JP7023180B2 - Sake rice analyzer - Google Patents

Description

The present invention relates to a sake rice analyzer that analyzes the morphology of sake rice.

In sake brewing, the processing process of raw rice such as rice milling, soaking, and steaming is an important process that influences the result of the subsequent brewing process. In particular, in soaking, in Daiginjo sake, etc., the work of adjusting the water absorption amount in 1% units, which is called limited water absorption, may be performed, and it is important to grasp the change in the water content of sake rice in detail to improve the quality of sake. It is important to plan. However, it is known that the water content of rice varies greatly depending on the condition of the raw rice (variety, rice polishing rate, moisture content, etc.) and immersion conditions (time, temperature, concentration, etc.).

Non-Patent Document 1 describes a technique entitled "Method for Measuring Immersion Cracks in White Rice for Sake Brewing". In this document, it is described that cracks occur in sake rice due to immersion, and cracks in sake rice are visually determined to calculate the rate of cracks in sake rice (immersion rate).

Shigeki Nakayama, Toru Takahashi, "Measuring method of immersion cracking of white rice for sake brewing", Iwate Industrial Research Institute Research Report, 2006, No. 13

With the technique described in Non-Patent Document 1, that is, the technique for determining whether or not the sake rice is cracked by human visual determination, it is difficult to continue to accurately determine the form of sake rice in a short time. be.
In particular, since the morphology of sake rice placed in a state of being immersed in water may change from moment to moment, it is required to develop a method for accurately determining the morphology of sake rice in a short time. , The technique described in Non-Patent Document 1 cannot meet such a demand.

The present invention has been made in view of the above problems, and an object of the present invention is to provide a sake rice analyzer capable of obtaining a determination result regarding the form of sake rice accurately in a short time.

The characteristic configuration of the sake rice analyzer according to the present invention for achieving the above object is an image data acquisition unit for acquiring image data obtained by photographing sake rice placed in a state of being immersed in water, and the image data acquisition unit. In the image data acquired by the company, an image processing unit that performs image processing including a process of extracting sake rice image data in a region containing only one sake rice and generates processed image data of the one sake rice. Using the machine learning execution result of the training data composed of the accumulated combination of the processed image data generated by the image processing unit and the information on the morphology of sake rice contained in the processed image data , As the form of the one sake rice in the processed image data generated by the image processing unit, the rice is not damaged, the rice is cracked, or the rice is cracked. A morphological determination unit that determines whether or not the cracks are in a state in which the gaps between the cracks are large , and a validity that determines the validity of the liquor rice determined by the morphological determination unit. The form is provided with a sex determination unit and a form storage unit that stores the determination result of the form determination unit in a state where the change history of the form of the one sake rice after the start of immersion in water can be known. The determination unit adds a combination of information on the form of sake rice determined to be highly appropriate by the validity determination unit and the processed image data for which the form has been determined to the existing training data. After executing the machine learning, the validity determination unit determines that the certainty of the form of the sake rice determined by the form determination unit is equal to or higher than a predetermined value, and the rice is soaked in water. When the change history of the above-mentioned form conforms to a predetermined change criterion, it is determined that the sake rice determined by the form determination unit has high validity, and the sake rice determined by the form determination unit. If the certainty of the above-mentioned form is less than the above-mentioned predetermined value, or if the change history of the above-mentioned form of sake rice after the start of immersion in water does not meet the predetermined change criteria, the above-mentioned The point is that the morphological determination unit determines that the validity of the above-mentioned form of sake rice is low .

