WO2021153646A1 - Model generation method, model generation device, abnormality detection method, abnormality detection device, and program - Google Patents

Abstract

Provided is a technique for detecting an abnormality in detection target data with high accuracy. One aspect of the present disclosure relates to a model generation method comprising: a step in which one or more processors input detection target data into an abnormality detection model and acquire an abnormality degree map from the abnormality detection model; a step in which the one or more processors acquire a region label containing a pair of region data indicating the region in the detection target data and a label value indicating whether the region is normal or abnormal; a step in which the one or more processors acquire an abnormality degree score from the abnormality degree map and the region data; a step in which the one or more processors acquire a loss value of the detection target data from the abnormality degree score and the label value; and a step in which the one or more processors update parameters of the abnormality detection model on the basis of the loss value.

Description

Model generation method, model generation device, abnormality detection method, abnormality detection device and program

This disclosure relates to a model generation method, a model generation device, an abnormality detection method, an abnormality detection device, and a program.

The anomaly detection problem that identifies whether a given data contains anomalies by machine learning can be widely applied in industrial applications. Examples include visual inspection, medical image diagnosis, abnormality detection of surveillance cameras, remote sensing, and automatic driving.

The object of the present disclosure is to provide a technique for detecting an abnormality in the data to be detected with high accuracy.

In order to solve the above problems, one aspect of the present disclosure is
A step in which one or more processors input detection target data into an abnormality detection model and acquire an abnormality degree map from the abnormality detection model.
A step in which the one or more processors obtain an area label including a pair of area data indicating an area in the detection target data and a label value indicating whether the area is normal or abnormal.
A step in which the one or more processors obtains an anomaly score from the anomaly map and the region data.
A step in which the one or more processors obtain a loss value of the detection target data from the abnormality score and the label value.
A step in which the one or more processors update the parameters of the anomaly detection model based on the loss value.
The present invention relates to a model generation method having.

It is the schematic which shows the abnormality detection processing by one Example of this disclosure. It is a block diagram which shows the functional structure of the model generation apparatus by one Example of this disclosure. It is a flowchart which shows the model generation process by one Example of this disclosure. It is a block diagram which shows the functional structure of the abnormality detection apparatus by one Example of this disclosure. It is a flowchart which shows the abnormality detection processing by one Example of this disclosure. It is a block diagram which shows the hardware configuration of the model generation apparatus and abnormality detection apparatus by one Example of this disclosure.

Hereinafter, embodiments of the present disclosure will be described with reference to the drawings.
[Summary of the present disclosure]
In the following examples, a model generation device that generates an abnormality detection model and an abnormality detection device that uses the generated abnormality detection model are disclosed.

As shown in FIG. 1, the model generator 100 according to an embodiment of the present disclosure is trained according to the training process described in detail below using the training labeled sample data stored in the database 300. You may train the anomaly detection model of the target. Here, the anomaly detection model is realized by, for example, a convolutional neural network, receives sample data such as image data as an input, and outputs an anomaly map (anomaly map) indicating an abnormality location in the image data. When the training of the abnormality detection model is completed, the model generation device 100 may provide the trained abnormality detection model to the abnormality detection device 200.

The abnormality detection device 200 inputs the sample data to be detected into the abnormality detection model, and whether the sample data to be detected is normal or abnormal based on the abnormality degree score derived from the abnormality degree map output from the abnormality detection model. You may judge.
[Model generator]
A model generator 100 according to an embodiment of the present disclosure will be described with reference to FIG. FIG. 2 is a block diagram showing a functional configuration of the model generation device 100 according to the embodiment of the present disclosure.

As shown in FIG. 2, the model generation device 100 may include a model processing unit 110, a label conversion unit 120, and a loss calculation unit 130.

The model processing unit 110 may input the detection target data into the abnormality detection model and acquire the abnormality degree map from the abnormality detection model. Specifically, the model processing unit 110 may input sample data such as image data among the sample data with a training label stored in the database 300 into the abnormality detection model to acquire an abnormality degree map. The abnormality degree map may indicate an abnormality portion in the sample data, and may be, for example, a two-dimensional map in which a portion having a high degree of abnormality is indicated by a large value and a portion other than the portion is indicated by a small value.

In one embodiment, the anomaly detection model may be a convolutional neural network, or it may detect anomalies based on the difference between normal sample data without anomalies and input detection target data. Or it may be an appropriate calculation model. The model processing unit 110 may update the parameters of the convolutional neural network of the anomaly detection model or the calculation model according to the training method described later.

The label conversion unit 120 may convert the label of the detection target data into an area label. Here, the area label may include a pair of area data indicating an area in the detection target data and a label value indicating whether the area is normal or abnormal. Specifically, the label conversion unit 120 converts the labels L1, L2 or L3 corresponding to the sample data input to the abnormality detection model among the training labeled sample data into the area labels RL1, RL2 or RL3, respectively. May be good. The area label may be composed of a pair of a binary image indicating an area in the sample data and a binary value "good" or "bad" indicating whether the area is normal or abnormal, and the label conversion unit 120 may be composed of a pair. , Area labels RL1, RL2 or RL3 may be generated corresponding to the type of label L1, L2 or L3 to be input.

