Recognition device, recognition processing method, and program

Abstract

To easily reproduce a recognition environment used in past recognition processing.SOLUTION: A recognition device acquires recognition target information to be a target of recognition processing, acquires information output in the past recognition processing performed using a model and indicating the content of the past recognition processing, and performs the recognition processing using the model on the recognition target information according to the information indicating the content of the past recognition processing.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to a recognition device, a recognition processing method, and a program, and relates to, for example, deformation detection processing for an image of a structure.

Background Art

[0002] Recognition devices and recognition services for recognizing recognition targets such as predetermined objects from information such as images are provided. In such a recognition device, learning of a model such as a CNN (Convolutional Neural Network) is performed using a large amount of learning data, and the recognition target can be recognized from the input information using the learned model.

[0003] For example, Patent Document 1 discloses a technique for detecting cracks from a captured image captured by an unmanned aerial vehicle using a learning model. In Patent Document 1, cracks are detected using two types of models: a special learning model capable of identifying a specific type of crack and a non-special learning model capable of identifying a plurality of types of cracks other than the specific type of crack.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] Users of recognition devices may want to compare the results of recognition processing on past information (e.g., past images of a structure) with the results of recognition processing on current information (e.g., current images of a structure). In such cases, it is desirable to maintain a recognition environment that can reproduce the same recognition results as in past recognition processing and apply it to recognition processing on different information in order to compare results. However, manually setting the recognition processing parameters to obtain the same recognition results as in past recognition processing is cumbersome. In particular, the learning model used in recognition processing may be updated in recognition devices. In such cases, in order to use the same learning model as in the past, it is necessary to maintain multiple versions of the learning model and for the user to identify and select the learning model used in past recognition processing from among the multiple versions, which is a cumbersome operation.

[0006] The present invention aims to make it possible to easily reproduce the recognition environment used in past recognition processing. [Means for solving the problem]

[0007] The recognition device according to an embodiment of the present invention has the following configuration. That is, A first acquisition means for acquiring information to be recognized that is the subject of recognition processing, multiple Model Selected models and tunable processing parameters Using Ta Recognition processing Model information that shows the contents of the model, and in order to store model information including the identification information of the model and the processing parameter information, the storage device stores the model information output Output means and From the aforementioned storage device, Information indicating the content of past recognition processing The aforementioned model information Get possible A second means of acquisition, before Model for the information to be recognized information A recognition means that performs recognition processing using, When the recognition means performs recognition processing using new model information different from the model information stored in the storage device, the setting means sets the new model information based on user input or automatically. When the recognition means performs recognition processing using the model information stored in the storage device, and the second acquisition means obtains from the storage device as information indicating the content of past recognition processing, the model identification information included in the model information is the identification information of a model that was previously available, the display means causes a user interface for the user to tune the processing parameters included in the model information to be displayed on a display device, It is equipped with. [Effects of the Invention]

[0008] The recognition environment used in past recognition processing can be easily reproduced.

Brief Description of Drawings

[0009] [Figure 1] A diagram showing a functional configuration example of a recognition device according to an embodiment. [Figure 2] A diagram showing a hardware configuration example of a recognition device according to an embodiment. [Figure 3] A diagram showing an example of a processing flow of a recognition processing method according to an embodiment. [Figure 4] A diagram showing an example of a UI for a user to set model information. [Figure 5] A diagram for explaining the output and input of model information. [Figure 6] A diagram showing an example of a UI displayed when a non-display model is called. [Figure 7] A diagram showing an example of a model management table. [Figure 8] A diagram for explaining the operation of the switching unit 111. [Figure 9] A diagram showing an example of a processing flow for presenting a display target model. [Figure 10] A diagram showing an example of a lookup table indicating the relationship of models. [Figure 11] A diagram showing an example of a UI for presenting the results of recognition processing using each model.

Embodiments for Carrying Out the Invention

[0010] Hereinafter, embodiments will be described in detail with reference to the accompanying drawings. Note that the following embodiments do not limit the invention according to the claims. Although a plurality of features are described in the embodiments, not all of these plurality of features are essential for the invention, and the plurality of features may be arbitrarily combined. Further, in the accompanying drawings, the same or similar configurations are denoted by the same reference numerals, and redundant descriptions are omitted.