According to the above-mentioned feature configuration, the form of sake rice (the form in which the rice is not damaged or the form in which the rice is not damaged) is obtained from the image data obtained by the image data acquisition unit, which is obtained by photographing the rice soaked in water. It is possible to determine whether the rice has cracks or the cracks in the rice have a large gap). In particular, in this feature configuration, the morphological determination unit does not determine the morphology of sake rice, but the processed image data generated by the image processing unit using the machine learning execution result of the accumulated training data. Determine the form of one sake rice. As a result, it is possible to obtain a consistent determination result in a short time regarding the form of sake rice. In addition, the image data referred to by the morphological determination unit for determining the morphology of the liquor rice is a liquor rice image in a region containing only one liquor rice in the image data acquired by the image data acquisition unit by the image processing unit. It is processed image data after performing image processing including processing for extracting data. That is, since the form determination unit can determine the form of sake rice from the image data (processed image data) in a state suitable for determining the form of sake rice, the possibility that the determination result is appropriate increases. ..
Therefore, it is possible to provide a sake rice analyzer capable of obtaining a determination result regarding the form of sake rice accurately in a short time.
In addition, machine learning is performed using image data (processed image data) in a state suitable for determining the form of sake rice as training data, and determination of the form of sake rice also determines the form of sake rice. It is performed on the image data (processed image data) in a state suitable for the processing. As a result, there is a high possibility that the determination result of the morphological determination unit is appropriate.
Further, the validity determination unit determines whether the morphology of the liquor rice determined by the morphology determination unit is high or low, and the morphology determination unit relates to the morphology of the liquor rice determined by the validity determination unit to be highly valid. Machine learning is executed by adding the combination of the information and the processed image data whose form has been determined to the existing training data. That is, it is possible to automatically accumulate appropriate training data and perform machine learning.
Further, the morphological storage unit stores the determination result of the morphological determination unit in a state where the change history of the morphology of one sake rice after the start of immersion in water can be known. As a result, it is possible to know what kind of change history one form of sake rice showed.
Furthermore, for example, when the form of one sake rice is continuously viewed, the form in which the cracks are formed, which is a state in which the cracks generated in the rice are enlarged, returns to the form in which the rice is not damaged. It never returns to the cracked form of sake rice. That is, if the morphology of sake rice after the start of immersion in water is appropriately determined by the morphological determination unit, the change history should conform to a predetermined change criterion. Therefore, in this characteristic configuration, in the validity determination unit, the certainty of the form of sake rice determined by the form determination unit is equal to or higher than a predetermined value, and the change in the form of sake rice after the start of immersion in water is started. When the history conforms to a predetermined change criterion, it is determined that the morphology of sake rice determined by the morphological determination unit is highly valid. As a result, appropriate training data can be accumulated and machine learning can be performed automatically.

The characteristic configuration of the sake rice analyzer according to the present invention for achieving the above object is an image data acquisition unit for acquiring image data obtained by photographing sake rice placed in a state of being immersed in water, and the image data acquisition unit. In the image data acquired by the company, an image processing unit that performs image processing including a process of extracting sake rice image data in a region containing only one sake rice and generates processed image data of the one sake rice. Using the execution result of machine learning of the accumulated training data, as the form of the one sake rice in the processed image data generated by the image processing unit, the rice is not damaged or the sake is not damaged. A morphological determination unit for determining whether the rice has a cracked form or a cracked form in which the gap between the cracks generated in the sake rice is large, and the determination of the morphological determination unit. The point is to include a morphological storage unit that stores the result in a state where the change history of the morphology of the one sake rice after the start of immersion in water can be known.

According to the above-mentioned feature configuration, the form of sake rice (the form in which the rice is not damaged or the form in which the rice is not damaged) is obtained from the image data obtained by the image data acquisition unit, which is obtained by photographing the rice soaked in water. It is possible to determine whether the rice has a cracked form or a cracked form in which the gap between the cracks generated in the sake rice is large. In particular, in this feature configuration, the morphological determination unit does not determine the morphology of sake rice, but the processed image data generated by the image processing unit using the machine learning execution result of the accumulated training data. Determine the form of one sake rice. As a result, it is possible to obtain a consistent determination result in a short time regarding the form of sake rice. In addition, the image data referred to by the morphological determination unit for determining the morphology of the liquor rice is a liquor rice image in a region containing only one liquor rice in the image data acquired by the image data acquisition unit by the image processing unit. It is processed image data after performing image processing including processing for extracting data. That is, since the form determination unit can determine the form of sake rice from the image data (processed image data) in a state suitable for determining the form of sake rice, the possibility that the determination result is appropriate increases. ..
Therefore, it is possible to provide a sake rice analyzer capable of obtaining a determination result regarding the form of sake rice accurately in a short time.
In addition, the morphological storage unit stores the determination result of the morphological determination unit in a state where the change history of the morphology of one sake rice after the start of immersion in water can be known. As a result, it is possible to know what kind of change history one form of sake rice has.