As an example, the label L1 of the detection target data may include a binary value (good / bad) indicating whether or not the entire detection target data contains an abnormality. For example, if the sample data contains an abnormality such as a scratch, the label of the sample data may be set to "bad", and if the sample data does not contain an abnormality and is normal, the sample is concerned. The label of the data may be set to "good". In this case, the area label RL1 corresponding to the detection target data may include a pair of the area data indicating the entire detection target data and the label value indicating the binary value. For example, as shown in the figure, the label conversion unit 120 uses the label L1 in which the entire sample data is set to “bad”, the area data indicating the entire sample data, and the label value “indicating that the area is abnormal”. It may be converted to the area label RL1 indicating the pair with “bad”.

Further, as another example, the label of the detection target data may include rectangular information indicating that the rectangular area in the detection target data contains an abnormality. For example, when the rectangular region in the sample data contains at least one abnormality such as a scratch, the label L2 of the sample data may indicate this rectangular region. When the sample data contains a plurality of abnormalities, a rectangular area may be set for each abnormal part, or one or more rectangular areas including all or a part of the abnormal parts are set. You may. In this case, the area label RL3 corresponding to the detection target data may include a pair of the area data indicating the rectangular area and the label value “bad” indicating that the area is abnormal. Further, the area label RL2 corresponding to the detection target data is a pair of the area data indicating the area excluding the rectangular area in the detection target data and the label value “good” indicating that the area excluding the rectangular area is normal. And may be further included. For example, as shown in the figure, the label conversion unit 120 uses the label L2 formed by the rectangular area set in “bad” as the area data indicating the area indicated by the diagonal line and the label indicating that the area is normal. The area label RL2 indicating a pair with the value "good", the area data indicating the area indicated by the diagonal line, and the area label RL3 indicating the pair with the label value "bad" indicating that the area is abnormal. It may be converted.

Further, as yet another example, the label L3 of the detection target data may include area information indicating that the designated area in the detection target data contains an abnormality. For example, when the designated area in the sample data contains at least one abnormality such as a scratch, the label L3 of the sample data may indicate this designated area. When the sample data contains a plurality of abnormalities, a designated area may be set for each abnormal part, or one or more designated areas including all or some abnormal parts may be set. May be done. In this case, the area label RL5 corresponding to the detection target data may include a pair of the area data indicating the designated area in the detection target data and the label value “bad” indicating that the designated area is abnormal. good. Further, the area label RL4 corresponding to the detection target data is a pair of the area data indicating the area excluding the designated area in the detection target data and the label value “good” indicating that the area excluding the designated area is normal. And may be further included. For example, as shown in the figure, the label conversion unit 120 displays the label L3 by the designated area set in "bad" with the area data indicating the area indicated by the diagonal line and the label indicating that the area is normal. The area label RL4 indicating a pair with the value "good", the area data indicating the area indicated by the diagonal line, and the area label RL5 indicating the pair with the label value "bad" indicating that the area is abnormal. It may be converted.

When the abnormality degree map and the area label are acquired in this way, the loss calculation unit 130 may acquire the abnormality degree score from the abnormality degree map and the area data of the area label. Specifically, the loss calculation unit 130 determines the region based on the anomaly degree map of the sample data and the binary image showing the region in the sample data (for example, the entire sample data, the binary image of the rectangular region or the designated region). A focused area anomaly map may be generated, or an anomaly score may be obtained from the area anomaly map by a pooling function.

For example, the loss calculation unit 130 collates the binary image of each pair of area labels with the anomaly degree map, and generates a partial anomaly degree map focusing only on the area shown in the binary image as an area anomaly degree map. Then, the loss calculation unit 130 may apply the pooling function to the generated area abnormality degree map to calculate the abnormality degree score representing the abnormality degree of the region. As a specific pooling function, a global polling function such as global max polling, global average polling, and global Lp polling may be used.

When the abnormality degree score is acquired in this way, the loss calculation unit 130 may acquire the loss value of the detection target data from the abnormality degree score for the area abnormality degree map and the label value of the area. Here, any loss function generally used in deep learning may be used for the calculation of the loss value, and for example, sigmoid cross entropy, contrast loss, triple loss, L2 softmax loss, etc. can be used. Is. The loss calculation unit 130 may calculate the abnormality degree score for all the area labels according to the above-mentioned procedure, or may calculate the loss value of each area label from the calculated abnormality degree score. Then, the loss calculation unit 130 may calculate the sum of the calculated loss values for the area label corresponding to each sample data, and the model processing unit uses the total of the calculated loss values as the loss value of the sample data. You may call me 110.