[0011] FIG. 1 shows the functional configuration of a recognition device according to an embodiment of the present invention. Information to be recognized (recognition target information) 102 is input to the recognition device 101, and a recognition result 103 obtained by performing a recognition process on the recognition target information 102 is output. The type of the recognition target is not particularly limited and may be, for example, an image, text information, voice information, or signal information. Also, the type of the recognition process is not particularly limited and may include, for example, a process of recognizing a recognition target from an image, a process of recognizing a summary sentence from text information, or a process of recognizing an abnormality from the impact sound of a structure or a sensor signal of a device. In the following embodiments, a case where the recognition device 101 recognizes a deformation occurring in a structure in an image, such as a crack, from the image will be described. The structure may be an infrastructure structure such as a tunnel or a bridge.

[0012] The recognition device 101 includes an information input unit (first acquisition unit) 104, a model information input unit (second acquisition unit) 110, and a recognition unit 105. The first acquisition unit 104 acquires recognition target information 102 that is the target of the recognition process. The second acquisition unit 110 acquires information indicating the content of a past recognition process that was output during a past recognition process performed using a model. The information indicating the content of the recognition process can include information for identifying the model, such as a model ID that is the identification information of the model. Also, the information indicating the content of the recognition process can include processing parameters used in the recognition process using the model. Further, the information indicating the content of the recognition process can include information indicating the recognition target. Hereinafter, the information indicating the content of the recognition process will be referred to as model information. That is, the second acquisition unit 110 acquires model information 109.

[0013] Here, the information indicating the content of past recognition processing may be associated with the results of past recognition processing. For example, the information indicating the results of past recognition processing may include the information indicating the content of past recognition processing. In this case, by the user inputting the results of past recognition processing to the second acquisition unit 110, the second acquisition unit 110 can acquire the information indicating the content of past recognition processing. The recognition device 101 may be implemented as a server or a cloud service, in which case the first acquisition unit 104 and the second acquisition unit 110 can acquire the recognition target information 102 and the model information 109 via the network.

[0014] The recognition unit 105 performs recognition processing on the information to be recognized using a model. Here, as will be described later, the recognition unit 105 can perform recognition processing according to information indicating the content of past recognition processing acquired by the second acquisition unit 110. For example, the recognition unit 105 can use a model used in past recognition processing and can also use processing parameters used in past recognition processing for recognition processing on the information to be recognized. As will be described later, the recognition device 101 may be equipped with multiple recognition units 105, in which case each recognition unit 105 corresponds to one or more models from among the multiple models.

[0015] The recognition device 101 may further include a model management unit 106, a setting unit 107, an output unit 108, and a switching unit 111. The model management unit 106 manages one or more models used by the recognition unit 105 to perform recognition processing on the information to be recognized. The model management unit 106 may also select a model to be used for recognition processing from the models it manages according to the model information. For example, the model management unit 106 can select a model identified by the model information from among the models it manages.

[0016] The setting unit 107 can set model information according to user input. The output unit 108 can output the results of the recognition processing performed by the recognition unit 105 on the recognition target information 102. In addition, the output unit 108 can output information (model information) indicating the content of the recognition processing performed by the recognition unit 105 in relation to the results of the recognition processing. The switching unit 111 switches the recognition unit 105 that performs recognition processing according to the model information.

[0017] The recognition device 101 according to this embodiment can be realized by a computer equipped with a processor and memory. Figure 2 shows an example of the hardware configuration of the recognition device 101 according to one embodiment. The CPU (Central Processing Unit) 201 executes the control program for the recognition device 101. The ROM 202 stores this control program and the like. The RAM 203 provides the work area for the CPU 201. The network IF 204 is used to communicate with applications outside the recognition device 101. The storage device 205 is, for example, a hard disk and stores data or programs. In this way, the functions of each part shown in Figure 1 and the like can be realized by a processor such as the CPU 201 executing a program stored in memory such as the ROM 202, RAM 203, or storage device 205.

[0018] Figure 3 shows the processing flow of a recognition processing method performed by a recognition device according to one embodiment of the present invention. In S301, the first acquisition unit 104 acquires an image which is the information to be recognized. In S302, the setting unit 107 determines whether or not to use model information 109 which indicates the contents of past recognition processing. If past model information is not used, the process proceeds to S303; if it is used, the process proceeds to S307.

