Detection device, detection method, and detection program

Abstract

To accurately detect the presence of black smoke.SOLUTION: A detection apparatus (1) includes a data acquisition unit (71) configured to acquire training data including a first sample image including black smoke and a background, and a second sample image including white smoke, a background, and no black smoke, a model generation unit (72) configured to generate a detection model by performing machine learning on the first sample image and the second sample image, and a detection unit (74) configured to detect presence or absence of black smoke by inputting an image captured by an image capturing unit (2) to the detection model.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a detection device, a detection method, and a detection program. [Background technology]

[0002] For example, it is desirable to monitor the presence or absence of smoke in a factory to detect whether a fire has occurred in the factory. Patent Document 1 discloses a smoke detection device including an image acquisition unit that acquires an image of a detection target, and a determination unit that determines the presence or absence of smoke by using a machine learning model that has completed machine learning to detect the presence or absence of smoke based on the image acquired by the image acquisition unit. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 2024-039883 Summary of the Invention [Problem to be solved by the invention]

[0004] The smoke detection device disclosed in Patent Document 1 has a problem in that it may mistakenly detect white smoke, which is not smoke caused by a fire, as black smoke. One aspect of the present invention aims to accurately detect the presence or absence of black smoke. [Means for solving the problem]

[0005] In order to solve the above problems, a detection device according to one aspect of the present invention includes: a data acquisition unit that acquires training data including a first sample image that includes black smoke and a background, a second sample image that includes white smoke and a background but does not include black smoke, and a detection result for the presence or absence of black smoke; a model generation unit that uses the training data acquired by the data acquisition unit to perform machine learning on the first sample image as a positive sample that includes the detection target and the second sample image as a negative sample that does not include the detection target, thereby generating a detection model that takes an image as an input and outputs a detection result for the presence or absence of black smoke; an image acquisition unit that acquires captured images from an imaging unit that images the monitored equipment; and a detection unit that detects the presence or absence of black smoke by inputting the captured images acquired by the image acquisition unit into the detection model generated by the model generation unit.

[0006] Moreover, a detection method according to one aspect of the present invention includes: a data acquisition step of acquiring training data including a first sample image that includes black smoke and a background, a second sample image that includes white smoke and a background but does not include black smoke, and a detection result for the presence or absence of black smoke; a model generation step of using the training data acquired by the data acquisition step to perform machine learning on the first sample image as a positive sample that includes the detection target and on the second sample image as a negative sample that does not include the detection target, thereby generating a detection model that takes an image as an input and outputs a detection result for the presence or absence of black smoke; an image acquisition step of acquiring captured images from an imaging unit that images the equipment to be monitored; and a detection step of detecting the presence or absence of black smoke by inputting the captured images acquired by the image acquisition step into the detection model generated by the model generation step. [Effects of the Invention]

[0007] According to one aspect of the present invention, the presence or absence of black smoke can be detected with high accuracy. [Brief explanation of the drawings]

[0008] [Figure 1]1 is a block diagram showing an example of a configuration of a detection device according to an embodiment of the present invention. [Figure 2] 4 is a flowchart showing an example of a detection process for detecting the presence or absence of black smoke, performed by a control unit included in the detection device shown in FIG. [Figure 3] 2A and 2B are diagrams showing a first sample image and a second sample image acquired by a data acquisition unit of a control unit included in the detection device shown in FIG. [Figure 4] 2 is a diagram showing a captured image generated by the imaging section shown in FIG. 1. FIG. [Figure 5] 10 is a flowchart showing an example of a detection model update process performed by a control unit included in the detection device shown in FIG. [Figure 6] 1. FIG. 4 is a diagram showing an example of a setting screen displayed on a display unit included in the detection device shown in FIG. DETAILED DESCRIPTION OF THE INVENTION

[0009] <Configuration of detection device 1> FIG. 1 is a block diagram showing an example of the configuration of a detection device 1 according to an embodiment of the present invention. The following description not only explains the detection device 1, but also describes a detection method for detecting the presence or absence of black smoke. The detection device 1 is a device that detects the presence or absence of black smoke in monitored equipment, and is, for example, a PC (Personal Computer). The detection device 1 also detects the occurrence or absence of a fire.

[0010] The detection device 1 is used to monitor a target facility for disaster prevention purposes. The detection device 1 and the image capture unit 2 are provided in a detection system that detects the presence or absence of black smoke. As shown in FIG. 1, the detection device 1 includes a communication unit 3, a display unit 4, an operation unit 5, a storage unit 6, and a control unit 7.

