WO2019026166A1

WO2019026166A1 - Work data classification system, work data classification method, and program

Info

Publication number: WO2019026166A1
Application number: PCT/JP2017/027787
Authority: WO
Inventors: 俊二菅谷
Original assignee: 株式会社オプティム
Priority date: 2017-07-31
Filing date: 2017-07-31
Publication date: 2019-02-07

Abstract

[Problem] To provide a work data classification system, a work data classification method, and a program capable of determining to which item of work newly acquired sensor data relates, and classifying the data accordingly. [Solution] This work data classification system is provided with: an acquisition module 211 which acquires a plurality of sets of sensor data relating to various items of work; an extraction module 212 which extracts feature quantities of the plurality of sets of sensor data; a storage module 231 which associates and stores the feature quantities with the items of work; a determination module 213 which determines which one of the stored feature quantities of the sets of sensor data is similar to a feature quantity of unknown sensor data; and a classification module 214 which, on the basis of the determination result, determines the item of work to which the unknown sensor data relates, and classifies the data accordingly.

Description

Work data classification system, work data classification method, and program

The present invention acquires sensor data regarding various operations such as image data from a camera, a log from a machine, a log of a smartphone terminal possessed by a user, etc., and the regularity and correlation of the series of operations. The present invention relates to a work data classification system, a work data classification method, and a program capable of classifying which work relating to unknown sensor data is by registering a work as a feature quantity and using it as a determination criterion.

Even in an environment where noise is present and can not be measured separately, a method has been proposed that allows each sensor data to be properly classified using only sensor data collected in time series from a large number of sensors. (Patent Document 1).

Japanese Patent Application Laid-Open No. 2016-099888

However, in the method of Patent Document 1, even under an environment where noise is present and can not be measured separately, each sensor data is appropriately classified using only sensor data collected in time series from many sensors. Although it is possible to do so, it is not possible to automatically classify what kind of operation data each sensor data is about.

For example, with regard to agricultural work, various sensor data are acquired from camera images, logs of agricultural machines, sensors of wearable terminals, etc., and it is automatic if the work being performed by the plurality of sensor data is rice planting work. There has never been a technique to classify by.

With respect to this problem, the present inventor extracts feature quantities of a plurality of sensor data and associates and stores a feature quantity and a specific work, thereby acquiring unknown sensor data, It was determined which feature amount the feature amount was similar to and stored, and it was possible to classify which work the unknown sensor data relates to based on the determination result.

The present invention acquires a plurality of sensor data relating to various tasks, determines a certain regularity and correlation of the series of tasks as feature quantities, and associates and stores the feature quantities and tasks. It is an object of the present invention to provide a work data classification system, a work data classification method, and a program capable of classifying which work the unknown sensor data relates to.

The present invention provides the following solutions.

The invention according to the first feature is
Acquisition means for acquiring multiple sensor data relating to various tasks;
Extracting means for extracting feature quantities of the plurality of sensor data;
Storage means for storing the feature quantity and the work in association with each other;
A determination unit that determines which feature of the stored sensor data is similar to the feature of the unknown sensor data;
Classification means for classifying which operation the unknown sensor data is about based on the result of the determination;
Provide a work data classification system comprising:

According to the invention as set forth in the first aspect, in the work data classification system, an acquisition means for acquiring a plurality of sensor data regarding various works, an extraction means for extracting a feature quantity of the plurality of sensor data, and the feature quantity Storage means for associating and storing the work and the operation, determination means for determining which feature amount of the stored sensor data is similar to the feature amount of the unknown sensor data, and the result of the determination, And categorizing means for categorizing which operation the unknown sensor data relates to.

The invention according to the first aspect is a category of a work data classification system, but the same action and effect can be obtained even with a work data classification method and a program.

An invention according to a second feature is the work data classification system according to the first feature, wherein
When there is a plurality of unknown sensor data, the classification unit classifies the plurality of feature data associated with the stored task into a plurality of tasks determined to be similar to the plurality of features. Provide a data classification system.

According to the invention as set forth in the second feature, in the work data classification system as the invention as set forth in the first feature, when there are a plurality of unknown sensor data, they are associated with the stored work in the classification means. The work is classified into a large number of tasks determined to be similar to a plurality of feature quantities.

