed since heading, the estimation accuracy of the evaluation index can be improved.

[0096] According to the sake rice quality prediction device 1 of this embodiment, as shown in Figure 4, the vegetation index calculation unit 1112 detects whether or not the sake rice has fallen over within a plot. If lodging is detected in a part of the plot, the lodged portion is excluded from the period thereafter, and the evaluation index for the remaining area is calculated. In this way, the accuracy of sake rice quality prediction can be improved.

[0097] According to the sake rice quality prediction device 1 of this embodiment, the judgment criteria in the lodging detection process by the vegetation index calculation unit 1112 are adjusted according to the height of the rice ears of sake rice. For example, comparing Yamada Nishiki and Rakufu Mai, Yamada Nishiki is taller and therefore more prone to lodging. Therefore, considering that Yamada Nishiki is more prone to lodging than Rakufu Mai, the accuracy of sake rice quality prediction can be improved by appropriately setting the criteria for determining the non-lodging state.

[0098] According to the sake rice quality prediction device 1 of this embodiment, the vegetation index calculation unit 1112 calculates the vegetation index from after heading until a certain period before harvest. As shown in Figure 3, sufficient estimation accuracy can be obtained even at the heading stage, so the vegetation index can be used from the heading stage. On the other hand, since the estimation accuracy decreases after 28 days have passed since heading, the calculation of the vegetation index can be stopped thereafter to prevent a decrease in estimation accuracy.

[0099] The sake rice quality prediction device 1 of this embodiment includes a plot determination unit 112 that determines the plots to be used for sake production using the vegetation index generated by the vegetation index calculation unit 1112. The plot determination unit 112 includes a class setting unit 1121 that sets a class by applying a threshold to the vegetation index of each plot, and a plot selection unit 1122 that selects a plot to be used for producing sake of the desired quality based on the class set by the class setting unit 1121.

[0100] This configuration allows for the assignment of different quality levels to each plot based on the vegetation index. This makes it possible to select the rice plots used in the sake brewing process according to the class required to achieve the desired sake quality, thereby bringing the resulting sake closer to the intended quality. Furthermore, since a production plan can be created in advance, preparations before the start of production become easier, leading to improved productivity in the production process.

[0101] According to the sake rice quality prediction device 1 of this embodiment, the section selection unit 1122 selects a section suitable for producing sake of the desired quality based on the sake quality evaluated in advance for sake produced using a predetermined class of sake rice shown in Table 1. More specifically, the section determination unit 112 determines the class necessary to achieve the desired sake quality based on the information in Table 1 and selects a section corresponding to that class. In this way, the sections to be used for sake rice production are determined in advance, making preparation before the start of production easier and improving the productivity of the production process.

[0102] According to the sake rice quality prediction device 1 of this embodiment, the section selection unit 1122 selects one or more sections to be used in the production of sake of a desired quality, based on the sake quality evaluated in advance for sake produced using, for example, multiple classes of sake rice shown in Table 2. In detail, the section determination unit 112 determines multiple classes and ratios to achieve the desired sake quality based on the information in Table 2, and selects a section corresponding to that class. In this way, since a class is defined for each section, it becomes possible to produce sake with complex qualities by blending multiple classes.

[0103] According to the sake rice quality prediction device 1 of this embodiment, the section selection unit 1122 uses a large-scale language model (LLM) to analyze the sake quality evaluated in advance and the target sake quality, and selects the section to be used for sake production based on the analysis results. Since the judgment of the aroma and flavor of sake is largely subjective, it is difficult to mechanically select the section to achieve the target sake quality by simply referring to the sake quality shown in Tables 1 and 2. Therefore, by utilizing an LLM with high language analysis capabilities to decompose the sake quality into multiple basic components and analyze them as combinations of these basic components, the selection of the section to achieve the target sake quality can be performed with higher accuracy.

[0104] According to the sake rice quality prediction device 1 of this embodiment, the plot determination unit 112 further comprises an adjustment unit 1123. The adjustment unit 1123 adjusts the threshold used by the class setting unit 1121 for class setting, based on at least one of the post-harvest state, state during the manufacturing process, and post-manufacturing evaluation of the sake rice in the plot selected by the class setting unit 1121, as shown in Tables 5 and 6. By providing a function to adjust the threshold in this way, the selection of plots to achieve the desired sake quality can be performed with higher accuracy.

[0105] According to the sake rice quality prediction device 1 of this embodiment, the plot determination unit 112 colors and displays the set plots in the orthomosaic image data according to the classes set by the class setting unit 1121. By visualizing each plot in this color-coded manner, it becomes easier to manage the predicted quality values ​​for each plot, improving the efficiency of harvesting and post-harvest transportation. Furthermore, the colored map can be used as a reference when a person verifies the validity of the selection results made by the plot selection unit 1122.