Another characteristic configuration of the sake rice analyzer according to the present invention is whether the morphological determination unit has, as the form of the crack, a vertical crack having a vertical cross section open along the long axis direction of the rice. Alternatively, the point is to determine whether or not the rice has a lateral crack with an open cross section along the minor axis direction.

According to the above-mentioned characteristic configuration, the morphological determination unit uses the image data (processed image data) in a state suitable for determining the morphology of the liquor rice as the morphology of the liquor rice, in which the liquor rice is not damaged. Or, it is a form in which cracks are generated in sake rice, or a form in which vertical cracks are generated in which the gaps between the cracks are large and the vertical cross section is opened along the long axis direction of sake rice. It is possible to determine whether or not the cracks have a large gap and the cross section of the rice is open along the short axis direction.

Yet another characteristic configuration of the liquor rice analyzer according to the present invention is the liquor rice in a region containing only one liquor rice in the image data acquired by the image data acquisition unit as the image processing performed by the image processing unit. It is a point including a process of extracting image data, a process of rotating a liquor rice image so that the one liquor rice has a predetermined posture, and a process of adjusting the contrast of the image.

According to the above feature configuration, the image data referred to by the morphological determination unit for determining the morphology of liquor rice is a region containing only one liquor rice in the image data acquired by the image data acquisition unit by the image processing unit. After performing image processing including a process of extracting the liquor rice image data, a process of rotating the liquor rice image so that one liquor rice has a predetermined posture, and a process of adjusting the contrast of the image. It is processed image data. That is, since the form determination unit can determine the form of sake rice from the image data (processed image data) in a state suitable for determining the form of sake rice, the possibility that the determination result is appropriate increases. ..

Another characteristic configuration of the liquor rice analyzer according to the present invention is that the morphological determination unit has information on the processed image data generated by the image processing unit and the morphology of liquor rice contained in the processed image data. The point is to execute the machine learning of the training data configured in combination with.

According to the above feature configuration, machine learning is performed using image data (processed image data) in a state suitable for determining the morphology of liquor rice as training data, and determination of the morphology of liquor rice is also performed. This is performed on the image data (processed image data) in a state suitable for determining the form of. As a result, there is a high possibility that the determination result of the morphological determination unit is appropriate.

Yet another characteristic configuration of the sake rice analyzer according to the present invention includes a validity determination unit for determining the validity level of the form of sake rice determined by the form determination unit, and the form determination unit is described. The machine learning is executed by adding a combination of information on the form of sake rice determined to be highly appropriate by the validity determination unit and the processed image data for which the form has been determined to the existing training data. There is a point to do.

According to the above characteristic configuration, the validity determination unit determines whether the morphology of the sake rice determined by the morphology determination unit is high or low, and the morphology determination unit determines that the validity is high by the validity determination unit. Machine learning is executed by adding a combination of information on the form of sake rice and the processed image data for which the form has been determined to the existing training data. That is, it is possible to automatically accumulate appropriate training data and perform machine learning.

It is a figure which shows the structure of the sake rice analyzer of 1st Embodiment. It is a flowchart explaining the form determination process. It is a figure explaining an example of image processing. It is a figure which shows the example of the form of sake rice. It is a flowchart explaining a machine learning process. It is a graph which shows the relationship between the number of learnings and the correct answer rate. It is a graph which shows the relationship between the number of learnings and the correct answer rate. It is a graph which shows the relationship between the number of learnings and the correct answer rate. It is a figure which shows the structure of the sake rice analyzer of 2nd Embodiment. It is a flowchart explaining a machine learning process. It is a flowchart explaining the validity determination process.

<First Embodiment>
The sake rice analyzer 10A (10) according to the first embodiment of the present invention will be described below with reference to the drawings.
FIG. 1 is a diagram showing the configuration of the sake rice analyzer 10A of the first embodiment. FIG. 2 is a flowchart illustrating a morphological determination process that is a part of the sake rice analysis method. FIG. 3 is a diagram illustrating an example of image processing. The sake rice analyzer 10A of the present embodiment includes an image data acquisition unit 11, an image processing unit 12, and a morphological determination unit 13. Further, the sake rice analyzer 10A includes a storage device 15 for storing information to be handled. The sake rice analyzer 10A is realized by using one or a plurality of computer devices having an information calculation processing function, an information input / output function, an information storage function, and the like. In that case, a program (liquor rice analysis program) that realizes the function of the image data acquisition unit 11, the function of the image processing unit 12, and the function of the morphological determination unit 13 on the computer device should be installed in the computer device. Just do it.