An invention according to a third feature is the work data classification system according to the first feature or the second feature, wherein
The determination means is a product of an inner product of a feature of the unknown sensor data and a feature of the sensor data as a product of an absolute value of the feature of the unknown sensor data and an absolute value of the feature of the sensor data. And a work data classification system characterized by determining that they are similar when closer than a predetermined range.

According to the invention of the third feature, in the work data classification system according to the first feature or the second feature, the determination means is characterized by the feature amount of the unknown sensor data and the feature of the sensor data. It is determined that the inner product with the amount is similar to the value of the product of the absolute value of the feature amount of the unknown sensor data and the absolute value of the feature amount of the sensor data when it is closer than a predetermined range.

An invention according to a fourth feature is the work data classification system according to any of the first feature through the third feature, wherein
The operation is characterized in that the determination means performs machine learning of a feature amount in the past to determine which feature amount in the feature amount of the sensor data is similar to the feature amount of the unknown sensor data. Provide a data classification system.

According to the invention as set forth in the fourth feature, in the work data classification system according to any one of the first feature through the third feature, the determination means machine-learns the past feature amount to obtain the feature data. It is determined which feature of the feature of the sensor data is similar to the feature of the unknown sensor data.

The invention according to the fifth feature is
Acquiring a plurality of sensor data on various tasks;
Extracting features of the plurality of sensor data;
Storing the feature amount and the work in association with each other;
Determining which feature of unknown sensor data is similar to the feature of the stored sensor data;
Classifying, based on the result of the determination, which operation the unknown sensor data relates to;
Provide a work data classification method comprising:

The invention according to the eighth feature is
In the work data classification system,
Acquiring multiple sensor data for various tasks,
Extracting features of the plurality of sensor data;
Storing the feature quantity and the work in association with each other;
Determining which feature of unknown sensor data is similar to the feature of the stored sensor data;
Classifying which operation the unknown sensor data relates to based on the result of the determination;
Provide a program to run the program.

According to the present invention, a plurality of sensor data relating to various tasks are acquired, the regularity and correlation of the series of tasks are judged as feature quantities, and the feature quantities and tasks are associated and stored. This makes it possible to provide a work data classification system, a work data classification method, and a program that can classify which unknown sensor data relates to which work.

FIG. 1 is a schematic diagram of a preferred embodiment of the present invention. FIG. 2 is a diagram showing the relationship between the functional blocks of the device 100 and the computer 200 and the respective functions. FIG. 3 is a flowchart of work data classification processing of acquiring unknown sensor data from the apparatus 100, performing extraction processing and determination processing of feature amounts by the computer 200, and classifying unknown sensor data. FIG. 4 is a flowchart of the case where unknown sensor data is classified by the computer 200 according to the number of determinations of the unknown sensor data. FIG. 5 is an example of a flowchart for determining whether the feature amount of unknown sensor data and the feature amount of stored sensor data are similar by the computer 200. FIG. 6 is a flowchart of the case where machine learning is performed on a combination of past work and sensor data feature amounts to determine which stored sensor data feature amounts of unknown sensor data feature amounts are similar to each other. is there. FIG. 7 is an example of a table showing data structures of work and sensor data stored in the storage unit 230. FIG. 8 is an example of a table showing the data structure of unknown sensor data. FIG. 9 is an example of a table including determination results of stored sensor data having feature amounts similar to unknown sensor data.

Hereinafter, the best mode for carrying out the present invention will be described with reference to the drawings. This is merely an example, and the technical scope of the present invention is not limited to this.

[Overview of work data classification system]
FIG. 1 is a schematic diagram of a preferred embodiment of the present invention. The outline of the present invention will be described based on FIG. The work data classification system comprises an apparatus 100, a computer 200, and a communication network 300.

In FIG. 1, the number of devices 100 is not limited to one, and may be plural. Here, a WEB camera is illustrated as the device 100A, and a wearable device is illustrated as an example of the device 100B. The computer 200 is not limited to an existing device, and may be a virtual device. Also, the computer 200 may be the same device as the device 100.

As shown in FIG. 2, the device 100 includes a sensor unit 10, a control unit 110, a communication unit 120, and a storage unit 130. The computer 200 also includes a control unit 210, a communication unit 220, a storage unit 230, and an input / output unit 240, as also shown in FIG. The control unit 210 cooperates with the communication unit 220 and the storage unit 230 to realize the acquisition module 211. Further, the control unit 210 cooperates with the storage unit 230 to implement an extraction module 212, a determination module 213, and a classification module 214. The storage unit 230 implements the storage module 231 in cooperation with the control unit 210. The communication network 300 may be a public communication network such as the Internet or a dedicated communication network, and enables communication between the device 100 and the computer 200.