Then, the model processing unit 110 may update the parameters of the abnormality detection model based on the loss value of each detection target data. For example, when the anomaly detection model is realized by a neural network, when the loss value of each sample data is acquired from the loss calculation unit 130, the model processing unit 110 optimizes any appropriate parameter such as the stochastic gradient descent method. According to the method, the parameters of the abnormality detection model may be updated based on the acquired loss value. Specifically, the model processing unit 110 may update the parameter for the loss value corresponding to the label value “good” so as to reduce the loss value, and the model processing unit 110 may update the parameter corresponding to the label value “bad”. The parameter may be updated so as to increase the loss value.

The model processing unit 110 may continue the update process until the end condition is satisfied, and provides the abnormality detection device 200 as a trained abnormality detection model with the abnormality detection model finally acquired after the end condition is satisfied. You may. Here, as the termination condition, for example, the above-mentioned procedure may be executed for a predetermined number of sample data.
[Model generation process]
Next, the model generation process according to the embodiment of the present disclosure may be described with reference to FIG. The model generation process may be realized by the model generation device 100 described above, and is realized, for example, by executing a program or instruction stored in a memory by one or more processors or processing circuits of the model generation device 100. May be done. FIG. 3 is a flowchart showing a model generation process according to an embodiment of the present disclosure.

As shown in FIG. 3, in step S101, the model generation device 100 may input the detection target data into the abnormality detection model, or may acquire the abnormality degree map from the abnormality detection model. Specifically, when sample data such as image data is acquired, the model generation device 100 may input the acquired sample data into the abnormality detection model to be trained, and acquires an abnormality degree map from the abnormality detection model. May be good. The anomaly degree map may be a two-dimensional map showing the position of an abnormal part in the sample data. For example, in the case of visual inspection, the real number indicating whether or not the sample data contains scratches and the said. It may consist of a pair with an image showing the position of the scratch. The image data for the visual inspection may be a monochrome image or a color image. Further, the detection target data may be any appropriate data according to the detection target, such as hyperspectral image data, depth image data, audio data, waveform data, and three-dimensional data.

In step S102, the model generation device 100 may convert the label of the detection target data into the area label. For example, when the detection target data is image data for visual inspection, a label indicating the presence or absence of scratches in the image data is stored in the database 300 as training data in association with the image data. The model generation device 100 may acquire a label corresponding to the image data input to the abnormality detection model in step S101 from the database 300, or may convert the label into area data as an intermediate representation. Here, the area data may include a pair of a binary image indicating an area in the image data and a binary value indicating whether the area is normal or abnormal, and is generated according to the label type. May be done. In the binary image, a large value may be assigned to an area with a high degree of anomaly, and a small value may be assigned to an area other than the normal area.

For example, if the image data label contains a binary value indicating whether the entire image data contains anomalies, as shown in L1 of FIG. 2, the model generator 100 has the model generator 100 as shown in RL1. A region label for the image data may be generated to include a pair of a binary image indicating the entire image data and a label value "bad" indicating the binary value.

Further, when the label of the image data includes rectangular information indicating that the rectangular region in the image data contains an abnormality as shown in L2 of FIG. 2, the model generator 100 indicates in RL2. A pair of a binary image indicating a rectangular area, a label value “bad” indicating that the area is abnormal, and a binary indicating an area excluding the rectangular area in the image data as shown in RL3. An area label for the image data may be generated to include a pair of the image and a label value "good" indicating that the area other than the rectangular area is normal.

Further, when the label of the image data includes the area information indicating that the designated area in the image data contains an abnormality as shown in L3 of FIG. 2, the model generator 100 indicates in RL4. A pair of a binary image indicating a designated area, a label value “bad” indicating that the designated area is abnormal, and a binary indicating an area excluding the designated area in the image data as shown in RL5. An area label for the image data may be generated so as to include a pair of an image and a label value "good" indicating that the area other than the designated area is normal.

In step S103, the model generation device 100 may acquire the anomaly degree score from the anomaly degree map and the area data. For example, when the anomaly degree map and the area label RL1 as shown in FIG. 2 are acquired for the sample data, the model generation device 100 creates the anomaly degree map and the binary image as the area data showing the entire image data. On the other hand, global max polling may be applied, and for example, a region score {1.28, bad} composed of an abnormality score “1.28” and a label value “bad” may be acquired.

Further, when the anomaly degree map and the area labels RL2 and RL3 as shown in FIG. 2 are acquired for the sample data, the model generator 100 may collate the anomaly degree map with the binary image of the RL2. , A partial anomaly degree map focusing only on the area of the label value “good” in the binary image may be generated as an area anomaly degree map. Then, the model generator 100 applies global max polling to the region abnormality degree map, and for example, the region score {0.30, which is composed of the abnormality degree score “0.30” and the label value “good”. Get good}. Further, the model generation device 100 collates the abnormality degree map with the binary image of RL3, and generates a partial abnormality degree map focusing only on the area of the label value “bad” in the binary image as the area abnormality degree map. May be good. Then, the model generator 100 applies global max polling to the region abnormality degree map, and for example, the region score {1.26, which is composed of the abnormality degree score “1.26” and the label value “bad”. Get bad}.