In the example described later, the morphological determination unit 13 predicts and determines the morphology of sake rice using an algorithm such as CNN (Convolutional Neural Network).
Further, the morphology determination unit 13 performs machine learning so that information regarding the morphology of sake rice is output when the processed image data generated by the image processing unit 12 is input. For example, machine learning adjusts a plurality of weight parameters of CNNs constituting the algorithm of the morphology determination unit 13.

The storage device 15 of the present embodiment determines the image data storage unit 15a for storing the image data captured by the imaging device 1, the training data storage unit 15b for storing the training data used in machine learning, and the morphology determination unit 13. It is provided with a morphological storage unit 15c for storing the result. Here, the morphological storage unit 15c may store the determination result of the morphological determination unit 13 in a state where the change history of the morphology of one sake rice after the start of immersion in water can be known.

The output device 2 is a display device capable of displaying image information, character information, and the like, and a printing device capable of printing the image information, character information, and the like on paper or the like.

FIG. 2 is a flowchart illustrating a form determination process for determining the form of sake rice with reference to image data.
In step # 10 of FIG. 2, the image data acquisition unit 11 acquires image data obtained by photographing sake rice arranged in a state of being immersed in water. That is, the image data acquisition step of acquiring the image data of the sake rice placed in the state of being immersed in water is executed. For example, the photographing apparatus 1 photographs a plurality of sake rice arranged in a soaked state to obtain one image data. Then, the image data is transmitted to the sake rice analyzer 10A and acquired by the image data acquisition unit 11. The image data acquired by the image data acquisition unit 11 can be stored in the image data storage unit 15a of the storage device 15. For example, the image data A shown in FIG. 3 is image data acquired by the image data acquisition unit 11.

In step # 11 of FIG. 2, the image processing unit 12 performs image processing including a process of extracting the sake rice image data of a region containing only one sake rice from the image data acquired by the image data acquisition unit 11. Generates processed image data of that one sake rice. That is, in the image data acquired in the image data acquisition step, image processing including a process of extracting the liquor rice image data of the region containing only one liquor rice is performed, and the processed image data of the one liquor rice is generated. The image processing step to be performed is executed. FIG. 3 is a diagram illustrating an example of image processing. In the present embodiment, as the image processing performed by the image processing unit 12, the processing for extracting the liquor rice image data in the region containing only one liquor rice from the image data acquired by the image data acquisition unit 11 and one of the processes. It includes a process of rotating the sake rice image so that the sake rice is in a predetermined posture, and a process of adjusting the contrast of the image.

For example, the image data B shown in FIG. 3 is an image obtained by extracting data including an entire image of one sake rice from the image data A acquired by the image data acquisition unit 11. The image data C shown in FIG. 3 is a process of extracting the sake rice image data of the region containing only one sake rice by removing the images of other sake rice other than one sake rice in the image data B. It is an image after going. The image data D shown in FIG. 3 is an image after processing to rotate the sake rice image so that one sake rice has a predetermined posture with respect to the image data C. In this case, the sake rice image is rotated so that the long axis of the sake rice is along the vertical direction. The image data E shown in FIG. 3 is an image after the image data D is subjected to a process of adjusting the orientation of sake rice (for example, a vertical inversion process, a horizontal inversion process, or the like). The image data F shown in FIG. 3 is an image after the image data E has been subjected to a process of adjusting the contrast of the image.

As described above, the image data as the input data referred to by the morphological determination unit 13 for determining the morphology of the liquor rice is one liquor rice in the image data acquired by the image processing unit 12 by the image data acquisition unit 11. Image processing including processing to extract liquor rice image data in an area containing only liquor, processing to rotate the liquor rice image so that one liquor rice has a predetermined posture, and processing to adjust the contrast of the image. It is the processed image data after it has been performed. That is, since the form determination unit 13 can determine the form of sake rice from the image data (processed image data) in a state suitable for determining the form of sake rice, the determination result may be appropriate. It will increase.