The device 100 includes a sensor unit 10 capable of acquiring an image, a machine log, and data of various sensors, and is a device capable of data communication with the computer 200. Here, as an example, a WEB camera is illustrated as the device 100A, and a wearable device is illustrated as the device 100B, but a digital camera, digital video, security camera, in-vehicle camera, 360 degree camera, industrial device, agricultural device, drone , A wearable device, or the like. Further, various sensor data may be stored in the storage unit 130.

The computer 200 is a computing device capable of data communication with the device 100. Here, a desktop computer is illustrated as an example, but in addition to a mobile phone, a portable information terminal, a tablet terminal, a personal computer, electric appliances such as a netbook terminal, a slate terminal, an electronic book terminal, a portable music player, etc. And wearable terminals such as smart glasses and head mounted displays.

In the work data classification system of FIG. 1, first, the storage module 231 of the computer 200 stores the work in which the feature quantities of a plurality of sensor data are associated with each other in the storage unit 230 (step S01). The association between the work and the feature quantities of a plurality of sensor data may be acquired from another computer or storage medium, or may be created by the computer 200. In addition, the storage unit 230 may be provided with a dedicated database.

FIG. 7 is an example of a table showing data structures of work and sensor data stored in the storage unit 230. Here, two sensor data are stored as work of rice planting 1, of which the feature of sensor A is v, the type of data is a moving image, the feature of sensor B is w, and the type of data is As that of the acceleration sensor. In addition, three sensor data are stored as a work of rice harvesting 1. Among them, the feature amount of sensor C is x, the type of data is moving image, the feature amount of sensor D is y, and the type of data is Indicates the machine log, the feature amount of the sensor E is z, and the data type is GPS. The data itself of each sensor may be stored together with these data. FIG. 7 shows an example in which the storage destination of the data itself of each sensor is described in the rightmost column of the table.

Returning to FIG. 1, the device 100 transmits unknown sensor data to the computer 200 (step S02), and the acquisition module 211 of the computer 200 acquires unknown sensor data (step S03). When there are a plurality of unknown sensor data, it is assumed that all are acquired as sensor data regarding one work. Here, although the flow of transmitting unknown sensor data from the device 100 is described, the acquisition module 211 instructs the device 100 to transmit unknown sensor data, and the device 100 receives the transmission of unknown sensor data in response thereto. You may go. The acquisition module 211 not only acquires sensor data acquired in real time by the device 100, but also acquires sensor data acquired by the device 100 in the past and stored in the storage unit 130. Good.

FIG. 8 is an example of a table showing the data structure of unknown sensor data. Here, the work consists of three sensor data, of which the feature of sensor F is s, the type of data is GPS, the feature of sensor G is t, the type of data is machine log, sensor H The feature amount of is u, which indicates that the type of data is a moving image. The data itself of each sensor may be stored together with these data. FIG. 8 shows an example in which the storage destination of the data itself of each sensor is described in the rightmost column of the table.

Referring back to FIG. 1 again, the extraction module 212 of the computer 200 extracts the feature amount of the unknown sensor data acquired in step S03 (step S04). Here, the feature amount may be, for example, according to the type of data, or may be analyzed according to the content of the data, and may be according to the feature amount. May be used as For example, if the data type is a moving image, it may be the feature that has been subjected to image analysis of the moving image, and if the data type is an acceleration sensor, it may be that the motion has been analyzed as the feature, For example, what analyzed the operation and time of the machine may be used as the feature amount, and in the case of GPS data, the location, height and weather may be used as the feature amount with reference to map data and weather data.

Next, the determination module 213 of the computer 200 determines which sensor data feature amount stored in the storage unit 230 is similar to the feature amount of unknown sensor data (step S05). Details of the determination method will be described later.

FIG. 9 is an example of a table including determination results of stored sensor data having feature amounts similar to unknown sensor data. Here, the feature of the sensor F is s and is similar to the sensor E, the feature of the sensor G is t and is similar to the sensor D, the feature of the sensor H is u and is similar to the sensor A Is shown as an example.