Further, when the anomaly degree map and the area labels RL4 and RL5 as shown in FIG. 2 are acquired for the sample data, the model generator 100 may collate the anomaly degree map with the binary image of the RL4. , A partial anomaly degree map focusing only on the area of the label value “good” in the binary image may be generated as an area anomaly degree map. Then, the model generator 100 applies global max polling to the region abnormality degree map, and for example, the region score {0.30, which is composed of the abnormality degree score “0.30” and the label value “good”. Get good}. Further, the model generation device 100 collates the abnormality degree map with the binary image of the RL5, and generates a partial abnormality degree map focusing only on the area of the label value “bad” in the binary image as the area abnormality degree map. Then, the model generator 100 applies global max polling to the region abnormality degree map, and for example, the region score {1.24, which is composed of the abnormality degree score “1.24” and the label value “bad”. Get bad}.

In step S104, the model generation device 100 may acquire the loss value of the detection target data from the abnormality degree score and the label value. For example, when the region score RS1 is acquired for the sample data, the model generator 100 applies the anomaly score to any appropriate loss function such as sigmoid cross entropy, cross entropy, triple loss, L2 softmax loss, and the like. The loss value of the sample data may be calculated. Further, when the region scores RS2 and RS3 are acquired for the sample data, the model generator 100 applies each abnormality degree score of RS2 and RS3 to the loss function, and the sum of the derived loss values is the loss of the sample data. It may be determined as a value. Similarly, when the region scores RS4 and RS5 are acquired for the sample data, the model generator 100 applies each abnormality score of RS4 and RS5 to the loss function, and the sum of the derived loss values is the sum of the derived loss values of the sample data. It may be determined as a loss value.

In step S105, the model generation device 100 may update the parameters of the abnormality detection model based on the loss value of the detection target data. When the anomaly detection model is realized by a neural network such as a convolutional neural network, the model generator 100 follows an appropriate parameter optimization method such as a stochastic gradient descent method, and the anomaly detection model is based on the acquired loss value. You may update the parameters of. At this time, the model generator 100 may update the parameter for the loss value corresponding to the label value “good” so as to reduce the loss value, and for the loss value corresponding to the label value “bad”, the parameter may be updated. , The parameters may be updated to increase the loss value.

In step S106, the model generator 100 may determine whether the end condition is satisfied. For example, the end condition may be that steps S101 to S105 are executed for a predetermined number of sample data. When the end condition is satisfied (S106: YES), the model generation process is completed, and the finally acquired abnormality detection model may be provided to the abnormality detection device 200. On the other hand, when the end condition is not satisfied (S106: NO), the model generator 100 may repeat steps S101 to S105 described above for the next sample data.
[Abnormality detection device]
Next, the abnormality detection device 200 according to the embodiment of the present disclosure will be described with reference to FIG. The abnormality detection device 200 may determine the presence or absence of an abnormality in unknown sample data by using the abnormality detection model generated by the model generation device 100. FIG. 4 is a block diagram showing a functional configuration of the abnormality detection device 200 according to the embodiment of the present disclosure.

As shown in FIG. 4, the abnormality detection device 200 may have an abnormality degree map generation unit 210 and an abnormality detection unit 220.

The abnormality degree map generation unit 210 may input the detection target data into the trained abnormality detection model and acquire the abnormality degree map from the trained abnormality detection model. Specifically, when the abnormality degree map generation unit 210 receives sample data such as unknown image data to be detected, the received sample data is input to the trained abnormality detection model and output from the trained abnormality detection model. The abnormality degree map may be sent to the abnormality detection unit 220. For example, in the visual inspection, a plurality of image data acquired by imaging the article to be inspected from various angles can be input to the abnormality detection device 200. In this case, the anomaly degree map generation unit 210 may generate an anomaly degree map for each type of image data by using a trained anomaly detection model corresponding to these a plurality of types of image data.

The abnormality detection unit 220 may acquire the abnormality degree score from the abnormality degree map according to the pooling function, or may determine whether the detection target data is normal or abnormal based on the abnormality degree score. Specifically, when the anomaly map is acquired from the anomaly map generation unit 210, the anomaly detection unit 220 applies any appropriate pooling function such as global max polling to the anomaly map to determine the anomaly score. You may. Then, the abnormality detection unit 220 may determine whether or not there is an abnormality in the sample data to be detected by determining whether or not the determined abnormality degree score is equal to or higher than a predetermined threshold value. For example, the abnormality detection unit 220 may determine that there is an abnormality in the sample data when the abnormality degree score is equal to or more than a predetermined threshold value, and when the abnormality degree score is less than a predetermined threshold value, the sample may be determined. The data may be determined to be normal.
[Abnormality detection processing]
Next, an abnormality detection process according to an embodiment of the present disclosure will be described with reference to FIG. The anomaly detection process is realized by the above-mentioned anomaly detection device 200, for example, by executing a program or instruction in which one or more processors or processing circuits of the anomaly detection device 200 are stored in one or more memories. It may be realized. FIG. 5 is a flowchart showing an abnormality detection process according to an embodiment of the present disclosure.