[0011] The communication unit 3 is a communication interface that communicates with devices external to the detection device 1. The display unit 4 is a display that displays a screen. The operation unit 5 is used by a user to operate the detection device 1, and is, for example, a mouse and keyboard. The storage unit 6 is, for example, a memory that stores information.

[0012] The control unit 7 controls each unit included in the detection device 1, and is, for example, a processor such as a CPU (Central Processing Unit). The control unit 7 has a data acquisition unit 71, a model generation unit 72, an image acquisition unit 73, a detection unit 74, a measurement unit 75, a setting unit 76, a reception unit 77, a display control unit 78, and a memory control unit 79. The display control unit 78 controls the display on the display unit 4, and the memory control unit 79 controls the storage of the storage unit 6.

[0013] <Configuration of Filming Unit 2> The photographing unit 2 is a camera that is provided in the monitored facility and photographs the monitored facility. The photographing unit 2 is capable of communicating with the communication unit 3 and transmits the generated photographed images to the communication unit 3. A plurality of photographing units 2 may be provided in the monitored facility.

[0014] <Detection process for the presence or absence of black smoke by the control unit 7> Fig. 2 is a flowchart showing an example of a detection process for detecting the presence or absence of black smoke performed by the control unit 7 included in the detection device 1 shown in Fig. 1. Fig. 3 is a diagram showing first sample images P11, P12 and second sample images P21, P22 acquired by the data acquisition unit 71 of the control unit 7 included in the detection device 1 shown in Fig. 1. Fig. 4 is a diagram showing captured images P31, P32 generated by the imaging unit 2 shown in Fig. 1.

[0015] 2, first, in the control unit 7, the data acquisition unit 71 acquires learning data (S1: data acquisition step). The learning data includes a first sample image including black smoke and a background, a second sample image including white smoke and a background but not black smoke, and a detection result of the presence or absence of black smoke. In the learning data, the first sample image is associated with a detection result indicating the presence of black smoke, and the second sample image is associated with a detection result indicating the absence of black smoke.

[0016] Examples of the first sample image include a first sample image P11 indicated by reference numeral 301 in Fig. 3 and a first sample image P12 indicated by reference numeral 302 in Fig. 3. The first sample image P11 includes a factory F1, black smoke B1, and a background BG1. The first sample image P11 is an image showing black smoke B1 emerging from the factory F1 due to a fire inside the factory F1. The background BG1 includes the factory F1, a building next to the factory F1, and a road.

[0017] The first sample image P12 includes a factory F2, black smoke B2, white smoke W1, and a background BG2. The first sample image P12 is an image showing black smoke B2 being emitted from a chimney CP1 of the factory F2 due to a fire inside the factory F2.

[0018] The first sample image P12 also shows white smoke W1 caused by a factor other than the fire, and the white smoke W1 is being emitted from the chimney CP2 of the factory F2. An example of a factor other than the fire is fuel combustion. The background BG2 includes the factory F2, the sky, and clouds. The first sample images P11 and P12 are images acquired by the data acquisition unit 71 from, for example, the Internet via the communication unit 3.

[0019] Examples of the second sample image include a second sample image P21 indicated by reference numeral 303 in FIG. 3 and a second sample image P22 indicated by reference numeral 304 in FIG. 3. The second sample image P21 includes a factory F3, white smoke W2, and a background BG3. The second sample image P21 is an image showing white smoke W2 that has been generated due to a cause other than a fire, and the white smoke W2 is being emitted from a chimney CP3 of the factory F3. The background BG3 includes the factory F3, the sky, and clouds. The second sample image P21 is an image acquired by the data acquisition unit 71 from the Internet via the communication unit 3.

[0020] The second sample image P22 includes a factory F4, white smoke W3, and a background BG4. The second sample image P22 is an image showing white smoke W3 that is being generated due to a cause other than a fire, and the white smoke W3 is being emitted from chimneys CP4 and CP5 of the factory F4. The background BG4 includes the factory F4, the sky, and clouds. The factory F4 is an example of equipment to be monitored by the detection device 1, and the photographing unit 2 may be installed in the factory F4 and photograph the factory F4. The second sample image P22 is, for example, an image acquired by the data acquisition unit 71 from the photographing unit 2 via the communication unit 3.

[0021] In response to a user's operation of the operation unit 5, the data acquisition unit 71 may acquire a plurality of first sample images and a plurality of second sample images from the Internet via the communication unit 3. In addition, in response to a user's operation of the operation unit 5, the plurality of first sample images and a plurality of second sample images may be associated with the detection result of the presence or absence of black smoke. The learning data is obtained by associating the plurality of first sample images and a plurality of second sample images with the detection result of the presence or absence of black smoke.