[0019] In S303, the setting unit 107 sets the model information. In this example, the setting unit 107 sets the model information based on user input. Figure 4 shows the user interface (UI) for the user of the recognition device 101 to set the model information. The UI shown in Figure 4 includes a list 401 for selecting a recognition target and a list 402 for selecting a model to be used for the recognition process. The user can use the list 402 to select a model to be used for the recognition process from one or more models corresponding to the recognition target. Furthermore, the UI shown in Figure 4 includes a slider 403 for specifying processing parameters to be used for the recognition process. The processing parameters may include, for example, a detection quantity parameter that controls the amount of the recognition target detected in the recognition process (for example, making the recognition target easier or harder to detect). The processing parameters may also include a noise reduction parameter that indicates the intensity of the noise reduction process as a preprocessing step for the recognition target information. However, the types of processing parameters are not limited to these.

[0020] Based on user input to this UI, the setting unit 107 can generate model information 404 that indicates the content of the recognition process. The model information 404 shown in Figure 4 indicates the model ID, which is the model identification information, the recognition target, and the processing parameters. In this way, in S303, the user can select a model or tune the processing parameters to obtain better recognition processing results. On the other hand, the method of setting the model information by the setting unit 107 is not limited to this method. For example, the setting unit 107 may set it automatically based on the attributes of the recognition target information (e.g., acquisition time, image brightness, or GPS information) and the type of recognition target.

[0021] In S304, the recognition unit 105 performs recognition processing on the information to be recognized using a model. The recognition unit 105 can perform recognition processing according to the model information set by the setting unit 107 in S303. In addition, the recognition unit 105 can perform recognition processing according to the model information indicating the content of past recognition processing acquired by the second acquisition unit 110 in S307. In this embodiment, the recognition unit 105 performs processing to recognize deformations in the image.

[0022] In S305, the output unit 108 outputs the result of the recognition processing in S304. In this embodiment, the output unit 108 can superimpose the location of the recognized cracks or other deformations onto the image, which is the information to be recognized, and output the superimposed image. The output unit 108 may also output vector data of the cracks. However, as mentioned above, the recognition processing performed by the recognition device according to the present invention is not particularly limited, and the output of the output unit 108 can take various forms.

[0023] In S306, the output unit 108 outputs model information indicating the content of the recognition process in S304. For example, the output unit 108 can output the model information 404 in Figure 4 as text or binary data in a predetermined format. The output unit 108 can output model information in association with the result of the recognition process. For example, information indicating the result of the recognition process may include model information. With this configuration, the user of the recognition device 101 can manage the model information used in the recognition process along with the result of the recognition process, and can find out the model information used to obtain the result of a specific recognition process. The output unit 108 can output the result of the recognition process and the model information to an external device via a network or the like in S305 and S306. On the other hand, the output unit 108 may store the result of the recognition process and / or the model information in a storage device provided by the recognition device 101 in S305 and S306, in which case the result of the recognition process and / or the model information can be output externally at any timing according to the user's request.

[0024] In S307, the second acquisition unit 110 acquires model information that indicates the content of past recognition processing. This model information is the model information that the output unit 108 output in S306 during past recognition processing. That is, as shown in Figure 5, at some point, the user of the recognition device 101 inputs an image to the recognition device 101, and the recognition device 101 outputs a recognition result 501 and model information 502 for this image. The recognition result 501 is obtained by superimposing the result of the recognition processing onto the input image. By acquiring the model information 502 in the second acquisition unit 110, the recognition device 101 can restore the settings used in past recognition processing. Subsequently, in S304, the recognition unit 105 can perform recognition processing on the recognition target information according to the model information acquired in S307.

[0025] On the other hand, in one embodiment, the model management unit 106 can manage a first group of models and a second group of models separately. For example, the model management unit 106 can select a model to be used for recognition processing from the first group of models based on model information set by the setting unit 107 (for example, according to user input), regardless of the model information acquired by the second acquisition unit 110. On the other hand, the model management unit 106 can select a model to be used for recognition processing from the second group of models according to the model information acquired by the second acquisition unit 110. In one embodiment, the model management unit 106 can perform both the process of determining a model to be used for recognition processing from the first group of models according to user input, and the process of selecting a model identified according to model information from the second group of models. Here, the models managed by the model management unit 106 may include display target models that are presented to the user in S303 and can be selected by the user from a list, and hidden models that are not presented to the user in S303 and cannot be selected by the user from a list. In this case, the first group of models consists of the models to be displayed, and the second group of models consists of the models to be displayed and the models to be hidden, with the first group being smaller than the second group of models.