As shown in FIG. 5, in step S201, the abnormality detection device 200 may input the detection target data into the trained abnormality detection model and acquire the abnormality degree map from the trained abnormality detection model.

In step S202, the abnormality detection device 200 may acquire the abnormality degree score from the abnormality degree map according to the pooling function.

In step S203, the abnormality detection device 200 may determine whether the abnormality degree score is equal to or higher than a predetermined threshold value. When the abnormality degree score is equal to or higher than a predetermined threshold value (S203: YES), the abnormality detection device 200 may determine that the sample data is abnormal in step S204. On the other hand, when the abnormality degree score is less than a predetermined threshold value (S203: NO), the abnormality detection device 200 may determine that the sample data is normal in step S205.
[Hardware configuration]
A part or all of each device (model generation device 100 or abnormality detection device 200) in the above-described embodiment may be composed of hardware, a CPU (Central Processing Unit), or a GPU (Graphics Processing Unit). ) Etc. may be composed of information processing of software (program) executed. When it is composed of information processing of software, the software that realizes at least a part of the functions of each device in the above-described embodiment is a flexible disk, a CD-ROM (Compact Disc-Read Only Memory), or a USB (Universal). Serial Bus) Information processing of software may be executed by storing it in a non-temporary storage medium (non-temporary computer-readable medium) such as a memory and reading it into a computer. In addition, the software may be downloaded via a communication network. Further, information processing may be executed by hardware by mounting the software on a circuit such as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array).

The type of storage medium that stores the software is not limited. The storage medium is not limited to a removable one such as a magnetic disk or an optical disk, and may be a fixed storage medium such as a hard disk or a memory. Further, the storage medium may be provided inside the computer or may be provided outside the computer.

FIG. 6 is a block diagram showing an example of the hardware configuration of each device (model generation device 100 or abnormality detection device 200) in the above-described embodiment. As an example, each device includes a processor 101, a main storage device 102 (memory), an auxiliary storage device 103 (memory), a network interface 104, and a device interface 105, which are connected via a bus 106. It may be realized as a computer 107.

The computer 107 of FIG. 6 includes one component for each component, but may include a plurality of the same components. Further, although one computer 107 is shown in FIG. 6, software is installed on a plurality of computers, and each of the plurality of computers executes the same or different part of the software. May be good. In this case, it may be a form of distributed computing in which each computer communicates via a network interface 104 or the like to execute processing. That is, each device (model generation device 100 or abnormality detection device 200) in the above-described embodiment realizes a function by executing an instruction stored in one or a plurality of storage devices by one or a plurality of computers. It may be configured as a system. Further, the information transmitted from the terminal may be processed by one or a plurality of computers provided on the cloud, and the processing result may be transmitted to the terminal.

Various operations of each device (model generation device 100 or abnormality detection device 200) in the above-described embodiment can be performed in parallel using one or more processors or by using a plurality of computers via a network. It may be executed. Further, various operations may be distributed to a plurality of arithmetic cores in the processor and executed in parallel processing. In addition, some or all of the processes, means, etc. of the present disclosure may be executed by at least one of a processor and a storage device provided on the cloud capable of communicating with the computer 107 via a network. As described above, each device in the above-described embodiment may be in the form of parallel computing by one or a plurality of computers.

The processor 101 may be an electronic circuit (processing circuit, Processing circuit, Processing circuitry, CPU, GPU, FPGA, ASIC, etc.) including a computer control device and an arithmetic unit. Further, the processor 101 may be a semiconductor device or the like including a dedicated processing circuit. The processor 101 is not limited to an electronic circuit using an electronic logic element, and may be realized by an optical circuit using an optical logic element. Further, the processor 101 may include a calculation function based on quantum computing.

The processor 101 can perform arithmetic processing based on data and software (programs) input from each device or the like having an internal configuration of the computer 107, and output the arithmetic result or control signal to each device or the like. The processor 101 may control each component constituting the computer 107 by executing an OS (Operating System) of the computer 107, an application, or the like.

Each device (model generation device 100 or abnormality detection device 200) in the above-described embodiment may be realized by one or a plurality of processors 101. Here, the processor 101 may refer to one or more electronic circuits arranged on one chip, or may refer to one or more electronic circuits arranged on two or more chips or two or more devices. You may point. When a plurality of electronic circuits are used, each electronic circuit may communicate by wire or wirelessly.

The main storage device 102 is a storage device that stores instructions executed by the processor 101, various data, and the like, and the information stored in the main storage device 102 is read out by the processor 101. The auxiliary storage device 103 is a storage device other than the main storage device 102. Note that these storage devices mean any electronic component capable of storing electronic information, and may be a semiconductor memory. The semiconductor memory may be either a volatile memory or a non-volatile memory. The storage device for storing various data in each device (model generation device 100 or abnormality detection device 200) in the above-described embodiment may be realized by the main storage device 102 or the auxiliary storage device 103, and may be realized by the processor 101. It may be realized by the built-in internal memory. For example, the storage unit in the above-described embodiment may be realized by the main storage device 102 or the auxiliary storage device 103.