Finally, the classification module 214 of the computer 200 classifies which operation the acquired unknown sensor data relates to (step S06). Here, from FIG. 7 and FIG. 9, the sensor F is similar to the sensor E of the rice harvesting 1, the sensor G is similar to the sensor D of the rice harvesting 1, and the sensor H is the sensor A of the rice planting 1 It is understood that it is similar to As a classification method, for example, in the case of prioritizing the number of determinations of sensor data, sensor F, sensor G, sensor H are based on the result that two of the three sensors are similar to rice harvesting and one is similar to rice planting. The work by can be classified as rice harvesting. Alternatively, if the type of data is prioritized and classified, for example, if the setting is to prioritize the classification result by the moving image, the sensor H emphasizes the determination result that the work is similar to the sensor A of rice planting 1, and the sensor Operations by F, sensor G, and sensor H can be classified as rice planting. Here, the case is described where it is determined that there is stored sensor data having a similar feature amount to unknown sensor data, but if it is determined that there is no similar stored sensor data, the classification module 214 It may be classified as unclassifiable or unclassified.

As described above, according to the present invention, a plurality of sensor data relating to various tasks are acquired, and the regularity and correlation possessed by the series of tasks are determined as feature quantities, and the feature quantities are associated with the tasks. By storing, it is possible to provide a work data classification system, a work data classification method, and a program that can classify which unknown sensor data relates to which work.

[Description of each function]
FIG. 2 is a diagram showing the relationship between the functional blocks of the device 100 and the computer 200 and the respective functions. The apparatus 100 includes a sensor unit 10, a control unit 110, a communication unit 120, and a storage unit 130. The computer 200 also includes a control unit 210, a communication unit 220, a storage unit 230, and an input / output unit 240. The control unit 210 cooperates with the communication unit 220 and the storage unit 230 to realize the acquisition module 211. Further, the control unit 210 cooperates with the storage unit 230 to implement an extraction module 212, a determination module 213, and a classification module 214. The storage unit 230 implements the storage module 231 in cooperation with the control unit 210. The communication network 300 may be a public communication network such as the Internet or a dedicated communication network, and enables communication between the device 100 and the computer 200.

The apparatus 100 includes, as the sensor unit 10, a sensor capable of acquiring an image, a machine log, and data of various sensors. Also, it is assumed that the obtained data has the accuracy necessary to extract the feature amount.

The control unit 110 includes a central processing unit (CPU), a random access memory (RAM), a read only memory (ROM), and the like.

A device for enabling communication with other devices as the communication unit 120, for example, an IMT-2000 standard such as a WiFi (Wireless Fidelity) compliant device compliant with IEEE 802.11 or a third generation or fourth generation mobile communication system A compliant wireless device is provided. It may be a wired LAN connection.

The storage unit 130 includes a storage unit of data using a hard disk or a semiconductor memory, and stores captured images, necessary data such as imaging conditions, and the like.

The computer 200 is a computing device capable of data communication with the device 100. Here, a desktop computer is illustrated as an example, but in addition to a mobile phone, a portable information terminal, a tablet terminal, a personal computer, electric appliances such as a netbook terminal, a slate terminal, an electronic book terminal, a portable music player, etc. And wearable terminals such as smart glasses and head mounted displays. The computer 200 is not limited to an existing device, and may be a virtual device. Also, the computer 200 may be the same device as the device 100.

The control unit 210 includes a CPU, a RAM, a ROM, and the like. The control unit 210 cooperates with the communication unit 220 and the storage unit 230 to realize the acquisition module 211. Further, the control unit 210 cooperates with the storage unit 230 to implement an extraction module 212, a determination module 213, and a classification module 214.

A device for enabling communication with other devices as the communication unit 220, for example, a wireless device compliant with IEEE 802.11 or a wireless device compliant with IMT-2000 such as a third generation or fourth generation mobile communication system Etc. It may be a wired LAN connection.

The storage unit 230 includes a storage unit of data using a hard disk or a semiconductor memory, and associates work with feature amounts of a plurality of sensor data, teacher data, data required for processing such as determination results, classification results, etc. Remember. The storage unit 230 implements the storage module 231 in cooperation with the control unit 210. In addition, the storage unit 230 may include a database for storing the work and the feature quantities of the plurality of sensor data associated with each other.

