Gas concentration feature quantity estimation device, gas concentration feature quantity estimation method, program, and gas concentration feature quantity inference model generation device

Abstract

The present invention is provided with: inspection data acquisition units 212, 211 for respectively acquiring a gas temperature value and a time-series pixel group inspection data item in which the number of vertical pixels and the number of horizontal pixels are not less than 2 and which is regionally extracted from an inspection data item of a gas distribution video indicating the existence region of gas in a space; and an estimation unit 2151 for calculating a gas concentration feature quantity estimation value corresponding to the time-series pixel group inspection data item acquired by the inspection data acquisition unit, by using an inference model obtained by machine learning using, as training data, time-series pixel group training data items which have the same size as the size of the time-series pixel group inspection data item and which are regionally extracted from training data items of gas distribution videos and gas concentration feature quantity values and gas temperature values corresponding to the time-series pixel group training data items.

Description

[Technical Field]

[0001] The present disclosure relates to a gas concentration feature estimation device and a gas concentration feature estimation method, and more particularly to a gas concentration feature estimation method based on an infrared image. [Background technology]

[0002] In facilities that use gas (hereinafter referred to as "gas facilities"), such as production facilities that produce natural gas and oil, production plants that use gas to produce chemical products, gas transmission facilities, petrochemical plants, thermal power plants, steel-related facilities, etc., the risk of gas leakage due to deterioration of the facilities over time or operational errors is recognized, and gas detection devices are installed to minimize gas leakage.

[0003] In gas detection, in addition to gas detection devices that utilize the change in the electrical properties of the detection probe when gas molecules come into contact with the probe, in recent years optical gas leak detection methods have been adopted that utilize the infrared absorption properties of gas to capture infrared video to detect gas leaks in the inspection area.

[0004] The infrared video gas detection method has the advantage that it can visualize gas through video, making it easier to detect the gas flow and other release conditions and leak locations compared to conventional detection probe methods. Another advantage is that the state of the leaked gas is recorded as video, which can be used as evidence of the occurrence of the gas leak and its repair.

[0005] As an example of this type of infrared gas detection device, Patent Document 1 discloses a technology for detecting the amplitude characteristics of the time-series brightness change of each pixel in an infrared image of a monitored object, thereby identifying the background temperature when gas is present and the background temperature when gas is not present, and estimating the concentration-thickness product. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] International Publication No. 2017 / 104617 Summary of the Invention [Problem to be solved by the invention]

[0007] Accurate estimation of the concentration thickness product is required for estimating the gas flow rate in gas facilities.

[0008] However, in gas plants and petrochemical plants, mechanical vibrations occurring during operation can cause mechanical vibrations in the imaging target, resulting in captured infrared images containing mechanical vibration noise. Alternatively, mechanical vibration noise can be transmitted from gas equipment to the imaging device, resulting in captured infrared images being affected by the mechanical vibration noise. In such cases, pixel-by-pixel brightness changes associated with the vibration noise can be detected even when no gas is present. Therefore, a particular problem with non-contact gas detection using infrared images is that mechanical vibration noise present in the measurement environment can affect the accuracy of the calculated gas concentration feature values.

[0009] The aspects of the present disclosure have been made in consideration of the above-mentioned problems, and aim to provide a gas concentration feature estimation device, a gas concentration feature estimation method, a program, and a gas concentration feature inference model generation device that can reduce the impact of mechanical vibration noise on measurements and accurately detect features representing gas concentrations in space from infrared gas distribution dynamic images. [Means for solving the problem]

[0010] A gas concentration feature estimation device according to one aspect of the present disclosure is characterized by comprising: an inspection data acquisition unit that acquires time-series pixel group inspection data of a gas distribution moving image having two or more vertical and horizontal pixels each, the time-series pixel group inspection data being an area extracted from inspection data of a gas distribution moving image representing a gas presence area in space, and a temperature value of the gas; and an estimation unit that calculates the gas concentration feature corresponding to the time-series pixel group inspection data acquired by the inspection data acquisition unit using an inference model machine-learned using as training data time-series pixel group teacher data of a gas distribution moving image of the same size as the time-series pixel group inspection data and gas temperature values ​​and gas concentration feature values ​​corresponding to the time-series pixel group teacher data. [Effects of the Invention]

[0011] According to one aspect of the present disclosure, it is possible to provide a gas concentration feature estimation device, a gas concentration feature estimation method, a program, and a gas concentration feature inference model generation device that can reduce the impact of mechanical vibration noise on measurement and accurately detect features representing gas concentration in space from infrared gas distribution dynamic images.