[0026] The provider of the recognition device 101 may update the model for purposes such as improving recognition performance. In this case, the model management unit 106 can manage the model before the update as a hidden model and the model after the update as a model to be displayed. In this way, a model that was previously a model to be displayed can be managed as a hidden model when updated. When the recognition device 101 is used for a new purpose, or when a new user uses the recognition device 101, and the model information is set in S303, the hidden model is concealed from the user. With this configuration, the new model to be displayed is usually used, thus discouraging the use of the old model. In addition, since the models presented to the user are limited to the new model and the number of models presented is reduced, usability is improved. On the other hand, when model information is acquired in S307 in order to reproduce a recognition environment used in the past, the recognition device 101 can use the hidden model indicated by the model information.

[0027] Therefore, in S308, the model management unit 106 determines whether the model indicated by the model information acquired in S307 is a model to be displayed. The model management unit 106 can determine whether the model is a model to be displayed or a model to be hidden according to the model ID contained in the model information. If the model is a model to be displayed, the process proceeds to S304, and the recognition unit 105 performs recognition processing on the recognition target information according to the model and processing parameters indicated by the model information acquired in S307. On the other hand, if the model is not a model to be displayed, the process proceeds to S309.

[0028] In S309, the model management unit 106 calls up the hidden model indicated by the model information acquired in S307. Figure 6 shows the UI displayed when the second acquisition unit 110 acquires model information in S307 and the model management unit 106 calls up the hidden model in S309. The model indicated by the model information acquired in S307 is "crackModel2021-1". On the other hand, as a result of the model update, the three display target models that can be selected for the recognition target "crack" at this point are "crackModel2023-1", "crackModel2023-2", and "crackModel2023-3". However, since "crackModel2021-1" has not been deleted from the model management unit 106, the previously used model is called up in Figure 6 by inputting past model information. On the other hand, as shown in Figure 4, in S303, three displayable models are shown as selectable for the recognition target "crack": "crackModel2023-1", "crackModel2023-2", and "crackModel2023-3". Thus, in S303, the hidden model "crackModel2021-1" is not displayed as an option. Furthermore, in S309, the recognition target, model, and / or processing parameters may be changed based on user input.

[0029] In S310, the switching unit 111 switches the recognition unit 105 according to the model being used. Depending on the model used for recognition processing, different recognition algorithms, libraries, and open-source software (OSS) may be used. Therefore, the recognition device 101 can use multiple recognition units 105. Here, each of the multiple recognition units 105 corresponds to one or more models managed by the model management unit 106. That is, one recognition unit 105 can be associated with each model. The switching unit 111 can switch the recognition unit 105 so that the recognition unit 105 corresponding to the model ID indicated by the model information is used in the recognition processing.

[0030] In this way, the hidden model is called according to the model ID indicated by the model information, and once the recognition unit 105 corresponding to this model is determined, the process proceeds to S304, where the recognition unit 105 performs recognition processing according to the called model and the processing parameters indicated by the model information. The processing from S304 onward is as described above.

[0031] The above process outputs model information indicating the content of the recognition process, and the settings of past recognition processes, i.e., the recognition environment, can be reproduced according to the output model information. Note that it is not necessary for the switching unit 111 to switch the recognition unit 105; the same recognition unit 105 may perform recognition processing using various models.

[0032] (Details of model management) The processing in S307-S310 will be explained in more detail. Figure 7 shows the model management table used by the model management unit 106 to manage models. In the table in Figure 7, a flag indicating whether or not it is a model to be displayed is associated with the model ID, and the ID of the recognition unit 105 that performs recognition processing using this model (recognition unit ID). The model management unit 106 also manages the actual model corresponding to the model ID. In this specification, the actual model means the structure and / or parameters that represent the model. For example, when using a convolutional neural network as a model, the actual model can be represented by structural information indicating the size and number of each layer, and / or parameters such as filter coefficients used in the convolutional processing performed between layers. The model management unit 106 in S30 4 When the recognition unit 105 performs recognition processing, information representing the actual model corresponding to the model ID can be supplied to the recognition unit 105, and the recognition unit 105 can construct the model using the supplied information.