Multiple processors may be connected (combined) to one storage device (memory), or a single processor may be connected. A plurality of storage devices (memory) may be connected (combined) to one processor. Each device (model generation device 100 or abnormality detection device 200) in the above-described embodiment is a plurality of processors connected (combined) to at least one storage device (memory) and the at least one storage device (memory). When configured, it may include a configuration in which at least one of the plurality of processors is connected (combined) to at least one storage device (memory). Further, this configuration may be realized by a storage device (memory) and a processor included in a plurality of computers. Further, a configuration in which the storage device (memory) is integrated with the processor (for example, a cache memory including an L1 cache and an L2 cache) may be included.

The network interface 104 is an interface for connecting to the communication network 108 wirelessly or by wire. As the network interface 104, an appropriate interface such as one conforming to an existing communication standard may be used. The network interface 104 may exchange information with the external device 109A connected via the communication network 108. The communication network 108 may be any one of WAN (Wide Area Network), LAN (Local Area Network), PAN (Personal Area Network), or a combination thereof, and may be a combination of the computer 107 and the external device 109A. Any information can be exchanged between them. An example of WAN is the Internet, an example of LAN is IEEE802.11, Ethernet (registered trademark), etc., and an example of PAN is Bluetooth (registered trademark), NFC (Near Field Communication), etc.

The device interface 105 is an interface such as USB that directly connects to the external device 109B.

The external device 109A is a device connected to the computer 107 via a network. The external device 109B is a device that is directly connected to the computer 107.

The external device 109A or the external device 109B may be an input device as an example. The input device is, for example, a device such as a camera, a microphone, a motion capture, various sensors, a keyboard, a mouse, or a touch panel, and gives the acquired information to the computer 107. Further, it may be a device including an input unit, a memory and a processor such as a personal computer, a tablet terminal, or a smartphone.

Further, the external device 109A or the external device 109B may be an output device as an example. The output device may be, for example, a display device such as an LCD (Liquid Crystal Display), a CRT (Cathode Ray Tube), a PDP (Plasma Display Panel), or an organic EL (Electro Luminescence) panel, and outputs audio or the like. It may be a speaker or the like. Further, it may be a device including an output unit such as a personal computer, a tablet terminal, or a smartphone, a memory, and a processor.

Further, the external device 109A or the external device 109B may be a storage device (memory). For example, the external device 109A may be a network storage or the like, and the external device 109B may be a storage such as an HDD.

Further, the external device 109A or the external device 109B may be a device having some functions of the components of each device (model generation device 100 or abnormality detection device 200) in the above-described embodiment. That is, the computer 107 may transmit or receive a part or all of the processing result of the external device 109A or the external device 109B.

In the present specification (including claims), the expression (including similar expressions) of "at least one (one) of a, b and c" or "at least one (one) of a, b or c" is used. When used, it includes any of a, b, c, ab, ac, bc, or abc. It may also include multiple instances of any element, such as a-a, a-b-b, a-a-b-b-c-c, and the like. It also includes adding elements other than the listed elements (a, b and c), such as having d, such as a-b-c-d.

In the present specification (including claims), when expressions such as "with data as input / based on / according to / according to" (including similar expressions) are used, unless otherwise specified. This includes the case where various data itself is used as an input, and the case where various data are processed in some way (for example, noise-added data, normalized data, intermediate representation of various data, etc.) are used as input. In addition, when it is stated that some result can be obtained "based on / according to / according to the data", it includes the case where the result can be obtained based only on the data, and other data other than the data. It may also include cases where the result is obtained under the influence of factors, conditions, and / or conditions. In addition, when it is stated that "data is output", unless otherwise specified, various data itself is used as output, or various data is processed in some way (for example, noise is added, normal). It also includes the case where the output is output (intermediate representation of various data, etc.).

In the present specification (including claims), when the terms "connected" and "coupled" are used, direct connection / coupling and indirect connection / coupling are used. , Electrically connected / combined, communicatively connected / combined, operatively connected / combined, physically connected / combined, etc. Intended as a term. The term should be interpreted as appropriate according to the context in which the term is used, but any connection / combination form that is not intentionally or naturally excluded is not included in the term. It should be interpreted in a limited way.

In the present specification (including claims), when the expression "A is configured to B" is used, the physical structure of the element A can perform the operation B. Including that the element A has a configuration and the permanent or temporary setting (setting / configuration) of the element A is set (configured / set) to actually execute the operation B. good. For example, when the element A is a general-purpose processor, the processor has a hardware configuration capable of executing the operation B, and the operation B is set by setting a permanent or temporary program (instruction). It suffices if it is configured to actually execute. Further, when the element A is a dedicated processor, a dedicated arithmetic circuit, or the like, the circuit structure of the processor actually executes the operation B regardless of whether or not the control instruction and data are actually attached. It only needs to be implemented.