[0012] This makes it possible to accurately detect features representing the gas concentration in the space even when mechanical vibration noise is observed in the gas equipment being inspected or in an environment where mechanical vibration noise is transmitted from the gas equipment to the imaging device. [Brief explanation of the drawings]

[0013] [Figure 1] 1 is a schematic configuration diagram of a gas concentration characteristic quantity estimation system 1 according to an embodiment. [Figure 2] 1 is a schematic diagram showing the relationship between the gas visualization imaging device 10 and an object 300 to be inspected. [Figure 3] FIG. 2 is a diagram showing the circuit configuration of a gas concentration characteristic quantity estimation device 20. [Figure 4] 1A is a functional block diagram of the control unit 21, and FIG. 1B is a functional block diagram of a machine learning unit 2151 in the control unit 21. FIG. [Figure 5]1A is a diagram showing the appearance of time-series pixel group inspection data of a gas distribution moving image obtained by extracting a region from inspection data of the gas distribution moving image; FIG. 1B is a diagram showing the appearance of time-series pixel group teacher data of a gas distribution moving image obtained by extracting a region from teacher data of the gas distribution moving image; and FIG. 1C is a schematic diagram showing the appearance of gas concentration features estimated from the time-series pixel group inspection data of the gas distribution moving image. [Figure 6] FIG. 1(a) is a schematic diagram showing the state of time-series pixel group data of a gas distribution moving image when there is no influence of mechanical vibration noise, and FIG. 1(b) is a schematic diagram showing the state of time-series pixel group data of a gas distribution moving image when there is an influence of mechanical vibration noise. [Figure 7] (a) is a schematic diagram showing the time change in brightness value in time-series pixel group data of a gas distribution moving image when affected by mechanical vibration noise, and (b) is a schematic diagram showing the position change in brightness value in time-series pixel group data. [Figure 8] FIG. 10(a) is a schematic diagram showing the time change in brightness values ​​representing gas distribution in time-series pixel group data of a gas distribution moving image, and FIG. 10(b) is a schematic diagram showing the position change in brightness values ​​representing gas distribution in time-series pixel group data. [Figure 9] 10(a) to 10(d) are diagrams showing a method for calculating a concentration-thickness product from light absorptance. [Figure 10] 10 is a flowchart showing the operation of the gas concentration feature quantity estimation device 20 in the learning phase. [Figure 11] 10 is a flowchart showing the operation of the gas concentration feature quantity estimation device 20 in the operation phase. [Figure 12] 10A is an example of a moving image of gas distribution in an inspection object, FIG. 10B is an example of an image of the inspection object in a light absorption image, and FIG. 10C is an example of a processed result of the light absorption image. [Figure 13] FIG. 2 is a diagram showing the circuit configuration of a machine learning data generation device 30. [Figure 14] FIG. 2 is a functional block diagram of a control unit 31. [Figure 15] 10A is a schematic diagram showing the data structure of the three-dimensional structure data DTstr, and FIG. 10B is a schematic diagram showing the data structure of the three-dimensional gas distribution image data DTgas. [Figure 16] FIG. 10 is a schematic diagram for explaining an outline of a concentration-thickness product calculation method in two-dimensional single-viewpoint gas distribution image conversion processing. [Figure 17] FIG. 10 is a schematic diagram for explaining an outline of a concentration-thickness product calculation method under the condition that a structure is present, in two-dimensional single-view gas distribution image conversion processing. [Figure 18] (a) is a schematic diagram showing an overview of density-thickness product image data, (b) is a schematic diagram for explaining an overview of a density-thickness product calculation method in light absorptance image conversion processing, and (c) is a schematic diagram showing an overview of light absorptance image data. [Figure 19] (a) is a schematic diagram showing an overview of the 3D data of the structure, (b) is a schematic diagram showing an overview of the extracted background location data, and (c) is a conceptual diagram showing a method for generating background image data based on the background location data. [Figure 20] 1A is a schematic diagram showing an outline of background image data, FIG. 1B is a schematic diagram showing an outline of light absorptance image data, and FIG. 1C is a schematic diagram showing an outline of light intensity image data. [Figure 21] 10 is a flowchart showing an outline of a two-dimensional single-viewpoint gas distribution image conversion process. [Figure 22] 10 is a flowchart showing an outline of a background image generation process. [Figure 23] 10 is a flowchart showing an outline of a light intensity image data generation process. DETAILED DESCRIPTION OF THE INVENTION

[0014] <Embodiment> <Configuration of Gas Concentration Feature Estimation System 1> An embodiment of the present disclosure is realized as a gas concentration feature system 1 that analyzes gas leaks from gas leak inspection images of gas equipment. The gas concentration feature system 1 according to the embodiment will be described in detail below with reference to the drawings.