[0033] When adding or updating a model used by the recognition device 101, the model ID and recognition unit ID of the new model can be added to the model management table. In this case, the model management unit 106 can set the flag indicating that only the latest model is a model to be displayed to Yes, and the flag for older models to No. In this case, older models will no longer be selectable when setting the normal model information in S303.

[0034] When using such a model management table, in S308 the model management unit 106 can use the model ID indicated by the model information and the table to determine whether the model indicated by the model ID is a model to be displayed. In S309, the model management unit refers to the table and retrieves the model corresponding to the model ID. Furthermore, in S310, the switching unit 111 refers to the table and switches the recognition unit 105 to use the recognition unit 105 that has the recognition unit ID corresponding to the model ID. Figure 8 is a diagram illustrating how the switching unit 111 switches the recognition unit 105 according to the model. The recognition unit 105 may be a library within the recognition device 101, or it may be a server or container that accepts recognition processing. In addition, each recognition unit 105 may be implemented in a separate cloud service or the like.

[0035] In the example shown in Figure 8, each of the multiple recognition units 105 operates on its corresponding container. Each container executes the recognition process as a batch process. Container 802 is the container for the recognition unit 105 corresponding to the model to be displayed and is normally running. The model management unit 106 manages snapshots (e.g., container images) of the containers corresponding to each of the multiple recognition units 105. Containers, which are the execution environments for applications like the recognition unit 105, and snapshots that preserve the state of the containers, can be provided, for example, using a cloud service.

[0036] In this case, the switching unit 111 can switch the recognition unit 105 to be called based on the model ID. Specifically, the switching unit 111 can start the recognition unit 105 using a snapshot of the container corresponding to the model selected by the model management unit 106 according to the model information. In Figure 8, the switching unit 111 starts the recognition unit 105 used for recognition processing on the container 803 using a snapshot of the recognition unit having a recognition unit ID corresponding to the model ID.

[0037] In the example shown in Figure 8, the model management unit 106 manages the latest models, "crackModel2023-001," "crackModel2023-002," and "crackModel2023-003," as models to be displayed. The recognition unit, which is launched on container 802 using a snapshot with the recognition unit ID "Recognizer003" corresponding to these models, is normally in operation. When model information is set based on user input in S303, and when it is determined in S308 that the model information indicates a model to be displayed, recognition processing is performed using this recognition unit.

[0038] On the other hand, if the model information indicates the hidden model ID "crackModel2021-001", the switching unit 111 refers to the model management table and determines that the recognition unit ID corresponding to this model ID is "Recognizer001". Based on this, the switching unit 111 uses a snapshot of the recognition unit ID "Recognizer001" to start the recognition unit corresponding to this model on container 803. Then, the switching unit 111 switches the recognition unit used for recognition processing from the recognition unit running on container 802 to the recognition unit running on container 803. Then, the recognition unit started on container 803 105 However, S30 4 Then, recognition processing is performed on the information to be recognized.

[0039] Thus, in the recognition device 101, the recognition unit 105 that performs recognition processing using the display target model may be in normal operation. When performing recognition processing using a non-display model, the switching unit 111 activates the recognition unit 105 that performs recognition processing using the non-display model, and the recognition processing can be performed by this recognition unit 105. In other words, the recognition unit 105 that uses the non-display model is normally not in operation and is activated as needed, while the recognition unit 105 that uses the display target model may be in constant operation.

[0040] In the embodiments described above, the correspondence between models and recognition units 105 was managed using the model management table shown in Figure 7. However, the method for determining the recognition unit 105 corresponding to a model is not limited to this method. For example, the model information may indicate the recognition unit ID corresponding to the model to be used. Alternatively, when the setting unit 107 sets the model information in S303, it may obtain user input to select a recognition unit 105 from one or more options, and determine which recognition unit 105 to use based on this user input.

[0041] As described above, the recognition device according to this embodiment can reproduce the recognition processing settings used in the past based on the model information output during past recognition processing. Furthermore, in the embodiment described above, the display target model and the previously unselectable, hidden model are managed in accordance with the addition or update of models. Even when the model information indicates a hidden model, it is possible to reproduce past settings using the hidden model when performing recognition processing according to previously output model information. With this method, under normal circumstances, only the new display target model is presented as an option for the user to use, while if the user wants to reproduce past settings, it is possible to use the old hidden model.