The input / output unit 240 is provided with the functions necessary to use the work data classification system. As an example for realizing the input, it is possible to provide a liquid crystal display for realizing a touch panel function, a keyboard, a mouse, a pen tablet, hardware buttons on the device, a microphone for performing voice recognition, and the like. Further, as an example for realizing the output, a form such as a liquid crystal display, a display of a PC, a display such as a projection on a projector, and an audio output can be considered. The present invention is not particularly limited in function by the input / output method.

[Work data classification process]
FIG. 3 is a flowchart of work data classification processing of acquiring unknown sensor data from the apparatus 100, performing extraction processing and determination processing of feature amounts by the computer 200, and classifying unknown sensor data. The processing executed by each module described above will be described along with this processing.

First, the storage module 231 of the computer 200 stores, in the storage unit 230, the work and the feature quantities of a plurality of sensor data associated with each other (step S301). The association between the work and the feature quantities of a plurality of sensor data may be acquired from another computer or storage medium, or may be created by the computer 200. In addition, the storage unit 230 may be provided with a dedicated database. The process of step S301 is skipped if there is already a stored work that associates the feature quantities of a plurality of sensor data with a work, and there is no one that associates the feature quantity of the sensor data with the work to be newly stored. It shall be good.

Returning to FIG. 3, the acquisition module 211 of the computer 200 requests the device 100 to transmit sensor data (step 302). In the case where there are a plurality of devices 100, transmission of sensor data is requested to all the devices.

In response to the sensor data transmission request from the computer 200, the control unit 110 of the device 100 stores the sensor data in the storage unit 130 (step S303).

Then, the device 100 transmits unknown sensor data to the computer 200 via the communication unit 120 (step S304).

The acquisition module 211 of the computer 200 acquires unknown sensor data (step S305). When there are a plurality of apparatuses 100, that is, when there are a plurality of unknown sensor data, it is assumed that all are acquired as sensor data regarding one operation. The acquisition module 211 may acquire not only sensor data acquired in real time by the device 100 but also sensor data acquired by the device 100 in the past and stored in the storage unit 130.

Referring back to FIG. 3 again, the extraction module 212 of the computer 200 extracts the feature amount of the unknown sensor data acquired in step S305 (step S306). Here, the feature amount may be, for example, according to the type of data, or may be analyzed according to the content of the data, and may be according to the feature amount. May be used as For example, if the data type is a moving image, it may be the feature that has been subjected to image analysis of the moving image, and if the data type is an acceleration sensor, it may be that the motion has been analyzed as the feature, For example, what analyzed the operation and time of the machine may be used as the feature amount, and in the case of GPS data, the location, height and weather may be used as the feature amount with reference to map data and weather data.

Next, the determination module 213 of the computer 200 determines which sensor data feature amount stored in the storage unit 230 is similar to the feature amount of unknown sensor data (step S307).

Finally, the classification module 214 of the computer 200 classifies which operation the acquired unknown sensor data relates to (step S308). Here, from FIG. 7 and FIG. 9, the sensor F is similar to the sensor E of the rice harvesting 1, the sensor G is similar to the sensor D of the rice harvesting 1, and the sensor H is the sensor A of the rice planting 1 It is understood that it is similar to As a classification method, for example, in the case of prioritizing the number of determinations of sensor data, sensor F, sensor G, sensor H are based on the result that two of the three sensors are similar to rice harvesting and one is similar to rice planting. The work by can be classified as rice harvesting. Alternatively, if the type of data is prioritized and classified, for example, if the setting is to prioritize the classification result by the moving image, the sensor H emphasizes the determination result that the work is similar to the sensor A of rice planting 1, and the sensor Operations by F, sensor G, and sensor H can be classified as rice planting. Here, the case is described where it is determined that there is stored sensor data having a similar feature amount to unknown sensor data, but if it is determined that there is no similar stored sensor data, the classification module 214 It may be classified as unclassifiable or unclassified.

[Work data classification processing according to the number of judgments of unknown sensor data]
FIG. 4 is a flowchart of the case where unknown sensor data is classified by the computer 200 according to the number of determinations of the unknown sensor data. The configuration is assumed to have the same configuration as the device 100 and the computer 200 of FIG. The processing corresponds to steps S307 and S308 in the flowchart of FIG. In the following, work data classification processing according to the number of determinations of unknown sensor data will be described as processing after the flow up to step S306 in FIG. 3. Here, for the sake of explanation, it is assumed that the data examples of the above-mentioned FIG. 7, FIG. 8 and FIG. 9 are used.