[0015] Fig. 1 is a schematic configuration diagram of a gas concentration feature system 1 according to an embodiment. As shown in Fig. 1, the gas concentration feature system 1 is composed of a plurality of gas visualization imaging devices 10, a gas concentration feature estimation device 20, a machine learning data generation device 30, and a storage means 40, all connected to a communication network N.

[0016] The communication network N is, for example, the Internet, and the gas visualization imaging device 10, the gas concentration feature estimation device 20, the multiple machine learning data generation devices 30, and the storage means 40 are connected so that they can exchange information with each other.

[0017] <Gas visualization imaging device 10> The gas visualization imaging device 10 is a device or system that uses infrared rays to capture an image of a monitoring target and provides an infrared image in which the gas is visualized, and comes in various forms, such as a stationary type installed in a gas facility or the like, a portable type that can be carried by an inspector, a type mounted on a drone, etc. The gas visualization imaging device 10 includes, for example, an imaging means (not shown) consisting of an infrared camera that detects infrared rays and captures an image, and an interface circuit (not shown) that outputs the image to a communication network N.

[0018] An infrared camera is an infrared imaging device that generates an infrared image based on infrared light and outputs the infrared image to the outside. For example, as described in publicly known documents (e.g., JP 2012-58093 A), an infrared camera can be used in a leak gas visualization imaging device that visualizes gas leaks from gas equipment by taking infrared videos of gas in the air using the infrared absorption properties of gas.

[0019] Images taken by infrared cameras are generally used to detect hydrocarbon gases. For example, infrared cameras equipped with an image sensor sensitive to at least a portion of the infrared wavelength range from 3 μm to 5 μm can detect hydrocarbon gases such as methane, ethane, ethylene, and propylene by detecting and imaging infrared light with wavelengths of 3.2 to 3.4 μm. Alternatively, infrared light with wavelengths of 4.52 to 4.67 μm can be used to detect other types of gases, such as carbon monoxide.

[0020] FIG. 2 is a schematic diagram showing the relationship between the gas visualization imaging device 10 and an inspection target 300. As shown in FIG. 2, the gas visualization imaging device 10 is installed so that the inspection target 300 is included in the field of view 310 of the infrared camera. The obtained inspection image is a video signal for transmitting images at, for example, 30 frames per second. The gas visualization imaging device 10 converts the captured image into a predetermined video signal. In this embodiment, the infrared image signal acquired from the infrared camera is restored to an image and processed as a moving image consisting of multiple frames. The image is an infrared photograph of the monitoring target, and has infrared intensity as pixel values.

[0021] Note that if the size of the gas distribution image or the number of frames as a video is excessively large, the amount of calculation for machine learning and machine learning-based determination will increase. In this embodiment, the number of pixels of the gas distribution image is, for example, 320 × 256 pixels, and the number of frames is, for example, 100.

[0022] The gas visualization imaging device detects the presence of gas by capturing changes in the amount of electromagnetic waves emitted from background objects with an absolute temperature of 0 (K) or higher. Changes in the amount of electromagnetic waves are mainly caused by the gas absorbing electromagnetic waves in the infrared range or by the gas itself generating blackbody radiation. The gas visualization imaging device 10 can capture gas leaks as images by photographing the monitored space, allowing for earlier gas leak detection and accurate location of the gas.

[0023] The visualized inspection image is temporarily stored in a memory or the like, and is transferred to and saved in the storage means 40 via the communication network N based on an operation input.

[0024] The gas visualization imaging device 10 is not limited to this, and may be any imaging device that can detect the gas being monitored. For example, if the gas being monitored is a gas that can be detected with visible light, such as water vapor that has turned into white smoke, a general visible light camera may be used.

[0025] <Storage means 40> The storage means 40 is a storage device that stores the inspection images transmitted from the gas visualization imaging device 10, and is configured to include a nonvolatile memory such as a hard disk, and stores the inspection images and light absorptance images in association with the identification information of the gas visualization imaging device 10. An administrator can read out the infrared images from the storage means 40 using an administration terminal (not shown) or the like, and grasp the state of the infrared images to be viewed.

[0026] <Gas concentration feature quantity estimation device 20> The gas concentration feature estimation device 20 is a device that acquires an inspection image of a monitoring target from the gas visualization imaging device 10, estimates gas feature quantities based on the inspection image, and notifies a user of gas detection via a display 24. The gas concentration feature estimation device 20 is realized, for example, as a computer including a general CPU (Central Processing Unit), RAM (Random Access Memory), and programs executed by these. As will be described later, the gas concentration feature estimation device 20 may further include a GPU (Graphics Processing Unit) and RAM as arithmetic units.