In the present specification (including claims), when a term meaning inclusion or possession (for example, "comprising / including" and "having", etc.) is used, the object of the term is used. It is intended as an open-ended term, including the case of containing or owning an object other than the indicated object. If the object of these terms that mean inclusion or possession is an expression that does not specify a quantity or suggests a singular (an expression with a or an as an article), the expression is interpreted as not being limited to a specific number. It should be.

In the present specification (including claims), expressions such as "one or more" or "at least one" are used in some places, and the quantity is specified in other places. Even if expressions that do not or suggest the singular (expressions with a or an as an article) are used, the latter expression is not intended to mean "one". In general, expressions that do not specify a quantity or suggest a singular (expressions with a or an as an article) should be interpreted as not necessarily limited to a particular number.

In the present specification, when it is stated that a specific effect (advantage / result) can be obtained for a specific configuration of an embodiment, unless there is a specific reason, one or more of the other configurations having the configuration. It should be understood that the effect can also be obtained in the examples of. However, it should be understood that the presence or absence of the effect generally depends on various factors, conditions, and / or states, etc., and that the effect cannot always be obtained by the configuration. The effect is merely obtained by the configuration described in the examples when various factors, conditions, and / or conditions are satisfied, and in the invention relating to the claim that defines the configuration or a similar configuration. , The effect is not always obtained.

In the present specification (including claims), when terms such as "maximize" are used, the global maximum value is obtained, the approximate value of the global maximum value is obtained, and the local maximum value is obtained. Should be interpreted as appropriate according to the context in which the term was used, including finding an approximation of the local maximum. It also includes probabilistically or heuristically finding approximate values of these maximum values. Similarly, when terms such as "minimize" are used, finding the global minimum, finding an approximation of the global minimum, finding the local minimum, and finding the local minimum. It should be interpreted as appropriate according to the context in which the term was used, including finding an approximation of the value. It also includes probabilistically or heuristically finding approximate values of these minimum values. Similarly, when terms such as "optimize" are used, finding a global optimal value, finding an approximation of a global optimal value, finding a local optimal value, and local optimization It should be interpreted as appropriate according to the context in which the term was used, including finding an approximation of the value. It also includes probabilistically or heuristically finding approximate values of these optimal values.

In the present specification (including claims), when a plurality of hardware performs a predetermined process, the respective hardware may cooperate to perform the predetermined process, or some hardware may perform the predetermined process. You may do all of the above. Further, some hardware may perform a part of a predetermined process, and another hardware may perform the rest of the predetermined process. In the present specification (including claims), when expressions such as "one or more hardware performs the first process and the one or more hardware performs the second process" are used. , The hardware that performs the first process and the hardware that performs the second process may be the same or different. That is, the hardware that performs the first process and the hardware that performs the second process may be included in the one or more hardware. The hardware may include an electronic circuit, a device including the electronic circuit, or the like.

In the present specification (including claims), when a plurality of storage devices (memory) store data, each storage device (memory) among the plurality of storage devices (memory) stores only a part of the data. It may be stored or the entire data may be stored.

Although the embodiments of the present disclosure have been described in detail above, the present disclosure is not limited to the individual embodiments described above. Various additions, changes, replacements, partial deletions, etc. are possible without departing from the conceptual idea and purpose of the present invention derived from the contents defined in the claims and their equivalents. For example, in all the above-described embodiments, when numerical values or mathematical formulas are used for explanation, they are shown as examples, and the present invention is not limited thereto. Further, the order of each operation in the embodiment is shown as an example, and is not limited to these.

This application asserts this in the interest of the priority of Japanese Patent Application No. 2020-013838, which was filed on January 30, 2020, and the entire contents of No. 2020-013838 are incorporated in this application. ..

Claims (21)