Regarding the unknown sensor data acquired in step S305, feature quantities of the respective sensor data have been extracted in step S306. Here, the number of acquired unknown sensor data is counted (step S401). In the example of FIG. 8 described above, the number of unknown sensor data is three.

Next, one unknown sensor data is selected (step S402). Here, it is assumed that the sensor F in FIG. 8 is selected.

The determination module 213 of the computer 200 determines which sensor data feature amount stored in the storage unit 230 is similar to the feature amount of the unknown sensor data (step S403). Here, as shown in FIG. 9, it is determined that the feature amount of the sensor F is similar to the sensor E.

Next, the determination module 213 confirms whether the determination of all unknown sensor data has been completed (step S404).

At this point in time, the process has not ended, so the process returns to step S402 to select one unknown sensor data. Here, it is assumed that the sensor G of FIG. 8 is selected.

Next, in step S403, the determination module 213 determines that the feature amount of the sensor G is similar to the sensor D as illustrated in FIG.

In step S404, the determination module 213 confirms again whether or not determination of all unknown sensor data has been completed.

Since determination of all sensor data has not been completed yet at this point either, the process returns to step S402 to select one unknown sensor data. Here, it is assumed that the sensor H of FIG. 8 is selected.

Next, in step S403, the determination module 213 determines that the feature amount of the sensor H is similar to that of the sensor A as illustrated in FIG.

In step S404, the determination module 213 confirms whether the determination of all unknown sensor data is completed.

Here, since the determination of all the sensor data is completed, the process proceeds to step S405, and the unknown sensor data is classified into a work having a large number determined to be similar to the feature amount of the associated sensor data. Here, the sensor F is similar to the sensor E of the rice harvesting 1, the sensor G is similar to the sensor D of the rice harvesting 1, and the sensor H is similar to the sensor A of the rice planting 1. From the above, of the three unknown sensor data, the number determined to be similar to the feature amount of the sensor data of rice harvesting 1 is two, and the number determined to be similar to the feature amount of the sensor data of rice planting 1 is It is one. Therefore, the classification module 214 classifies the sensor F, the sensor G, and the sensor H as work data related to rice planting.

As described above, according to the present invention, a plurality of sensor data relating to various tasks are acquired, and the regularity and correlation possessed by the series of tasks are determined as feature quantities, and the feature quantities are associated with the tasks. A work data classification system capable of appropriately classifying which work the unknown sensor data relates to by storing work data classification processing according to the number of judgments of unknown sensor data It becomes possible to provide a data classification method and program.

[Feature amount determination process]
FIG. 5 is an example of a flowchart for determining whether the feature amount of unknown sensor data and the feature amount of stored sensor data are similar by the computer 200. The configuration is assumed to have the same configuration as the device 100 and the computer 200 of FIG. In addition, the process corresponds to step S307 in the flowchart of FIG. In the following, feature amount determination processing will be described as processing after the flow up to step S306 in FIG. 3. Here, for the sake of explanation, it is assumed that the data examples of the above-mentioned FIG. 7, FIG. 8 and FIG. 9 are used.

Regarding the unknown sensor data acquired in step S305, feature quantities of the respective sensor data have been extracted in step S306. Here, one unknown sensor data is selected (step S501). Here, it is assumed that the sensor F in FIG. 8 is selected.

Next, the determination module 213 of the computer 200 selects stored sensor data to be compared (step S502). Here, it is assumed that the sensor A in FIG. 7 is selected.

The determination module 213 obtains an inner product of the feature amount s of the sensor F which is unknown sensor data and the feature amount v of the sensor A which is stored sensor data to be compared (step S503).

Further, the determination module 213 obtains the product of the absolute value of the feature amount s of the sensor F, which is unknown sensor data, and the absolute value of the feature amount v of the sensor A, which is stored sensor data to be compared (step S504).

The determination module 213 obtains a difference between the inner product obtained in step S503 and the product obtained in step S504 (step S505).

Then, when the difference obtained in step S506 is smaller than the predetermined range, the determination module 213 determines that the sensor F resembles the sensor A (step S506), and the difference is equal to or larger than the predetermined range. It is determined that the sensor F is not similar to the sensor A (step S507). Here, it is determined that the sensor F is not similar to the sensor A.