[0027] The following describes the configuration of the gas concentration characteristic quantity estimation device 20. Fig. 3 is a diagram showing the circuit configuration of the gas concentration characteristic quantity estimation device 20.

[0028] The gas concentration feature estimation device 20 is a server computer for estimating light absorptance based on an inspection image. The gas concentration feature estimation device 20 reads out an inspection image stored in the storage means 40, or receives an inspection image from the gas visualization imaging device 10, estimates the light absorptance of the inspection image, generates a light absorptance image, and outputs the image to the storage means 40 via the communication network N for storage.

[0029] 3 is a schematic diagram showing the circuit configuration of gas concentration feature quantity estimation device 20. As shown in FIG. 3, gas concentration feature quantity estimation device 20 includes a control unit 21, a communication circuit 22, a storage device 23, a display 24, and an operation input unit 25.

[0030] The display 24 is, for example, a liquid crystal panel, and displays the display screen generated by the control unit 21.

[0031] Operation input unit 25 is an input device through which an operator inputs information to operate gas concentration feature estimation apparatus 20. For example, operation input unit 25 may be realized as an input device such as a keyboard or a mouse, or as a single device that functions as both a display device and an input device, such as a touch panel with a touch sensor disposed on the front surface of display 24.

[0032] The control unit 21 is configured to include a CPU, RAM, and ROM, and the CPU executes a program (not shown) stored in the ROM to realize each function of the gas concentration feature estimation device 20. Specifically, the control unit 21 estimates gas concentration feature values ​​of a gas distribution image based on an inspection image acquired from the communication circuit 22, creates an image of the gas concentration feature values, and outputs the image of the gas concentration feature values ​​to the communication circuit 22. Specifically, the control unit 21 estimates light absorptance or gas concentration thickness product as the gas concentration feature values, creates a light absorptance image or gas concentration thickness product image, and outputs the image to the communication circuit 22.

[0033] [On estimation of gas concentration features] FIG. 4( a ) is a functional block diagram of the control unit 21 .

[0034] 4(a), the gas concentration feature estimation device 20 includes an inspection image acquisition unit 211, a gas temperature value inspection data acquisition unit 212, a teacher image acquisition unit 213, a gas feature teacher data acquisition unit 214, a machine learning unit 2151, a learning model storage and inference unit 2152, and a gas concentration feature output unit 216. The machine learning unit 2151 and the learning model storage and inference unit 2152 constitute a gas concentration feature estimation unit 215.

[0035] The inspection image acquisition unit 211 is a circuit that acquires pixel group inspection data of the gas distribution moving image (hereinafter, may be referred to as "gas distribution pixel group inspection data") from the gas visualization imaging device 10. For example, an image capture board or other device that captures image data into a processing device such as a computer can be used.

[0036] The inspection data of the gas distribution moving image is an infrared image captured by an infrared camera with sensitivity to wavelengths of 3.2 to 3.4 μm. For example, it may be an image that visualizes the gas leak area under inspection, or it may be a luminance signal that indicates the gas presence area in space as a high concentration. The image size may be, for example, 320×256 pixels in length and width. Furthermore, the inspection data of the gas distribution moving image is a moving image including multiple frames of time-series data. Gain adjustment, offset adjustment, image inversion, etc. may be performed as necessary for subsequent processing.

[0037] The pixel group test data is test data in which the number of vertical and horizontal pixels is two or more, and which is an area extracted from the test data of the gas distribution moving image. The pixel group test data may also be composed of pixel groups with a number of pixels less than the number of pixels in a frame of the gas distribution moving image. This is because the smaller the number of pixels in the pixel group test data, the more areas can be extracted from a single test image for learning, reducing the amount of test image data required for learning. Furthermore, a larger number of pixels in the pixel group test data requires preparing various background variations for learning. Specifically, it is preferable that the time-series pixel group test data have a frame pixel count of three to seven, both vertical and horizontal. For example, the pixel group test data may have a vertical x horizontal pixel count of 4 x 4, or N frames (N is a natural number), or N may be approximately 100 frames for 10 seconds at 10 frames per second (fps).

[0038] 5(a) is a diagram showing the appearance of time-series pixel group inspection data of a gas distribution moving image obtained by extracting regions from the inspection data of the gas distribution moving image. When the gas type is a hydrocarbon gas such as methane gas, the inspection data of the gas distribution moving image captured by an infrared camera and the pixel group inspection data will have the appearance shown in FIG. 5(a).