In step # 12 of FIG. 2, the morphology determination unit 13 is one of the processed image data (image data F of FIG. 3) generated by the image processing unit 12 using the machine learning execution result of the accumulated training data. As the form of sake rice, cracks occur in which the rice is not damaged, the rice is cracked, or the cracks in the rice have large gaps. It is determined whether the form is the same. That is, as one form of sake rice in the processed image data generated in the image processing step using the machine learning execution result of the accumulated training data, the rice is not damaged or the rice is not damaged. A morphological determination step of determining whether the rice has a cracked form or a cracked form in which the gap between the cracks generated in the sake rice is large is executed. As the form of the crack, the morphological determination unit 13 has a form in which a vertical crack having an open vertical cross section along the long axis direction of the sake rice has occurred, or a horizontal cross section having an open cross section along the minor axis direction of the sake rice has occurred. It is also determined whether the form is cracked.

The morphological determination unit 13 has an algorithm configured so that when the processed image data generated by the image processing unit 12 is input, information regarding the morphology of sake rice is output. For example, the CNN executed by the morphological determination unit 13 performs a convolution process on the input layer into which the processed image data generated by the image processing unit 12 is input and the processed image data input to the input layer. A hierarchical network is composed of a convolutional layer for obtaining a feature map, a pool layer for reducing the feature map output from the convolutional layer, a fully connected layer for connecting all units, and an output layer. The combination of the convolution layer and the pool layer is repeatedly provided a plurality of times, and a plurality of fully connected layers are also provided.

In this embodiment, four types of information (no damage, cracks, vertical cracks, horizontal cracks) are assumed as the information regarding the form of sake rice that is the determination result of the form determination unit 13, and therefore, the unit of the output layer of the CNN. The number will be four. In the output layer, the softmax function can be used as the likelihood function. Since the softmax function divides the value of each unit by the cumulative value of all the units of the output layer and normalizes it, the likelihood of each unit of the output layer takes a value between 0 and 1. Then, the morphology determination unit 13 determines the class having the maximum likelihood as a classification class, that is, the morphology (determination result) of sake rice. The determination result of the morphology determination unit 13 is stored in the morphology storage unit 15c. Here, the morphological storage unit 15c may store the determination result of the morphological determination unit 13 in a state where the change history of the morphology of one sake rice after the start of immersion in water can be known.

The determination result of the form determination unit 13 can also be output from the output device 2. For example, it is also possible to calculate the ratio of sake rice having vertical cracks or horizontal cracks among a plurality of sake rice and output it from the output device 2. Further, when the determination result of the morphology determination unit 13 is stored in the morphology storage unit 15c in a state where the change history of the morphology of one sake rice after the start of immersion in water can be known, vertical cracking occurs. It is also possible to calculate a change over time in the proportion of sake rice in which horizontal cracks occur and output it from the output device 2.

FIG. 4 is a diagram showing an example of the form of sake rice. Specifically, FIG. 4A is a diagram showing a form in which sake rice is not damaged. FIG. 4B shows a form in which the sake rice is cracked. FIGS. 4 (c) and 4 (d) show a form in which cracks are generated, which is a state in which the gaps between the cracks generated in sake rice are large. As the form of cracking, a horizontal crack having an open cross section along the short axis direction of sake rice (FIG. 4 (c)) and a vertical crack having an open vertical cross section along the long axis direction of sake rice occur. There is another form (FIG. 4 (d)). Further, FIG. 4B shows a form in which cracks are formed along the minor axis direction in sake rice. As described above, the morphological determination unit 13 has a form in which vertical cracks are formed in which the vertical cross section along the long axis direction of the sake rice is open, or the cross section along the minor axis direction of the sake rice. It is also determined that is a form in which an open lateral crack has occurred.

As described above, in the present embodiment, the morphological determination unit 13 does not determine the morphology of sake rice, but the image processing unit 12 generates the image processing unit 12 using the machine learning execution result of the accumulated training data. The morphology of one sake rice in the processed image data is determined. As a result, it is possible to obtain a consistent determination result in a short time regarding the form of sake rice.