[0022] After the data acquisition unit 71 acquires the learning data, the model generation unit 72 generates a detection model that takes an image as an input and outputs a detection result of the presence or absence of black smoke (S2: model generation step). Specifically, the model generation unit 72 uses the learning data acquired by the data acquisition unit 71 to perform machine learning on a first sample image as a positive sample that includes the detection target. In addition, the model generation unit 72 performs machine learning on a second sample image as a negative sample that does not include the detection target.

[0023] The negative samples may include things other than the detection target. The model generation unit 72 generates a detection model by performing machine learning on the first sample image and the second sample image.

[0024] 3, the first sample image P11 includes black smoke B1 as the detection target, and the first sample image P12 includes black smoke B2 as the detection target. The second sample images P21 and P22 do not include the black smoke that is the detection target. The storage control unit 79 stores the detection model generated by the model generation unit 72 in the storage unit 6.

[0025] After the model generation unit 72 generates the detection model, the image acquisition unit 73 acquires the captured image from the imaging unit 2 via the communication unit 3 (S3: image acquisition step). Note that the image acquisition unit 73 may acquire the captured image from the imaging unit 2 at predetermined intervals. The captured image acquired by the image acquisition unit 73 is a video or a still image.

[0026] Examples of the photographed images include photographed image P31 indicated by reference numeral 401 in Fig. 4 and photographed image P32 indicated by reference numeral 402 in Fig. 4. Photographed image P31 includes a factory F4, white smoke W4, and a background BG5. Photographed image P31 is an image showing white smoke W4 that has been generated due to a cause other than a fire, and the white smoke W4 is being emitted from chimneys CP4 and CP5 of the factory F4. The background BG5 includes the factory F4, the sky, and clouds.

[0027] The photographed image P32 includes a factory F4, black smoke B3, white smoke W5, and a background BG6. The photographed image P32 is an image showing black smoke B3 being emitted from a chimney CP4 of the factory F4 due to a fire inside the factory F4. In the photographed image P32, white smoke W5, which is caused by a factor other than the fire, is being emitted from a chimney CP5 of the factory F4. The background BG6 includes the factory F4, the sky, and clouds.

[0028] The detection unit 74 determines whether black smoke is present in the photographed image acquired by the image acquisition unit 73 (S4). Specifically, the detection unit 74 detects the presence or absence of black smoke by inputting the photographed image acquired by the image acquisition unit 73 into a detection model stored in the storage unit 6 (detection step). The detection model stored in the storage unit 6 is a detection model generated by the model generation unit 72.

[0029] For example, when the detection unit 74 inputs the photographed image P31 acquired by the image acquisition unit 73 into the detection model, the detection unit 74 determines that there is no black smoke in the photographed image P31. On the other hand, when the detection unit 74 inputs the photographed image P32 acquired by the image acquisition unit 73 into the detection model, the detection unit 74 determines that there is black smoke B3 in the photographed image P32.

[0030] If the detection unit 74 determines that there is no black smoke in the captured image (NO in S4), the processing of the control unit 7 returns to step S3. On the other hand, if the detection unit 74 determines that there is black smoke in the captured image (YES in S4), the memory control unit 79 stores the captured image acquired by the image acquisition unit 73 in the memory unit 6 (S5).

[0031] At this time, the memory control unit 79 stores the captured image acquired by the image acquisition unit 73 in a first folder in the memory unit 6. Furthermore, the display control unit 78 causes the display unit 4 to display warning information indicating a warning. The display control unit 78 may also cause the display unit 4 to display a message indicating that black smoke is present in the captured image. Furthermore, an alarm control unit (not shown) included in the control unit 7 causes an alarm unit (not shown) included in the detection device 1 to issue a warning. The alarm control unit may also cause the alarm unit to issue a message indicating that black smoke is present in the captured image. The alarm unit is a sound output unit that outputs sound.

[0032] Next, the measurement unit 75 measures the elapsed time from the time when the memory control unit 79 stored the captured image in the memory unit 6, and determines whether a predetermined time has elapsed from the time when the memory control unit 79 stored the captured image in the memory unit 6 (S6).

[0033] If the measurement unit 75 determines that the predetermined time has not elapsed since the memory control unit 79 stored the captured image in the memory unit 6 (NO in S6), the process returns to step S6. On the other hand, if the measurement unit 75 determines that the predetermined time has elapsed since the memory control unit 79 stored the captured image in the memory unit 6 (YES in S6), the process of the control unit 7 returns to step S3.