[0039] The gas temperature value test data acquisition unit 212 is a circuit that acquires gas temperature values ​​corresponding to the pixel group test data of the gas distribution image. The captured image is input as a brightness value, and the gas temperature is input as a temperature value measured by a thermometer and converted into a brightness value. However, they may also be input as temperature values. The gas temperature value may be a value obtained by converting the temperature measurement data of the gas atmosphere measured using a thermometer TE when capturing the gas distribution moving image into a brightness value. In this example, as shown in FIG. 5(a), the average value of 4x4 pixels and N frames in the time-series pixel group test data is used. The pixel group test data and gas temperature value test data are images captured and measured under the same conditions and the temperature values ​​related to those images.

[0040] The data set consisting of the pixel group inspection data and the gas temperature value inspection data is output to the inference unit 2152 as inspection target data for the inference unit 2152 .

[0041] The teacher image acquisition unit 213 is a circuit that receives input of time-series pixel group teacher data of a gas distribution moving image obtained by extracting regions from the gas distribution moving image generated by the machine-learning data generation device 30. FIG. 5(b) illustrates the state of time-series pixel group teacher data of a gas distribution moving image obtained by extracting regions from the teacher data of the gas distribution moving image when an optical absorptance image is used as the gas distribution moving image. As shown in FIG. 5(b), the time-series pixel group teacher data is an image having the same format as the pixel group inspection data generated by the gas visualization imaging device 10, and is a moving image including multiple frames of time-series data. Again, the time-series pixel group teacher data preferably has a frame pixel count of 3 to 7 in both the vertical and horizontal directions. For example, the vertical and horizontal pixel count may be 4×4, 10 frames per second (fps) for 10 seconds, and N frames (N is a natural number), where N is, for example, approximately 100 frames.

[0042] Furthermore, learning can be performed more effectively by changing the number of frame pixels in the time-series pixel group training data depending on the magnitude of the vibration noise. For example, with 3x3 pixels, it is possible to learn the effects of vibration noise of less than 1 pixel above, below, left, or right. With 4x4 pixels, it is possible to learn the effects of vibration noise of up to about 1.5 pixels above, below, left, or right. With 7x7 pixels, it is possible to learn the effects of deviations of less than 3 pixels above, below, left, or right.

[0043] The gas feature teacher data acquisition unit 214 is a circuit that acquires, as representative values ​​for the time-series pixel group teacher data, gas temperature values ​​and gas concentration feature teacher data corresponding to time-series pixel group teacher data extracted from regions of the gas distribution moving image generated by the machine learning data generation device 30. The time-series pixel group teacher data, gas temperature values, and gas concentration feature values ​​are images generated under the same conditions and parameters related to the images. As shown in FIG. 5(b), the temperature value may be the average value (average value of N frames) corresponding to the time-series pixel group teacher data.

[0044] The gas temperature value may be a gas temperature value corresponding to the brightness of the time-series pixel group teacher data, and the gas concentration feature may be a light absorptance or a gas concentration thickness product corresponding to the time-series pixel group teacher data.

[0045] Optical absorptance is the rate at which light is absorbed when gas is present in a space, and is expressed as a value between 0 and 1, with a gas-free state being taken as 0. The optical absorptance can be converted into a concentration-thickness product using the gas's spectral absorption coefficient. As shown in Figure 5(b), the optical absorptance and concentration-thickness product may be calculated using the average values ​​(average values ​​for 4 x 4 pixels x N frames) corresponding to the time-series pixel group training data.

[0046] The data set consisting of the training data of the time-series pixel group training data, the training data of the gas temperature values, and the training data of the gas concentration feature amounts is output to the machine learning unit 2151.

[0047] In addition, if the acquired time-series pixel group teacher data is not in the same format as the pixel group test data acquired by the test image acquisition unit 211, the teacher image acquisition unit 213 may perform processing such as cutting out or enlarging / reducing the data so that it has the same format.

[0048] The machine learning unit 2151 is a circuit that performs machine learning and generates an inference model based on time-series pixel group teacher data obtained by region extraction from the teacher data of the gas distribution moving image received by the teacher image acquisition unit 213, and a data set consisting of gas temperature values ​​and gas concentration features corresponding to the time-series pixel group teacher data received by the gas feature teacher data acquisition unit 214. The inference model is formed to estimate gas concentration features possessed by the gas distribution moving image based on the pixel group inspection data of the gas distribution moving image and the gas temperature values ​​and gas concentration features possessed by the gas distribution moving image.

[0049] For machine learning, for example, a convolutional neural network (CNN) can be used, and known software such as PyTorch can be used.