FIG. 5 is a flowchart illustrating a machine learning process that is a part of the sake rice analysis method.
In step # 20, the image data acquisition unit 11 acquires image data of sake rice. That is, the image data acquisition step of acquiring the image data of the sake rice placed in the state of being immersed in water is executed. Then, in step # 21, the image processing unit 12 performs image processing to generate processed image data. Next, in step # 22, the morphology determination unit 13 trains training data composed of a combination of processed image data generated by the image processing unit 12 and information on the morphology of sake rice contained in the processed image data. It is stored in the data storage unit 15b, and machine learning of the training data is executed in step # 23. For example, one training data is created by labeling one processed image data with correct answer data determined by humans for one form of sake rice contained in the processed image data. Will be done. The training data storage unit 15b stores a plurality of training data created in this way.

In the present embodiment, the softmax function is used as the likelihood function in the output layer of the CNN executed by the morphology determination unit 13. Since the softmax function divides the value of each unit by the cumulative value of all the units of the output layer and normalizes it, the likelihood of each unit of the output layer takes a value between 0 and 1. Then, the morphological determination unit 13 uses an error back propagation method to reduce the error between the estimated data (“probability”) generated in the output layer and the correct answer data (“1”). , The plurality of weight parameters of the CNN to be executed are modified, and an algorithm for outputting information on the morphology of sake rice when the processed image data generated by the image processing unit 12 is input is obtained as the execution result of machine learning.

In this way, machine learning is performed using image data (processed image data) in a state suitable for determining the morphology of liquor rice as training data, and the above-mentioned determination of the morphology of liquor rice is also performed on liquor rice. This is performed on the image data (processed image data) in a state suitable for determining the morphology. As a result, there is a high possibility that the determination result of the morphology determination unit 13 is appropriate.

Next, the effect obtained by performing machine learning using the processed image data as in the present embodiment as training data will be described.
6 to 8 are graphs showing the relationship between the number of learnings and the correct answer rate. Specifically, FIG. 6 is a graph showing the relationship between the number of learnings and the correct answer rate when machine learning is performed using the training data in which the correct answer data is labeled in the image data B of FIG. FIG. 7 is a graph showing the relationship between the number of learnings and the correct answer rate when machine learning is performed using the image data D of FIG. 3 as training data. FIG. 8 is a graph showing the relationship between the number of learnings and the correct answer rate when machine learning is performed using the image data F of FIG. 3 as training data.

As shown in FIG. 6, when the image data B is used as training data, the correct answer rate is less than 80% even if the number of learnings reaches 20,000. This 80% correct answer rate corresponds to, for example, the correct answer rate when a person who is skilled to some extent determines the form of sake rice from image data. On the other hand, when machine learning is performed using the training data in which the correct answer data is labeled on the image data D, the number of learning times is about 12000 and the correct answer rate exceeds 80%. Further, when machine learning is performed using the training data in which the correct answer data is labeled in the image data F, the number of learning times is about 5000 times and the correct answer rate exceeds 80%. In this way, by using the processed image data after image processing, that is, the image data in a state suitable for determining the morphology of sake rice as training data, the correct answer rate even if the number of learnings is small. Was able to obtain the effect of rising.

<Second Embodiment>
The content of the machine learning process of the sake rice analyzer 10B (10) of the second embodiment is different from that of the above embodiment. The sake rice analyzer 10B of the second embodiment will be described below, but the description of the same configuration as that of the above embodiment will be omitted.

FIG. 9 is a diagram showing the configuration of the sake rice analyzer 10B of the second embodiment. As shown in the figure, the sake rice analyzer 10B of the second embodiment further includes a validity determination unit 14 for determining the validity of the form of sake rice determined by the morphology determination unit 13. Then, the morphology determination unit 13 adds a combination of information on the morphology of sake rice determined to be highly valid by the validity determination unit 14 and the processed image data for which the morphology has been determined to the existing training data. And perform machine learning.
Also in this embodiment, the sake rice analyzer 10B is realized by using one or a plurality of computer devices having an information calculation processing function, an information input / output function, an information storage function, and the like. In that case, a program (liquor rice analysis program) for realizing the functions of the image data acquisition unit 11, the image processing unit 12, the morphology determination unit 13, and the validity determination unit 14 on a computer device is provided. , You can install it on the computer device.

The morphological storage unit 15c stores the determination result of the morphological determination unit 13 in a state where the change history of the morphology of one sake rice after the start of immersion in water can be known. For example, the morphological storage unit 15c stores the determination results of the morphological determination unit 13 from the past to the present for each of the plurality of sake rice to be morphologically determined, in a state where the change history of the morphology can be known. is doing. As a result, it is possible to know what kind of change history one form of sake rice showed.