[0034] In the above, the first sample image includes not only black smoke but also the background, and the second sample image includes not only white smoke but also the background. Therefore, by inputting the captured image into the detection model, the detection unit 74 can accurately detect the presence or absence of black smoke by taking into account not only the black or white smoke in the captured image but also the background of the captured image.

[0035] Furthermore, the model generation unit 72 performs machine learning on the first sample image as a positive sample, and on the second sample image as a negative sample. This reduces the likelihood of the detection unit 74 mistakenly detecting white smoke as black smoke, and the detection unit 74 can accurately detect the presence or absence of black smoke because it can distinguish between black smoke and white smoke. Furthermore, the user does not need to constantly monitor the monitored equipment, and the detection device 1 can constantly monitor the monitored equipment.

[0036] The detection unit 74 inputs the captured image into the detection model and detects the presence or absence of black smoke, thereby enabling early detection of black smoke occurring in the monitored equipment. This allows the user to respond quickly to a fire and reduce damage caused by the fire. Furthermore, when the weather conditions are bad or when it is currently nighttime, it is difficult for the user to determine the presence or absence of black smoke. Even in such cases, the detection unit 74 can detect the presence or absence of black smoke.

[0037] The learning data acquired by the data acquisition unit 71 in step S1 may include a second sample image that includes the monitored equipment as a background. For example, the learning data includes the second sample image P22 shown by reference numeral 304 in Fig. 3. The second sample image P22 includes a factory F4 as a background and was captured by the factory F4 as an example of the monitored equipment.

[0038] Therefore, the captured image input to the detection model in step S4 is an image of the monitored facility. Therefore, the model generation unit 72 performs machine learning using the second sample image, which includes the monitored facility as the background, as a negative sample, and the detection unit 74 can accurately detect the presence or absence of black smoke while taking into account the background of the monitored facility included in the captured image.

[0039] The learning data acquired by the data acquisition unit 71 in step S1 may include first sample images generated by a generation AI (Artificial Intelligence) for first sample images including the monitored equipment as a background. In addition, in the learning data, the number of first sample images including the monitored equipment as a background is greater than the number of first sample images not including the monitored equipment. Furthermore, in the learning data, the number of second sample images including the monitored equipment as a background is greater than the number of second sample images not including the monitored equipment.

[0040] As a result, in step S4 described below, the detection unit 74 places more importance on the monitored equipment as the background than on anything other than the monitored equipment. In addition, the detection unit 74 uses the captured image generated by the imaging unit 2 that captures the monitored equipment, and therefore can accurately detect the presence or absence of black smoke.

[0041] <Detection Model Update Process by Control Unit 7> Fig. 5 is a flowchart showing an example of a detection model update process performed by the control unit 7 included in the detection device 1 shown in Fig. 1. The control unit 7 executes the update process shown in Fig. 5 in parallel with the detection process shown in Fig. 2. If the answer is YES in step S4 shown in Fig. 2, the display control unit 78 in the control unit 7 causes the display unit 4 to display the captured image acquired by the image acquisition unit 73 in step S3 shown in Fig. 2 (S11). If the answer is YES in step S4, this means that the detection unit 74 has detected black smoke.

[0042] In step S11, for example, consider a case where the display control unit 78 displays the captured image P32 on the display unit 4, as indicated by reference numeral 402 in Fig. 4. In this case, the display control unit 78 displays, in the captured image P32, a portion determined by the detection unit 74 in step S4 to contain black smoke, with a frame L1.

[0043] When the display control unit 78 causes the display unit 4 to display the captured image, the receiving unit 77 receives (S12) a selection of whether the detection result of the detection unit 74 detects the presence or absence of black smoke for the captured image displayed on the display unit 4 is appropriate or not. For example, when the receiving unit 77 receives a selection of whether the detection result of the presence or absence of black smoke is appropriate or not, the display control unit 78 causes the display unit 4 to display selection items for receiving a selection of whether the detection result of the presence or absence of black smoke is appropriate or not.

[0044] The user can select an option displayed on the display unit 4 by operating the operation unit 5. In this way, the reception unit 77 receives a selection of whether or not the detection result of the presence or absence of black smoke is appropriate, based on the user's operation of the operation unit 5.

[0045] The accepting unit 77 determines whether a selection that the detection result of the presence or absence of black smoke by the detecting unit 74 is inappropriate has been accepted (S13). When the accepting unit 77 determines that a selection that the detection result of the presence or absence of black smoke is appropriate has been accepted (NO in S13), the control unit 7 ends the processing shown in FIG.