[0042] In the embodiment described above, the model information output by the output unit 108 indicates the model ID, the object to be recognized, and the processing parameters used for the recognition process, and the model management unit 106 manages the model corresponding to the model ID. When the second acquisition unit 110 acquires the model information, recognition processing is performed using the model corresponding to the model ID indicated by the model information. On the other hand, the model information may also include information representing the actual model as described above.

[0043] In this case, when the second acquisition unit 110 acquires model information indicating an entity for the model, the recognition unit 105 can perform recognition processing on the information to be recognized using the model according to the model information. For example, the output unit 108 can output a model information file containing information representing the entity of the model (e.g., model structure information and / or parameters) in a predetermined format. When the second acquisition unit 110 acquires this file, the model management unit 106 can restore the entity of the model based on the model information and pass it to the recognition unit 105. In this case, the second acquisition unit 110 can acquire model information indicating an entity for a model different from the model managed by the model management unit 106, and the recognition unit 105 can perform recognition processing using the model according to this model information.

[0044] With this configuration, the model information output from the recognition device 101 contains information representing the actual model used in the recognition process, and the model can be restored from the model information if it is desired to reproduce the recognition environment. Therefore, even if the recognition device 101 deletes old models when updating the model, the past recognition environment can be reproduced. In this embodiment as well, the switching unit 111 can switch the recognition unit 105 that performs recognition processing using the model. In this case, the model management unit 106 or the model information can hold information that identifies the recognition unit corresponding to the model (for example, a recognition unit ID).

[0045] (Presentation of alternative models to be displayed based on model information) In the above embodiment, the past recognition environment was reproduced using a hidden model according to the model information output in the past. On the other hand, if it is sufficient to reproduce a recognition environment in which a generally similar recognition result can be obtained, the recognition device 101 may present a display target model corresponding to the hidden model indicated by the model information. With such a configuration, when the model information indicates the model before the update, the updated model can be presented to the user, thereby facilitating the replacement of the old model with the updated model.

[0046] Such an embodiment can be realized by performing the process shown in Figure 9 when, in S308, the model information indicates a hidden model, and it is determined that the model identified by the model information is not included in the models to be displayed. That is, first, in S1004, the model management unit 106 presents the user with a model selected from the models to be displayed. Here, the model management unit 106 can present a model to be displayed that corresponds to the hidden model indicated by the model information. For example, the model management unit 106 may manage information indicating the association between hidden models and displayed models. Based on such information, the model management unit 106 may determine the model to be displayed that corresponds to the hidden model identified by the model information and is presented to the user. Specifically, it is possible to determine in advance the updated model that is similar to the model before the update, create a lookup table showing these relationships, and select the updated model to be displayed that corresponds to the hidden model before the update based on this lookup table. Figure 10 shows an example of such a lookup table, where the model to be displayed ID is uniquely determined for each hidden model ID.

[0047] As another example, the display target model corresponding to the hidden model indicated by the model information may be determined based on the similarity of the recognition results. In this case, the recognition unit 105 can perform recognition processing on the recognition target information, a part thereof, or other information using each of the hidden model indicated by the model information and the multiple display target models. The display target model that gives the recognition result closest to the hidden model is then determined to be the display target model corresponding to the hidden model.

[0048] In S1005, the recognition unit 105 performs recognition processing using the hidden model identified by the model information, and recognition processing using the display target model presented to the user in S1004. The recognition unit 105 can perform these recognition processes on the recognition target information, a part thereof, or other information. The model management unit 106 can then present the results of the recognition processing obtained in this way to the user.

[0049] Figure 11 shows an example of a UI that displays the model ID of the displayable model corresponding to the hidden model indicated by the model information, and the results of the recognition process using this displayable model. Here, the displayable model (ID: "crackModel2023-3") corresponding to the hidden model (ID: "crackModel2021-1") indicated by the model information, and the recognition results using each model are presented. Based on these recognition results, the user can decide whether or not to change the model.

[0050] In S1006, the model management unit 106 obtains user instructions indicating whether or not the user agrees to the model change. For example, if the user agrees to the model change, they can press the "Change Model" button on the UI shown in Figure 11. If the user agrees to the model change, the process proceeds to S1007, where the model management unit 106 decides to use the display target model presented in S1004 for the recognition process.