[0106] The present invention can be embodied and modified in various ways without departing from its broad spirit and scope. Furthermore, the embodiments described above are for illustrative purposes only and do not limit the scope of the invention. In other words, the scope of the invention is determined not by the embodiments, but by the claims. Various modifications made within the scope of the claims and the equivalent meaning of the invention are considered to be within the scope of the invention.

[0107] In other words, the calculation process of the evaluation index using multiple vegetation indices (NDVI and GNDVI) by the evaluation index calculation unit 1113 can be omitted. That is, although the plot determination process in Figure 11, performed by the plot determination unit 112, is explained using the evaluation index, it may also be performed using the vegetation index directly. Also, although the map generation process in Figure 9, performed by the evaluation map generation unit 111, uses the evaluation index from step S144 onwards, the process in step S143 may be omitted, and only one of the vegetation indices may be used.

[0108] 1: Quality prediction device, 3: Multispectrum camera, 4: Fixed-point camera, 10: Field, 11: Control unit, 111: Evaluation map generation unit, 1111: Plot setting unit, 1112: Vegetation index calculation unit, 1113: Evaluation index calculation unit, 112: Plot determination unit, 1121: Class setting unit, 1122: Plot selection unit, 1123: Adjustment unit

Claims

1. A sake rice quality prediction device comprising: a plot setting unit for setting multiple plots in image data including multispectral information of a sake rice field; and a vegetation index calculation unit for calculating a vegetation index for each plot set by the plot setting unit.

2. The sake rice quality prediction device according to claim 1, further comprising: a vegetation index calculation unit that calculates a first vegetation index using first wavelength light and a second vegetation index using second wavelength light having a different wavelength from the first wavelength light; and an evaluation index calculation unit that calculates an evaluation index indicating the quality of sake rice at harvest time for each plot using the first vegetation index and the second vegetation index calculated by the vegetation index calculation unit.

3. The sake rice quality prediction device according to claim 2, wherein the evaluation index calculation unit generates the evaluation index by combining the first vegetation index and the second vegetation index in a ratio corresponding to the number of days elapsed since heading.

4. The sake rice quality prediction device according to claim 3, wherein the first wavelength light is red light, the second wavelength light is green light, and when the sake rice is Yamada Nishiki, the longer the number of days elapsed after heading, the higher the ratio of the second vegetation index in the processing of the evaluation index calculation unit.

5. The sake rice quality prediction device according to claim 1, wherein the vegetation index calculation unit detects whether or not sake rice has fallen over within the plot, and if lodging of sake rice is detected in a part of the area within the plot, it calculates the vegetation index for the period thereafter using the part of the plot excluding the part where lodging was detected.

6. The sake rice quality prediction device according to claim 5, wherein the criteria for determining whether or not lodging occurs in the lodging detection process performed by the vegetation index calculation unit are changed according to the height of the rice ears of sake rice.

7. The vegetation index calculation unit calculates the vegetation index from after heading until a certain period before the planned harvest time, as described in claim 1.

8. A sake rice quality prediction device according to claim 1, comprising a plot determination unit that determines the plots to be used for sake production using the vegetation index generated by the vegetation index calculation unit, wherein the plot determination unit comprises a class setting unit that sets a class using a threshold for the vegetation index of each plot, and a plot selection unit that selects the plots to be used for sake production of a desired sake quality based on the classes set by the class setting unit.

9. The sake rice quality prediction device according to claim 8, wherein the section selection unit selects the section to be used for producing sake of the desired quality based on the sake quality that has been evaluated in advance for sake produced using one of the sake rice classes.

10. The sake rice quality prediction device according to claim 9, wherein the section selection unit selects one or more sections to be used for producing sake of the desired quality based on the sake quality that has been evaluated in advance for sake produced using a plurality of the classes of sake rice.

11. The sake rice quality prediction device according to claim 8, wherein the section determination unit further comprises an adjustment unit, the adjustment unit adjusts the threshold used by the class setting unit to set the class based on at least one of the following: the state of the sake rice in the section selected by the class setting unit before or at harvest; the state of the sake manufacturing process using the sake rice in the selected section; and an evaluation of the state of the sake produced using the sake rice in the selected section.

12. The sake rice quality prediction device according to claim 8, wherein the section determination unit displays the image data by coloring the sections set by the section setting unit according to the class set by the class setting unit.

13. A method for predicting the quality of sake rice at harvest time, comprising setting multiple plots in image data including multispectral information obtained by photographing a sake rice field, and calculating a vegetation index for each of the set plots.

Images