[0046] On the other hand, if the accepting unit 77 determines that the selection that the detection result of the presence or absence of black smoke is inappropriate has been accepted (YES in S13), the storage control unit 79 stores the captured image acquired by the image acquisition unit 73 in step S3 shown in FIG. 2 in the storage unit 6 (S14). At this time, the storage control unit 79 stores the captured image in a second folder in the storage unit 6. The second folder is a folder different from the first folder in which the captured image is stored in step S5, and is a folder for storing the captured image to be used for further machine learning.

[0047] After the memory control unit 79 stores the captured image in the memory unit 6, the reception unit 77 receives parameter settings (S15). Specifically, the reception unit 77 receives settings of a learning rate and a number of learning iterations in machine learning as parameters in response to a user's operation of the operation unit 5. The learning rate is a parameter indicating how much the model generation unit 72 changes the weight parameters at one time in optimizing machine learning in step S16, which will be described later. The number of learning iterations is the number of times the model generation unit 72 executes machine learning in step S16, which will be described later.

[0048] After the reception unit 77 receives the parameter settings, the model generation unit 72 updates the detection model by again performing machine learning on the captured image stored in the second folder in the storage unit 6 in step S14 (S16). Here, if the reception unit 77 receives a selection in step S13 that the detection result for the presence or absence of black smoke is inappropriate, the model generation unit 72 updates the detection model.

[0049] In step S16, the model generation unit 72 performs machine learning again on the learning data acquired by the data acquisition unit 71 in step S1 shown in Fig. 2 and the captured image stored in the second folder in step S14. The model generation unit 72 also associates the captured image stored in the second folder in step S14 with a detection result opposite to the detection result of the presence or absence of black smoke by the detection unit 74 in step S4 shown in Fig. 2, and performs machine learning again.

[0050] The model generation unit 72 performs machine learning again based on the learning rate and the number of learning times received in step S15 by the reception unit 77. The detection model updated in step S16 is used again by the detection unit 74 in step S4 shown in FIG.

[0051] In the above, when black smoke is detected by the detection unit 74, the user can check from the captured image displayed on the display unit 4 whether the detection of black smoke by the detection unit 74 is accurate, and select whether the detection result is appropriate. Furthermore, if a selection that the detection result is inappropriate is received, the model generation unit 72 updates the detection model by machine learning the captured image. This can improve the accuracy of black smoke detection by the detection unit 74 using the detection model.

[0052] Furthermore, by having the model generation unit 72 automatically update the detection model, the update of the detection model can be completed more quickly than if the user operated the operation unit 5 to update the detection model.

[0053] <Configuration of the setting screen SP1> Fig. 6 is a diagram showing an example of a setting screen SP1 displayed on the display unit 4 included in the detection device 1 shown in Fig. 1. The display control unit 78 causes the display unit 4 to display the setting screen SP1 in response to a user's operation of the operation unit 5. As shown in Fig. 6, the setting screen SP1 includes selection items SL1, SL2, and SL3, a setting item SE1, and a determination button BT1. The reception unit 77 receives the selection of the selection items SL1 to SL3 and the setting of the setting item SE1 through the user's operation of the operation unit 5.

[0054] The user can select either selection item SL1 or SL2 by operating the operation unit 5. Selection item SL1 is an item for selecting a folder in which a created model is saved as a folder in which a detection model is saved. The created model is a model created by performing machine learning on a default model (described later) in step S16 shown in FIG. 5.

[0055] Selection item SL2 is an item for selecting a folder in which a default model is saved as a folder in which a detection model is saved. The default model is a model created by a user operating operation unit 5. The folder in which the detection model is saved is stored in storage unit 6.

[0056] The user can select a detection model from selection item SL3 by operating operation unit 5. Selection item SL3 is an item for selecting a detection model. When selection item SL1 is selected, selection item SL3 becomes an item for selecting at least one created model saved in a folder in storage unit 6. When selection item SL3 is selected, at least one created model is displayed.

[0057] When selection item SL2 is selected, selection item SL3 becomes an item for selecting at least one default model saved in a folder in storage unit 6. When selection item SL3 is selected, at least one default model is displayed.

[0058] The user can set a threshold value that serves as a detection standard for detecting the presence or absence of black smoke by the detection unit 74 from the setting item SE1 by operating the operation unit 5. The setting item SE1 is an item for setting the threshold value.

[0059] The threshold value of setting item SE1 is a numerical value indicating the strictness of the detection unit 74's determination of black smoke in step S4 shown in Fig. 2. As the threshold value increases, the possibility that the detection unit 74 will detect black smoke decreases, and as the threshold value decreases, the possibility that the detection unit 74 will detect black smoke increases. In addition, the user can operate the operation unit 5 to press the determination button BT1 to cause the control unit 7 to execute the detection process shown in Fig. 2.