[0050] FIG. 4(b) is a schematic diagram illustrating an overview of the logical configuration of the machine learning model. The machine learning model includes an input layer 51, an intermediate layer 52-1, an intermediate layer 52-2, ..., an intermediate layer 52-n, and an output layer 53, and the interlayer filters are optimized through learning. For example, if the gas distribution video has 4x4 pixels and 100 frames, the input layer 51 receives a 4x4x100 three-dimensional tensor containing the pixel values ​​of the gas distribution video. The intermediate layer 52-1 is, for example, a convolutional layer, and receives a 4x4x100 three-dimensional tensor generated by a convolution operation from the data in the input layer 51. The intermediate layer 52-2 is, for example, a pooling layer, and receives a three-dimensional tensor obtained by resizing the data in the intermediate layer 52-1. The intermediate layer 52-n is, for example, a fully connected layer, and converts the data in the intermediate layer 52-(n-1) into a two-dimensional vector representing coordinate values. The configuration of the intermediate layers is an example, and the number n of intermediate layers is approximately 3 to 5, but is not limited to this. Also, while FIG. 4(b) depicts each layer as having the same number of neurons, each layer may have any number of neurons. The machine learning unit 2151 receives as input a data set consisting of time-series pixel group training data, gas temperature value training data, and gas concentration feature data, performs training using the gas concentration feature as the correct answer, generates a machine learning model, and outputs the model to the learning model storage and inference unit 2152. When the gas concentration feature estimation device 20 includes a GPU and RAM as a computing device, the machine learning unit 2151 may be implemented by a GPU and software.

[0051] In general, in machine learning, parameters for convolution filter processing and the like used in image recognition, etc., are automatically adjusted through a learning process to construct a processing system that can perform processing similar to human shape recognition and recognition of changes over time. The machine learning model of gas concentration feature estimation device 20 according to this embodiment extracts specific frequency component data that appears in time-series pixel group training data, and estimates the relationship between the brightness value of the specific frequency component data and the gas temperature value and gas concentration feature, thereby constructing an inference model that estimates gas concentration feature corresponding to pixel group inspection data.

[0052] The optical absorptance and the concentration-thickness product are parameters that can be calculated by estimating the background temperature with gas and the background temperature without gas from the time change for each pixel, as described in known documents, for example, Patent Document 1. For this reason, the gas feature teacher data acquisition unit 214 is configured to acquire the average value of frames in the time-series pixel group teacher data as teacher data.

[0053] However, there is a problem in that the inference model based on the time-series pixel group training data is affected by mechanical vibration noise, which reduces the accuracy of the calculated gas concentration feature amount.

[0054] Figure 6(a) is a schematic diagram showing the state of time-series pixel group data of a gas distribution moving image when there is no influence of mechanical vibration noise, and (b) is a schematic diagram showing the state of time-series pixel group data of a gas distribution moving image when there is an influence of mechanical vibration noise.

[0055] When there is no influence of mechanical vibration noise, there is no change in the brightness value of each pixel between successive frames of the time-series pixel group data of the gas distribution moving image, as shown in FIG. 6(a). On the other hand, when there is influence of mechanical vibration noise, there is a reversible change in the brightness value of each pixel between successive frames of the time-series pixel group data of the gas distribution moving image, as shown in FIG. 6(b). In the example shown in FIG. 6(b), in frame N, the brightness values ​​of the pixels in the first and third rows increase downward by a brightness value corresponding to 2°C or 4°C compared to frames N-1 and N+1. In other words, in these pixels, a change in brightness due to the vibration noise is detected even when there is no gas distribution.

[0056] Figure 7(a) shows the time change in brightness values ​​in the time-series pixel group data of a gas distribution video image when affected by mechanical vibration noise, and (b) is a schematic diagram showing the positional change in brightness values ​​in the time-series pixel group data. As shown in Figure 7(a), brightness changes when affected by mechanical vibration noise are observed as in-phase brightness changes with aligned timing in the time direction, and the direction of change in brightness values ​​in the pixel group data is independent of and usually different from the direction of gas flow, as shown in (b).

[0057] In contrast, the change in brightness value representing the gas distribution takes a different form.

[0058] Figure 8(a) shows the time change in brightness values ​​representing gas distribution in the time-series pixel group data of a gas distribution video image, and (b) is a schematic diagram showing the positional change in brightness values ​​representing gas distribution in the time-series pixel group data. As shown in Figure 8(a), brightness changes representing gas distribution are observed as brightness changes accompanied by a timing (phase) shift due to gas flow, and the direction of change in brightness values ​​in the pixel group data is the same as the direction of gas flow, as shown in (b). In this way, by using time-series information consisting of multiple pixel groups, it is possible to distinguish between brightness changes due to mechanical vibration noise and brightness changes due to gas flow.