Next, the determination module 213 confirms whether or not the determination of all stored sensor data is completed (step S508), and when the determination is not completed, the process returns to step S502 to continue the process, and the determination is If it has ended, the process proceeds to the next step S509. That is, when all the determinations as to whether the sensor F which is unknown sensor data is similar to the sensor A, the sensor B, the sensor C, the sensor D and the sensor E which are stored sensor data are all finished, The process proceeds to step S509. Here, it is assumed that the sensor F is determined to be similar to the sensor E only.

Finally, the determination module 213 confirms whether the determination of all unknown sensor data has been completed (step S509). If the determination is not completed, the process returns to step S501 to continue the processing, and the determination is completed. If yes, the feature amount determination process ends. That is, when the determination of which stored sensor data the sensor F, the sensor G, and the sensor H, which are unknown sensor data, are similar to each other is completed, the process is ended. Here, it is determined that the sensor F is similar to the sensor E only, the sensor G is similar to the sensor D only, and the sensor H is only similar to the sensor A.

After the above-described feature amount determination processing, the process returns to step S308 in the flowchart of FIG. 3, and the classification module 214 of the computer 200 classifies to which work the acquired unknown sensor data relates. In step S308, it is determined that the sensor F is similar to the sensor E of the rice harvesting 1, the sensor G is similar to the sensor D of the rice harvesting 1, and the sensor H is similar to the sensor A of the rice planting 1. Perform classification based on the result. As an example of the classification method, as described above with reference to FIG. 4, the sensor F is prioritized based on the result that priority is given to the determination number of sensor data and two out of three sensors are similar to rice harvesting and one is similar to rice planting. Work classified by G and sensor H is classified as rice harvesting. As another classification method, in the example of prioritizing the type of data, for example, if the setting is to prioritize the classification result by the moving image, for example, it is determined that the sensor H resembles the sensor A of rice planting 1 It is also conceivable to classify the work by the sensor F, the sensor G and the sensor H as rice planting, with emphasis on Here, the case has been described where it is determined that the sensor F is similar to the sensor E only, the sensor G is similar to the sensor D only, and the sensor H is only similar to the sensor A. If there are a plurality of similar stored sensor data, it may be possible to hold a plurality of centers as similar sensors, or it is determined in step S505 when the determination of all stored sensor data is completed in step S508. The one with the smallest difference between the inner product and the product may be held as a similar sensor. Further, if it is determined that there is no stored sensor data similar to the unknown sensor data, the classification module 214 may set the classification as unclassifiable or unclassified.

When the feature quantity vector of unknown sensor data is a and the feature quantity vector b of stored sensor data is a feature quantity determination method of the flow chart of FIG. 5, an inner product “absolute value a × absolute value b × The cosine θ ′ utilizes that it is considered that the orientations of the feature quantity vector a and the feature quantity vector b are more similar as θ is smaller. That is, when the angle θ approaches 0, the value of the cosine θ approaches 1, so the value of the inner product such as absolute value a × absolute value b × cosine θ is also absolute value a × absolute value b × 1 = absolute value a × It approaches the absolute value b. That is, when the difference between “absolute value a × absolute value b × cosine θ” and “absolute value a × absolute value b” is within a predetermined range, ie, a value closer to 0, the feature quantity vector a And the feature quantity vector b can be said to be similar. However, this is merely an example of the feature amount determination method, and the determination method in step S308 is not limited to this method.

As described above, according to the present invention, a plurality of sensor data relating to various tasks are acquired, and the regularity and correlation possessed by the series of tasks are determined as feature quantities, and the feature quantities are associated with the tasks. By storing, it is determined which stored sensor feature quantity the feature quantity of unknown sensor data is similar to, and based on that, it is classified which work related to the unknown sensor data is data about It is possible to provide a work data classification system, a work data classification method, and a program that can be performed.

[Feature amount determination processing using machine learning]
FIG. 6 is a flowchart of the case where machine learning is performed on a combination of past work and sensor data feature amounts to determine which stored sensor data feature amounts of unknown sensor data feature amounts are similar to each other. is there. The configuration is assumed to have the same configuration as the device 100 and the computer 200 of FIG.

First, the storage module 231 of the computer 200 stores combinations of work and feature quantities of a plurality of sensor data in the storage unit 230 as teacher data (step S601). The combination of the work and the feature quantities of a plurality of sensor data may be acquired from another computer or storage medium, or may be created by the computer 200. Alternatively, an operation classified as a feature of unknown sensor data classified in the past may be used as teacher data. Further, the storage unit 230 may be provided with a database dedicated to teacher data.