[0060] <Variation 1> 2, consider the second sample images included in the learning data acquired by the data acquisition unit 71. Among the second sample images included in the learning data, the second sample images including the monitored equipment may include second sample images including multiple monitored equipment in different environments.

[0061] Specifically, one second sample image includes multiple second sample images, and the environments of the monitored equipment differ among the multiple second sample images. For example, one second sample image may include multiple second sample images of factory F4 taken under different weather conditions. In this case, for example, one second sample image may include a second sample image of factory F4 taken under sunny weather, a second sample image of factory F4 taken under cloudy weather, and a second sample image of factory F4 taken under rainy weather.

[0062] Furthermore, one second sample image may include multiple second sample images of factory F4 taken at different times. For example, one second sample image may include a second sample image of factory F4 taken in the morning, a second sample image of factory F4 taken in the afternoon, and a second sample image of factory F4 taken in the evening.

[0063] One second sample image includes multiple second sample images, that is, one second sample image is divided into multiple regions, and a second sample image is arranged in each of the multiple regions.

[0064] As described above, the second sample image includes multiple monitored facilities in different environments. Therefore, in step S2 shown in Fig. 2, the model generation unit 72 performs machine learning on the second sample image including multiple monitored facilities in different environments as a negative sample. This enables the detection unit 74 to more accurately detect the presence or absence of black smoke by taking into account the background of the monitored facilities and also the differences in the environments.

[0065] The learning data may include a plurality of second sample images, and the environments of the monitored equipment may differ among the plurality of second sample images. For example, the learning data may include a plurality of second sample images of factory F4 taken under different weather conditions. The learning data may also include a plurality of second sample images of factory F4 taken at different times.

[0066] <Variation 2> 2, the setting unit 76 may change the detection criteria by which the detection unit 74 detects the presence or absence of black smoke, depending on the environment at the time when the captured image is output by the imaging unit 2. Specifically, the setting unit 76 may change the detection criteria depending on the imaging time, which is the time at which the captured image is output by the imaging unit 2.

[0067] In this case, the setting unit 76 may change the threshold value that serves as the detection criterion for the setting item SE1 shown in FIG. 6, or may change a detection criterion other than the threshold value for the setting item SE1. The data acquiring unit 71 constantly acquires the current time from the Internet via the communication unit 3. The setting unit 76 identifies the shooting time based on the current time acquired by the data acquiring unit 71 and the time point at which the captured image is output by the imaging unit 2. The setting unit 76 may also identify the shooting time from information about the shooting time included in the data of the captured image output by the imaging unit 2.

[0068] The setting unit 76 may also change the detection criteria depending on the weather conditions at the time when the captured image is output by the imaging unit 2. In this case, the data acquisition unit 71 acquires weather condition information from the Internet via the communication unit 3 at any time. The setting unit 76 identifies the weather conditions at the time when the captured image is output by the imaging unit 2, based on the weather condition information acquired by the data acquisition unit 71 and the time when the captured image is output by the imaging unit 2.

[0069] Here, the brightness of the captured image changes depending on the environment. Therefore, by having the setting unit 76 change the detection criteria depending on the environment, the detection unit 74 can accurately detect the presence or absence of black smoke depending on the environment. As described above, the setting unit 76 sets the detection criteria.

[0070] <Variation 3> 2, the setting unit 76 may change the brightness of the captured image output from the capturing unit 2 in accordance with the environment at the time when the captured image is output by the capturing unit 2. Specifically, the setting unit 76 may change the brightness of the captured image in accordance with the capturing time, which is the time when the captured image is output by the capturing unit 2. The setting unit 76 may also change the brightness of the captured image in accordance with the weather conditions at the time when the captured image is output by the capturing unit 2. As described above, the setting unit 76 sets the brightness of the captured image.

[0071] <Variation 4> The control unit 7 may further include an imaging control unit (not shown) that controls the imaging unit 2. The imaging control unit controls the zoom magnification of the imaging unit 2. Furthermore, if multiple imaging units 2 are provided in the monitored facility, the imaging control unit may switch between the multiple imaging units 2 to output the captured image to the communication unit 3. In this case, in step S3 shown in FIG. 2, the image acquisition unit 73 acquires the captured image from the imaging unit 2 switched by the imaging control unit.