[0051] The model management unit 106 may further delete the hidden model indicated by the model information in S1008. That is, in accordance with the instructions in S1007 from the user to whom the display target model was presented, the model management unit 106 may delete the model identified by the model information. As a result of such deletion, even if the model information acquired by the second acquisition unit 110 indicates this hidden model, this hidden model will become unusable. Various methods can be considered for deleting a model. For example, the model may be physically deleted from the model management unit 106, or the model may be logically deleted from the model management unit 106 so that it becomes unusable. On the other hand, if the recognition device 101 provides services to multiple users or organizations, the model may be logically deleted only for users or organizations that have agreed to the modification or deletion of the model, so that these users or organizations cannot use the model. After that, the process proceeds to S304, and the recognition unit 105 performs recognition processing using the display target model presented in step S1004.

[0052] On the other hand, if the user does not agree to the model change, the process proceeds to S1009, where the model management unit 106 decides to use the hidden model indicated by the model information for the recognition process. Subsequently, the process proceeds to S304, where the recognition unit 105 performs the recognition process using this hidden model. If the hidden model indicated by the model information is no longer supported, for example, because a predetermined amount of time has elapsed since it became a hidden model, the model management unit 106 may issue a warning at S1009 and terminate the process.

[0053] According to this embodiment, the past recognition environment can be reproduced based on the model information output during past recognition processing, and after the model is updated, a model similar to the previous model and the recognition results can be presented, thereby proposing a recognition environment similar to the past one. This configuration can encourage the user to migrate to the new model.

[0054] Another embodiment of the present invention provides a recognition device comprising: a recognition unit 105 that performs recognition processing using a model on recognition target information to be recognized; and an output unit 108 that outputs information indicating the content of the recognition processing performed by the recognition unit 105 in relation to the result of the recognition processing. By using the information indicating the content of the recognition processing output by such a recognition device, it becomes possible to easily reproduce the recognition environment in future recognition processing. The configuration of the recognition unit 105 and the output unit 108 is the same as in the above embodiment, and this recognition device may further have the above-described functional units.

[0055] The recognition device described above may be composed of multiple devices. For example, the recognition device according to one embodiment may be composed of multiple information processing devices connected via a network. In particular, if the recognition device 101 is a server that receives requests from a user terminal, the functions of the recognition unit 105 may be executed by another server. With such a configuration, even if the processing of the recognition unit 105 takes time, the processing of the recognition unit 105 can be performed asynchronously with respect to the request. Furthermore, although the functions of the information processing device shown in Figure 2 can be realized by a computer, some or all of the functions of the information processing device may be realized by dedicated hardware.

[0056] Furthermore, as described above, the recognition unit 105 can take various forms, such as a library within the recognition device 101, or a server or cloud service separate from the recognition device 101. The snapshot 801 of the recognition unit shown in Figure 8 can also take various forms, such as a container image, server image, dynamic link library, cloud service version, or script, depending on the form of the recognition unit 105.

[0057] (Other examples) The present invention can also be realized by supplying a program that implements one or more of the functions of the above-described embodiments to a system or device via a network or storage medium, and by having one or more processors in the computer of that system or device read and execute the program. It can also be realized by a circuit (e.g., an ASIC) that implements one or more functions.

[0058] The invention is not limited to the embodiments described above, and various modifications and variations are possible without departing from the spirit and scope of the invention. Accordingly, claims are attached to disclose the scope of the invention. [Explanation of Symbols]

[0059] 101: Recognition device, 104: First acquisition unit, 105: Recognition unit, 106: Model management unit, 107: Setting unit, 108: Output unit, 110: Second acquisition unit, 111: Switching unit

Claims

1. A first acquisition means for acquiring recognition target information that is the subject of recognition processing, Model information that indicates the content of recognition processing using a model selected from multiple models and tunable processing parameters, and an output means for outputting the model information to a storage device in order to store the model information which includes the identification information of the model and the processing parameter information, A second acquisition means capable of acquiring the model information from the aforementioned storage device as information indicating the content of past recognition processing, A recognition means that performs recognition processing on the aforementioned information to be recognized using model information, When the recognition means performs recognition processing using new model information different from the model information stored in the storage device, the setting means sets the new model information based on user input or automatically. When the recognition means performs recognition processing using the model information stored in the storage device, and the second acquisition means obtains from the storage device as information indicating the content of the past recognition processing, the model identification information included in the model information is the identification information of a model that was previously available, the display means displays a user interface on a display device for the user to tune the processing parameters included in the model information, A recognition device characterized by comprising:

2. The recognition device according to claim 1, characterized in that the information indicating the content of the past recognition process is associated with the result of the past recognition process.