[0059] Therefore, rather than inputting only information on time changes, the effects of mechanical vibration noise can be reduced by using machine learning to determine whether the timing is aligned due to mechanical vibration noise or whether the phase is not aligned due to the influence of gas flow.

[0060] The learning model holding and inference unit 2152 is a circuit that holds the machine learning model generated by the machine learning unit 2151, performs inference using the machine learning model, and estimates and outputs gas concentration feature amounts corresponding to the pixel group inspection data of the gas distribution moving image acquired by the inspection image acquisition unit 211. In this embodiment, the light absorptance or gas concentration thickness product is specified as the gas concentration feature amount and output as the average value of the pixel group inspection data of the input gas distribution moving image.

[0061] [Parameter conversion between gas concentration features] The gas concentration feature output unit 216 is a circuit that generates a display image for displaying the second image output by the learning model storage and inference unit 2152 on the display 24. Fig. 5(c) is a schematic diagram showing the state of gas concentration features estimated from the time-series pixel group inspection data of the gas distribution moving image when a light absorptance image is used as the gas distribution moving image. The average value of the light absorptance of 4x4 pixels and N frames in the time-series pixel group inspection data is output.

[0062] Here, the optical absorptance can be converted into a concentration-thickness product using the spectral absorption coefficient of the gas.

[0063] In the case of a learner that uses the concentration-thickness product, the concentration-thickness product can be calculated directly by creating a trained model for a specific gas species, such as methane. On the other hand, in the case of optical absorption, the concentration-thickness product for various gas species can be calculated by specifying the gas species later.

[0064] 9(a) to 9(d) are diagrams showing a method for calculating the concentration-thickness product from the light absorptance.

[0065] The method shown in Figure 9(a) is a method for experimentally obtaining relationship data between optical absorptance and concentration-thickness product for each gas type. For example, optical absorptance is measured using a gas cell filled with a gas of a predetermined concentration-thickness product, and the measurement is repeated while changing the concentration-thickness product of the filled gas to obtain relationship data between optical absorptance and concentration-thickness product. The obtained relationship data is then used to obtain the concentration-thickness product through interpolation. An interpolation method such as spline interpolation may be used as the interpolation method. The black dots in Figure 9(a) represent data for methane gas actually measured using a gas cell, and the solid black line represents the result of cubic spline interpolation. Cubic spline coefficients are calculated for the intervals between the measurement points to obtain the concentration-thickness product value corresponding to the optical absorptance value, and it can be seen that the interpolation was successful.

[0066] The method shown in Figure 9(b) involves experimentally obtaining data on the relationship between optical absorptance and concentration-thickness product for each gas species in advance, then performing mathematical parameter fitting using the obtained relationship data to obtain an approximate equation for calculating the concentration-thickness product from optical absorptance. The black dots in Figure 9(b) represent data on methane gas actually measured using a gas cell, and the black dotted line is the interpolation result using the approximate equation obtained by mathematical parameter fitting based on the measurement data, and it can be seen that the interpolation was performed well.

[0067] The method shown in Figure 9(c) is a method for obtaining data on the relationship between optical absorptance and concentration-thickness product through theoretical calculations for gas species for which an optical absorption line database is available. For example, by using an optical absorption line database called HITRAN, it is possible to obtain detailed spectral extinction coefficients. By calculating optical absorptance while changing the concentration-thickness product value, data on the relationship between optical absorptance and concentration-thickness product can be obtained. The black dots in Figure 9(c) represent data on methane gas obtained using HITRAN, and the solid black line is the result of cubic spline interpolation. It can be seen that good interpolation was achieved by calculating cubic spline coefficients for the intervals between measurement points and determining the concentration-thickness product value corresponding to the optical absorptance value.

[0068] In the method shown in Figure 9(d), the relationship data between optical absorptance and concentration-thickness product is obtained using the same method as in Figure 9(c), and the relationship data obtained using the same method as in Figure 9(b) is interpolated. The black dots in Figure 9(d) are data related to methane gas obtained using HITRAN, and the black dotted line is the interpolation result using an approximation formula obtained by fitting mathematical parameters based on this data, and it can be seen that the interpolation was successful.

[0069] It should be noted that even when the gas species is different, such as propane gas, the concentration-thickness product can be calculated using a similar calculation method.

[0070] Furthermore, even if a learning machine is created to estimate the concentration-thickness product of methane, the output concentration-thickness product of methane gas can be converted into light absorption rate using the above method, and then further converted into the concentration-thickness product of propane gas as another gas species.

[0071] As described above, the gas concentration characteristic estimation device 20 can calculate the light absorption rate independent of the gas species as the gas concentration characteristic rate, and can also determine the concentration-thickness product for a specific gas species based on the calculated light absorption rate.