Next, the determination module 213 of the computer 200 performs machine learning of the determination method using the teacher data (step S602). As a method of machine learning here, it is assumed to use supervised learning (Supervised Learning). Based on a large number of teacher data stored in the storage unit 230 by the storage module 231, the machine determines which feature quantity of sensor data it is determined to be similar to the feature quantity of sensor data of which operation learn. The processes in steps S601 and S602 may be skipped if machine learning of the determination method is unnecessary. In addition, it is desirable to use the flow of FIG. 3 because the accuracy is expected to be low while the number of teacher data is small. The advantage of machine-learning the judgment module 213 by supervised learning is that the judgment accuracy is further improved without using human hands by using work classified as feature amounts of unknown sensor data classified in the past as teacher data. It is a point that can be improved. However, since it is assumed that it takes time for machine learning, the process of step S602 may be performed in a time zone or the like where the load on the work data classification system is small.

The subsequent processes of step S603 to step S609 correspond to the processes of step S302 to step S308 of FIG. 3 and thus the description thereof is omitted here.

As described above, according to the present invention, the combination of the feature amount and the operation is used as teacher data, and the supervised learning is performed by the determination module 213, whereby any feature value of the unknown sensor data is stored. It is possible to improve the accuracy with which it is determined to be similar to the feature amount, thereby classifying the operation relating to which unknown sensor data is data, and capable of causing classification accuracy to be negotiated. It becomes possible to provide a system, a work data classification method, and a program.

The above-described means and functions are realized by a computer (including a CPU, an information processing device, and various terminals) reading and executing a predetermined program. The program may be provided, for example, from a computer via a network (SaaS: software as a service), a flexible disk, a CD (CD-ROM, etc.), a DVD (DVD-ROM, DVD) Provided in the form of being recorded in a computer readable recording medium such as a RAM, a compact memory, etc. In this case, the computer reads the program from the recording medium, transfers the program to an internal storage device or an external storage device, stores it, and executes it. Alternatively, the program may be recorded in advance in a storage device (recording medium) such as, for example, a magnetic disk, an optical disk, or a magneto-optical disk, and may be provided from the storage device to the computer via a communication line.

As mentioned above, although embodiment of this invention was described, this invention is not limited to these embodiment mentioned above. Further, the effects described in the embodiments of the present invention only list the most preferable effects resulting from the present invention, and the effects according to the present invention are limited to those described in the embodiments of the present invention is not.

100 devices, 200 computers, 300 communication networks

Claims

Acquisition means for acquiring multiple sensor data relating to various tasks;
Extracting means for extracting feature quantities of the plurality of sensor data;
Storage means for storing the feature quantity and the work in association with each other;
A determination unit that determines which feature of the stored sensor data is similar to the feature of the unknown sensor data;
Classification means for classifying which operation the unknown sensor data is about based on the result of the determination;
Work data classification system comprising:
In the case where there are a plurality of unknown sensor data, the classification means is classified into a work having a large number determined to be similar to a plurality of feature quantities associated with the stored work. The work data classification system according to Item 1.
The determination means is a product of an inner product of a feature of the unknown sensor data and a feature of the sensor data as a product of an absolute value of the feature of the unknown sensor data and an absolute value of the feature of the sensor data. The work data classification system according to claim 1 or 2, wherein it is determined that they are similar when they are closer than a predetermined range.
The determination means is characterized by performing machine learning of a feature amount in the past to determine which feature amount in the feature amount of the sensor data is similar to the feature amount of the unknown sensor data. The work data classification system according to any one of claims 1 to 3.
Acquiring a plurality of sensor data on various tasks;
Extracting features of the plurality of sensor data;
Storing the feature amount and the work in association with each other;
Determining which feature of unknown sensor data is similar to the feature of the stored sensor data;
Classifying, based on the result of the determination, which operation the unknown sensor data relates to;
Work data classification method comprising:
In the work data classification system,
Acquiring multiple sensor data for various tasks,
Extracting features of the plurality of sensor data;
Storing the feature quantity and the work in association with each other;
Determining which feature of unknown sensor data is similar to the feature of the stored sensor data;
Classifying which operation the unknown sensor data relates to based on the result of the determination;
A program to run a program.