FIG. 10 is a flowchart illustrating the machine learning process of the second embodiment. This machine learning process is performed in addition to the machine learning process (FIG. 5) described in the first embodiment. More specifically, in step # 30, the validity determination unit 14 determines the validity of the current form of sake rice determined by the morphology determination unit 13. That is, the validity determination step of determining the validity of the form of sake rice determined in the form determination step executed by the form determination unit 13 is executed. FIG. 11 is a flowchart illustrating the validity determination process performed by the validity determination unit 14. In step # 40, the validity determination unit 14 determines whether or not the certainty of the current form of sake rice determined by the form determination unit 13 is equal to or higher than a predetermined value. For example, when the maximum likelihood calculated by the softmax function used in the output layer of the morphological determination unit 13 is the above-mentioned certainty, the maximum likelihood calculated by the softmax function is equal to or higher than a predetermined value (for example, 0). If it is (0.8 or more, etc.), the validity determination unit 14 determines that the certainty of the current form of sake rice is equal to or higher than a predetermined value, and proceeds to step # 41.

Next, in step # 41, the validity determination unit 14 determines whether or not the change history of the form of sake rice conforms to a predetermined standard. The predetermined criteria are the following six patterns. Then, if the change history of the form of sake rice conforms to any of the criteria of the six patterns, the validity determination unit 14 shifts to step # 42 and determines that the validity is high.

Specifically, for example, when the form of one sake rice is continuously viewed, the rice is not damaged from the cracked form in which the gap between the cracks generated in the rice is enlarged. It does not return to its morphology, nor does it return to the morphology of cracked rice. That is, if the morphology of sake rice after the start of immersion in water is appropriately determined by the morphological determination unit 13, the change history should conform to a predetermined change criterion. Therefore, in this feature configuration, the validity determination unit 14 has a certainty of the form of sake rice determined by the form determination unit 13 of a predetermined value or more, and the form of sake rice after the start of immersion in water. When the change history of No. 3 conforms to a predetermined change criterion, it is determined that the form of sake rice determined by the form determination unit 13 is highly valid. As a result, appropriate training data can be accumulated and machine learning can be performed automatically.

On the other hand, the validity determination unit 14 determines that the certainty is not equal to or higher than the predetermined value in the step # 40, and the change history of the liquor rice morphology does not conform to the predetermined criteria in the step # 41. If it is determined, the process proceeds to step # 43 and it is determined that the validity is low.

As described above, in the validity determination unit 14, the certainty of the form of the sake rice determined by the form determination unit 13 is equal to or higher than a predetermined value, and the form of the sake rice after the start of immersion in water is started. When the change history conforms to a predetermined change criterion, it is determined that the form of sake rice determined by the morphological determination unit 13 is highly valid, and the certainty of the form of sake rice determined by the morphological determination unit 13 is high. Is less than a predetermined value, or if the change history of the morphology of sake rice after starting soaking in water does not meet the predetermined change criteria, the sake rice determined by the morphological determination unit 13 It is judged that the validity of the form of is low.

Next, in step # 31, the combination of the information on the form of sake rice determined to be highly valid and the processed image data is used as training data. Then, in step # 32, the training data is added to the existing training data to execute machine learning.

In this way, the validity determination unit 14 determines whether the morphology of the sake rice determined by the morphology determination unit 13 is high or low, and the morphology determination unit 13 is determined by the validity determination unit 14 to have high validity. Machine learning is executed by adding a combination of information on the morphology of sake rice and processed image data for which the morphology has been determined to existing training data. That is, it is possible to automatically accumulate appropriate training data and perform machine learning.

<Another Embodiment>
<1>
In the above embodiment, the configuration of the sake rice analyzer 10 has been described with reference to specific examples, but the configuration can be changed as appropriate.

<2>
In the above embodiment, an example in which the morphological determination unit 13 predicts and determines the morphology of sake rice using a CNN (Convolutional Neural Network) has been described, but the algorithm constituting the morphological determination unit 13 is not limited to the above-mentioned CNN. , Support vector machine (SVM), k-nearest neighbor algorithm, discriminant function, etc. can be configured using various other algorithms.