[0072] Furthermore, the photography control unit may cause the multiple photography units 2 to output the multiple photographed images output from the multiple photography units 2 as a single photographed image to the communication unit 3. The single photographed image includes the multiple photographed images output from the multiple photography units 2. In other words, with one photographed image divided into multiple areas, a photographed image output from the photography unit 2 is arranged in each of the multiple areas. In this case, in step S3 shown in FIG. 2, the image acquisition unit 73 acquires the multiple photographed images output from the multiple photography units 2 as a single photographed image.

[0073] <Variation 5> The display control unit 78 may constantly display the captured image output from the imaging unit 2 on a display unit separate from the display unit 4. The separate display unit is provided in a detection system that detects the presence or absence of black smoke. This allows the user to monitor the monitored equipment on the separate display unit while the user is monitoring the monitored equipment. Furthermore, when multiple imaging units 2 are provided in the monitored equipment, the display control unit 78 may display multiple captured images output from the multiple imaging units 2 on the separate display unit. In this case, the display area of ​​the separate display unit is divided into multiple areas, and the captured image output from the imaging unit 2 is arranged in each of the multiple areas.

[0074] <Variation 6> 2, consider a case where the image acquisition unit 73 acquires a plurality of captured images from a plurality of imaging units 2. In this case, in step S11 shown in FIG. 5, the display control unit 78 may cause the display unit 4 to display the plurality of captured images acquired by the image acquisition unit 73 in step S3. At this time, the display control unit 78 causes the display unit 4 to display a list of the plurality of captured images. Furthermore, in step S12, the accepting unit 77 accepts a selection of whether the detection result by the detection unit 74 as to whether black smoke is present or not, for each of the plurality of captured images displayed in a list on the display unit 4.

[0075] <Variation 7> 5, the reception unit 77 may receive, through a user operation on the operation unit 5, a designation of a location where black smoke is present in the captured image displayed on the display unit 4 in step S11. In this case, in step S16, the model generation unit 72 associates the captured image stored in the second folder in step S14 with the designation of the location where black smoke is present received by the reception unit 77, and performs machine learning again.

[0076] This allows the user to specify the location where black smoke is present, allowing the model generation unit 72 to efficiently perform machine learning, thereby further improving the accuracy with which the detection unit 74 detects the presence or absence of black smoke.

[0077] <Software implementation example> The functions of the detection device 1 (hereinafter referred to as the "device") can be realized by a detection program that causes a computer to function as the device, and a detection program that causes a computer to function as each control block of the device (particularly each part included in the control unit 7).

[0078] In this case, the device includes a computer having at least one control device (e.g., a processor) and at least one storage device (e.g., a memory) as hardware for executing the detection program. The control device and storage device execute the detection program, thereby realizing the functions described in the above embodiments.

[0079] The detection program may be stored non-transitory on one or more computer-readable storage media. The storage media may or may not be included in the device. In the latter case, the detection program may be provided to the device via any wired or wireless transmission medium.

[0080] Furthermore, some or all of the functions of the control blocks can be realized by logic circuits. For example, an integrated circuit in which a logic circuit that functions as each of the control blocks is formed is also included in the scope of the present invention. In addition, the functions of the control blocks can also be realized by, for example, a quantum computer.

[0081] <Summary> The detection device according to a first aspect of the present invention comprises: a data acquisition unit that acquires learning data including a first sample image that includes black smoke and a background, a second sample image that includes white smoke and a background but does not include black smoke, and a detection result for the presence or absence of black smoke; a model generation unit that uses the learning data acquired by the data acquisition unit to perform machine learning on the first sample image as a positive sample that includes the detection target and the second sample image as a negative sample that does not include the detection target, thereby generating a detection model that takes an image as an input and outputs a detection result for the presence or absence of black smoke; an image acquisition unit that acquires captured images from an imaging unit that images the equipment to be monitored; and a detection unit that detects the presence or absence of black smoke by inputting the captured images acquired by the image acquisition unit into the detection model generated by the model generation unit.

[0082] A detection device according to aspect 2 of the present invention may be configured such that, in aspect 1 above, the learning data acquired by the data acquisition unit includes the second sample image in which the monitored equipment is included as a background.

[0083] A detection device according to aspect 3 of the present invention may be configured such that, in aspect 2 above, the second sample image containing the monitored equipment includes second sample images containing multiple monitored equipment in different environments.

[0084] The detection device according to aspect 4 of the present invention may be configured in any one of aspects 1 to 3 above, further comprising a setting unit that changes the detection criteria by which the detection unit detects the presence or absence of black smoke depending on the environment at the time the captured image is output by the imaging unit.