1. A step in which one or more processors input detection target data into an abnormality detection model and acquire an abnormality degree map from the abnormality detection model.
   A step in which the one or more processors obtains an anomaly score from the anomaly map and area data indicating an area in the detection target data.
   A step in which the one or more processors obtains a loss value of the detection target data from the abnormality score and the label value of the region.
   A step in which the one or more processors update the parameters of the anomaly detection model based on the loss value.
   Model generation method having.
2. The one or more processors include a step of acquiring an area label including a pair of an area data indicating an area in the detection target data and a label value of the area.
   The model generation method according to claim 1, wherein the step of acquiring the area label is performed by converting the label of the detection target data into the area label.
3. The detection target data is image data and
   The anomaly degree map shows anomalous parts in the image data.
   The anomaly score indicates the anomaly of the region.
   The model generation method according to claim 1 or 2, wherein the area label includes a pair of a binary image indicating an area in the image data and a binary value indicating whether the area is normal or abnormal. ..
4. The step of acquiring the anomaly degree score is to generate a region anomaly degree map of the region based on the anomaly degree map and the binary image, and acquire the anomaly degree score from the region anomaly degree map by a pooling function. Item 3. The model generation method according to item 3.
5. The label of the detection target data includes a binary value indicating whether or not the entire detection target data contains an abnormality.
   The model generation method according to any one of claims 1 to 4, wherein the area label corresponding to the detection target data includes a pair of an area data indicating the entire detection target data and a label value indicating the binary value.
6. The label of the detection target data includes rectangular information indicating that the rectangular region in the detection target data contains an abnormality.
   The area label corresponding to the detection target data indicates a pair of the area data indicating the rectangular area, the label value indicating that the area is abnormal, and the area excluding the rectangular area in the detection target data. The model generation method according to any one of claims 1 to 5, further comprising a pair of region data and a label value indicating that the region other than the rectangular region is normal.
7. The label of the detection target data includes area information indicating that the designated area in the detection target data contains an abnormality.
   The area label corresponding to the detection target data includes a pair of the area data indicating the designated area, a label value indicating that the designated area is abnormal, and an area excluding the designated area in the detection target data. The model generation method according to any one of claims 1 to 6, further comprising a pair of the indicated area data and a label value indicating that the area other than the designated area is normal.
8. With one or more memories
   With one or more processors
   Have,
   The one or more processors
   Input the detection target data into the anomaly detection model, acquire the anomaly degree map from the anomaly detection model, and
   An abnormality score is obtained from the abnormality degree map and the area data indicating the area in the detection target data, and the abnormality degree score is acquired.
   The loss value of the detection target data is acquired from the abnormality degree score and the label value of the area, and the loss value is obtained.
   Update the parameters of the anomaly detection model based on the loss value.
   A model generator configured as such.
9. The one or more processors acquire an area label including a pair of an area data indicating an area in the detection target data and a label value of the area.
   The model generation device according to claim 8, wherein the acquisition of the area label is performed by converting the label of the detection target data into the area label.
10. The detection target data is image data and
    The anomaly degree map shows the anomalous part of the image data.
    The anomaly score indicates the anomaly of a region in the image data.
    The model generator according to claim 8 or 9, wherein the area label includes a pair of a binary image indicating an area in the image data and a binary value indicating whether the area is normal or abnormal.
11. 10. The one or more processors generate a region anomaly map of the region based on the anomaly map and the binary image, and acquire the anomaly score from the region anomaly map by a pooling function. The model generator described.
12. The label of the detection target data includes a binary value indicating whether or not the entire detection target data contains an abnormality.
    The model generation device according to any one of claims 8 to 11, wherein the area label corresponding to the detection target data includes a pair of an area data indicating the entire detection target data and a label value indicating the binary value.
13. The label of the detection target data includes rectangular information indicating that the rectangular region in the detection target data contains an abnormality.
    The area label corresponding to the detection target data indicates a pair of the area data indicating the rectangular area, the label value indicating that the area is abnormal, and the area excluding the rectangular area in the detection target data. The model generator according to any one of claims 8 to 12, comprising a pair of region data and a label value indicating that the region other than the rectangular region is normal.
14. The label of the detection target data includes area information indicating that the designated area in the detection target data contains an abnormality.
    The area label corresponding to the detection target data includes a pair of the area data indicating the designated area, a label value indicating that the designated area is abnormal, and an area excluding the designated area in the detection target data. The model generator according to any one of claims 8 to 13, comprising a pair of the area data shown and a label value indicating that the area other than the designated area is normal.
15. The process of inputting the detection target data into the abnormality detection model and acquiring the abnormality degree map from the abnormality detection model,
    The process of acquiring the anomaly degree score from the anomaly degree map and the area data indicating the area in the detection target data, and
    The process of acquiring the loss value of the detection target data from the abnormality score and the label value of the area, and
    The process of updating the parameters of the abnormality detection model based on the loss value, and
    A program that causes one or more computers to run.
16. A step in which one or more processors input detection target data into a trained anomaly detection model and acquire an anomaly degree map from the trained anomaly detection model.
    A step in which the one or more processors obtains an anomaly score in a predetermined area from the anomaly map.
    A step in which the one or more processors determine the state of the data to be detected based on the anomaly score.
    Anomaly detection method with.
17. The abnormality detection method according to claim 16, wherein the abnormality degree score is acquired according to a pooling function.
18. The detection target data is a plurality of image data, and is
    The abnormality detection method according to claim 16 or 17, wherein the abnormality detection model acquires an abnormality degree map for the plurality of image data taken from different angles.
19. The abnormality detection method according to any one of claims 16 to 18, wherein the step of determining the detection target data is a step of determining whether the detection target data is normal or abnormal.
20. With one or more memories
    With one or more processors
    Have,
    The one or more processors
    Input the detection target data into the trained anomaly detection model, acquire the anomaly degree map from the trained anomaly detection model, and obtain the anomaly degree map.
    The anomaly score is obtained from the anomaly map according to the pooling function.
    The state of the detection target data is determined based on the abnormality degree score.
    Anomaly detection device configured as such.
21. The process of inputting the detection target data into the trained anomaly detection model and acquiring the anomaly degree map from the trained anomaly detection model,
    The process of acquiring the anomaly score from the anomaly map according to the pooling function,
    A process of determining the state of the detection target data based on the abnormality score, and
    A program that causes one or more computers to run.

Images