<3>
The configurations disclosed in the above embodiments (including other embodiments) can be applied in combination with the configurations disclosed in other embodiments as long as there is no contradiction, and are disclosed in the present specification. The embodiment described is an example, and the embodiment of the present invention is not limited to this, and can be appropriately modified without departing from the object of the present invention.

INDUSTRIAL APPLICABILITY The present invention can be used for a sake rice analyzer capable of obtaining a determination result regarding the form of sake rice accurately in a short time.

1 Imaging device 2 Output device 10 (10A, 10B) Sake rice analyzer 11 Image data acquisition unit 12 Image processing unit 13 Form determination unit 14 Validity determination unit 15 Storage device 15a Image data storage unit 15b Training data storage unit 15c Form storage Department

Claims (6)

An image data acquisition unit that acquires image data of sake rice placed in a state of being immersed in water, and an image data acquisition unit.
In the image data acquired by the image data acquisition unit, image processing including extraction of liquor rice image data in a region containing only one liquor rice is performed to generate processed image data of the one liquor rice. Image processing unit and
Using the machine learning execution result of the training data composed of the accumulated combination of the processed image data generated by the image processing unit and the information on the morphology of sake rice contained in the processed image data , As the form of the one sake rice in the processed image data generated by the image processing unit, the rice is not damaged, the rice is cracked, or the rice is cracked. A morphological determination unit that determines whether or not a crack has occurred, which is a state in which the gap between the cracks generated in the rice is large.
The validity determination unit for determining the validity of the form of sake rice determined by the morphology determination unit, and the validity determination unit for determining the validity of the form.
It is provided with a morphological storage unit that stores the determination result of the morphological determination unit in a state where the change history of the morphology of the one sake rice after the start of immersion in water can be known.
The morphological determination unit adds a combination of information on the morphology of sake rice determined to be highly valid by the validity determination unit and the processed image data for which the morphology is determined to the existing training data. And execute the machine learning
The validity determination unit is
The certainty of the form of sake rice determined by the form determination unit is equal to or higher than a predetermined value, and the change history of the form of sake rice after the start of immersion in water conforms to the predetermined change criteria. If so, it is determined that the form of sake rice determined by the form determination unit is highly valid.
When the certainty of the form of sake rice determined by the form determination unit is less than the predetermined value, or the change history of the form of sake rice after the start of immersion in water is used as a predetermined change criterion. A sake rice analyzer that determines that the form of sake rice determined by the form determination unit is not valid if it is not compatible . An image data acquisition unit that acquires image data of sake rice placed in a state of being immersed in water, and an image data acquisition unit.
In the image data acquired by the image data acquisition unit, image processing including extraction of liquor rice image data in a region containing only one liquor rice is performed to generate processed image data of the one liquor rice. Image processing unit and
Using the machine learning execution result of the accumulated training data, as the form of the one sake rice in the processed image data generated by the image processing unit, the rice is not damaged or the sake is liquor. A morphological determination unit that determines whether the rice has a cracked form or a cracked form in which the gap between the cracks generated in the sake rice is large.
A sake rice analyzer comprising a morphology storage unit that stores the determination result of the morphology determination unit in a state in which the change history of the morphology of the one sake rice after the start of immersion in water is known. The form determination unit executes the machine learning of the training data composed of a combination of the processed image data generated by the image processing unit and information on the form of sake rice contained in the processed image data. The sake rice analyzer according to claim 2 . It is provided with a validity determination unit for determining the validity of the form of sake rice determined by the morphology determination unit.
The morphological determination unit adds a combination of information on the morphology of sake rice determined to be highly appropriate by the validity determination unit and the processed image data for which the morphology has been determined to the existing training data. The sake rice analyzer according to claim 3 , wherein the machine learning is executed. The morphological determination unit has, as the form of the crack, a form in which a vertical crack having a vertical cross section along the long axis direction of sake rice has occurred, or a cross section having a cross section along the minor axis direction of sake rice has opened. The sake rice analyzer according to any one of claims 1 to 4, which determines whether or not the form has a lateral crack. As the image processing performed by the image processing unit, a process of extracting liquor rice image data in a region containing only one liquor rice from the image data acquired by the image data acquisition unit, and the one liquor rice are predetermined. The liquor rice analyzer according to any one of claims 1 to 5, which comprises a process of rotating a liquor rice image so as to be in the posture of the above, and a process of adjusting the contrast of the image.

Images