[0085] A detection device according to a fifth aspect of the present invention is in any of the first to fourth aspects described above, and further includes a display control unit that, when the detection unit detects black smoke, causes the captured image acquired by the image acquisition unit to be displayed on a display unit, and a reception unit that receives a selection of whether the detection result of the detection unit for the presence or absence of black smoke is appropriate for the captured image displayed on the display unit, and the model generation unit may be configured to update the detection model by machine learning the captured image acquired by the image acquisition unit when the reception unit receives a selection that the detection result for the presence or absence of black smoke is inappropriate.

[0086] A detection method according to a sixth aspect of the present invention includes: a data acquisition step of acquiring training data including a first sample image that includes black smoke and a background, a second sample image that includes white smoke and a background but does not include black smoke, and a detection result for the presence or absence of black smoke; a model generation step of using the training data acquired by the data acquisition step to perform machine learning on the first sample image as a positive sample that includes the detection target and on the second sample image as a negative sample that does not include the detection target, thereby generating a detection model that takes an image as an input and outputs a detection result for the presence or absence of black smoke; an image acquisition step of acquiring captured images from an imaging unit that images the equipment to be monitored; and a detection step of detecting the presence or absence of black smoke by inputting the captured images acquired by the image acquisition step into the detection model generated by the model generation step.

[0087] The detection device according to each aspect of the present invention may be realized by a computer. In this case, the detection program of the detection device that causes the computer to operate as each part (software element) of the detection device to realize the detection device on a computer, and the computer-readable recording medium on which the program is recorded, also fall within the scope of the present invention.

[0088] <Additional Notes> The present invention is not limited to the above-described embodiments, and various modifications are possible within the scope of the claims. Configurations obtained by appropriately combining multiple technical means disclosed in the embodiments are also included in the technical scope of the present invention. [Explanation of symbols]

[0089] 1 Detector 2 Photography unit 4 Display unit 71 Data acquisition unit 72 Model generation unit 73 Image acquisition unit 74 Detection unit 76 Setting unit 77 Reception unit 78 Display control unit P11, P12 First sample image P21, P22 Second sample image P31, P32 Photographed image F4 Factory (monitored equipment) B1, B2, B3: Black smoke W1, W2, W3: White smoke BG1, BG2, BG3, BG4 Background

Claims

1. a data acquisition unit that acquires learning data including a first sample image including black smoke and a background, a second sample image including white smoke and a background but not black smoke, and a detection result of the presence or absence of black smoke; a model generation unit that uses the learning data acquired by the data acquisition unit to perform machine learning on the first sample image as a positive sample including the detection target and the second sample image as a negative sample not including the detection target, thereby generating a detection model that takes an image as an input and outputs a detection result of the presence or absence of black smoke; an image acquisition unit that acquires a captured image from an image capture unit that captures an image of the monitored facility; a detection unit that detects the presence or absence of black smoke by inputting the captured image acquired by the image acquisition unit into the detection model generated by the model generation unit.

2. The detection device according to claim 1 , wherein the learning data acquired by the data acquisition unit includes the second sample image in which the monitored facility is included as a background.

3. The detection device according to claim 2, characterized in that the second sample images including the monitored equipment include second sample images including a plurality of monitored equipment in different environments.

4. 3. The detection device according to claim 1, further comprising a setting unit that changes a detection criterion for detecting the presence or absence of black smoke by the detection unit in accordance with an environment at the time when the captured image is output by the imaging unit.

5. a display control unit that displays the captured image acquired by the image acquisition unit on a display unit when the detection unit detects black smoke; and a receiving unit that receives a selection of whether or not a detection result of the presence or absence of black smoke by the detecting unit is appropriate for the captured image displayed on the display unit, 3. The detection device according to claim 1, wherein, when the reception unit receives a selection that the detection result of the presence or absence of black smoke is inappropriate, the model generation unit updates the detection model by performing machine learning on the captured image acquired by the image acquisition unit.

6. a data acquisition step of acquiring learning data including a first sample image including black smoke and a background, a second sample image including white smoke and a background but not black smoke, and a detection result of the presence or absence of black smoke; a model generation step of using the learning data acquired by the data acquisition step to perform machine learning on the first sample image as a positive sample including the detection target and the second sample image as a negative sample not including the detection target, thereby generating a detection model that uses an image as an input and outputs a detection result of the presence or absence of black smoke; an image acquisition step of acquiring a captured image from an image capturing unit that captures an image of the monitored facility; a detection step of detecting the presence or absence of black smoke by inputting the photographed image acquired in the image acquisition step into the detection model generated in the model generation step.

7. A detection program for causing a computer to function as the detection device according to claim 1, the detection program causing a computer to function as the data acquisition unit, the model generation unit, the image acquisition unit, and the detection unit.

Images