3. The recognition device according to claim 1 or 2, characterized in that the output means outputs information indicating the content of the recognition processing performed by the recognition means in relation to the result of the recognition processing.

4. The information indicating the content of the past recognition process includes identification information of the model used for the recognition process in the past recognition process, The recognition device further comprises a management means for managing models used in the recognition process and selecting a model to be used in the recognition process from the managed models according to the identification information of the model. The recognition device according to any one of claims 1 to 3, characterized in that the recognition means performs recognition processing on the recognition target information using the model selected by the management means.

5. The recognition device according to claim 4, characterized in that the management means selects a model identified by the model identification information from among the models being managed.

6. The management means can perform both the process of selecting a model to be used for the recognition process from a first group of models according to user input, and the process of selecting a model identified by the identification information of the model from a second group of models. The recognition device according to claim 4 or 5, characterized in that the first group of models is smaller than the second group of models.

7. The recognition device according to claim 6, characterized in that the management means presents to the user a model selected from the first model group if the model identified by the model identification information is not included in the first model group.

8. The recognition device according to claim 7, characterized in that the management means presents to the user the result of the recognition process using the model identified by the model identification information and the result of the recognition process using the model presented to the user.

9. The recognition device according to claim 8, wherein the management means deletes a model identified by the model identification information in accordance with an instruction from the user to which a model selected from the first group of models has been presented.

10. The recognition device according to any one of claims 7 to 9, characterized in that the management means manages information indicating the association between a model identified by the model identification information and a model selected from the first group of models and presented to the user.

11. Each of the recognition devices comprises multiple recognition means corresponding to one or more models among the models being managed. The recognition device according to any one of claims 4 to 10, further comprising a switching means for switching the recognition means that performs the recognition processing according to the identification information of the model.

12. The recognition device according to claim 11, characterized in that each of the plurality of recognition means operates on a corresponding container.

13. The recognition device according to claim 12, characterized in that the management means manages snapshots of containers corresponding to each of the plurality of recognition means, and the switching means activates the recognition means using the snapshot of the container corresponding to the model selected by the management means.

14. The recognition device according to any one of claims 4 to 13, characterized in that, in response to the second acquisition means acquiring model information indicating an entity of a model different from the model being managed, the recognition means performs recognition processing on the recognition target information using the different model according to the model information.

15. The recognition device according to any one of claims 4 to 14, characterized in that the model information includes a model ID which is identification information of the model, a recognition target, and the processing parameters.

16. The recognition device according to any one of claims 1 to 15, wherein the processing parameters include at least one of a parameter for controlling the detectability of a recognition target detected in the recognition process and a parameter for indicating the intensity of noise reduction processing as preprocessing for the recognition target information.

17. The recognition device according to any one of claims 1 to 16, characterized in that the object of the recognition process is an image, and the recognition process is a process of recognizing an object to be recognized in the image.

18. The recognition device according to claim 17, characterized in that the object of recognition is a deformation occurring in a structure in the image.

19. A recognition processing method performed by a recognition device, A process to acquire information to be recognized that is the target of the recognition process, A step of outputting model information to a storage device in order to store model information that includes identification information of the model and information of the processing parameters, which is model information that indicates the content of recognition processing using a model selected from multiple models and tunable processing parameters, A step of performing recognition processing on the aforementioned information to be recognized using model information, The model information can be obtained from the aforementioned storage device as information indicating the content of past recognition processing, When performing recognition processing using new model information different from the model information stored in the storage device, the new model information is set based on user input or automatically. A step in which recognition processing is performed using the model information stored in the storage device, and the model identification information included in the model information obtained from the storage device as information indicating the content of the past recognition processing is the identification information of a model that was previously available, includes displaying a user interface on a display device for the user to tune the processing parameters included in the model information; A recognition processing method characterized by comprising:

20. A program for causing a computer to function as a recognition device according to any one of claims 1 to 18.

Images