[0072] Alternatively, by adopting a configuration in which the gas concentration feature quantity is a gas concentration thickness product, a method of directly calculating the concentration thickness product of a specific gas species as the gas concentration feature quantity may be adopted, which can improve accuracy and simplify calculations when the gas species are fixed.

[0073] (Operation of the gas concentration feature estimation device) The operation of the gas concentration characteristic quantity estimation device 20 according to this embodiment will be described below with reference to the drawings.

[0074] 10 is a flowchart showing the operation of the gas concentration feature estimation device 20 in the learning phase. The following description explains a method for constructing an inference model that estimates optical absorptance as a gas concentration feature. However, the optical absorptance in the following process may be replaced with the gas concentration-thickness product, and the concentration-thickness product may be estimated as the gas concentration feature.

[0075] First, gas distribution dynamic image training data is created based on a three-dimensional fluid simulation (step S110). The three-dimensional optical reflection image data may be based on a three-dimensional optical illumination analysis simulation. For example, a three-dimensional concept model of the gas facility is created using commercially available three-dimensional CAD (Computer-Aided Design) software, and a three-dimensional optical illumination analysis simulation is performed using commercially available three-dimensional optical illumination analysis simulation software taking into account the structural model. The three-dimensional optical reflection image data obtained as a result of the simulation is converted into a two-dimensional image observed from a predetermined viewpoint.

[0076] In this case, the learning phase uses images containing areas with and without gas, using subjects with different background temperatures. Furthermore, images without vibration noise and images with vibration noise are combined on the image to create learning data. Images with vibration noise are created by generating background data with vibration noise by vibrating the original background data in a planar direction through simulation, and then superimposing a gas distribution image on this background data with vibration noise.

[0077] Next, a region of time-series pixel group teacher data of a predetermined size (4×4 pixels×N frames) is extracted from the gas distribution image teacher data (step S120).

[0078] Next, the temperature value and light absorptance of the gas corresponding to the time-series pixel group teacher data are calculated (step S130). The temperature value and light absorptance of the gas may be average values ​​of 4×4 pixels×N frames in the time-series pixel group teacher data.

[0079] Next, a data set consisting of a combination of time-series pixel group teacher data of the gas distribution image, temperature values, and light absorptance is acquired. The time-series pixel group teacher data is acquired as a training image by the teacher image acquisition unit 213, and the corresponding temperature values ​​and light absorptance are acquired as ground truth data by the gas feature teacher data acquisition unit 214. At this time, image data that has been subjected to processing such as gain adjustment may be acquired as necessary.

[0080] Next, data is input into a convolutional neural network to perform machine learning and construct an inference model that estimates the light absorption rate (step S150). As a result, parameters are optimized by trial and error through deep learning, and a machine-learned model is formed. The formed machine-learned model is stored in the learning model storage and inference unit 2152.

[0081] Through the above operations, an inference model is formed that estimates the light absorption rate based on the characteristics of the time-series pixel group training data of the gas distribution image.

[0082] FIG. 11 is a flowchart showing the operation of the gas concentration characteristic quantity estimation device 20 in the operation phase.

[0083] First, gas distribution moving image inspection data is acquired by the gas visualization imaging device 10 (step S210), and the temperature value of the imaging environment is measured (step S220). The gas distribution moving image inspection image is an infrared image captured by the infrared camera of the gas visualization imaging device 10, and is a moving image showing gas distribution that visualizes the gas leak part of the inspection target.

[0084] Next, a region of time-series pixel group inspection data of a predetermined size (4 × 4 pixels × N frames) is extracted from the gas distribution image data (step S230). A portion may be cut out from each frame of the captured image so as to include all pixels that detect gas, and time-series pixel group inspection data may be generated as frames of the gas distribution image.

[0085] Next, a data set consisting of a combination of time-series pixel group inspection data of the gas distribution image and temperature values ​​is acquired (step S240). The time-series pixel group inspection data is acquired as an image to be inspected by the inspection image acquisition unit 211, and the corresponding temperature values ​​are acquired by the gas temperature value inspection data acquisition unit 212. At this time, image data that has been processed, such as gain adjustment, may be acquired as necessary. The time-series pixel group inspection data is image data in the same format as the time-series pixel group training data, and is composed of multiple frames of time-series data. Offset component subtraction and gain adjustment may be performed on the inspection image.

[0086] Next, the machine-learned model is used to calculate an estimated light absorption rate for the time-series pixel group inspection data (step S250). By using the machine-learned inference model formed in step S150, the time-series pixel group training data of the gas distribution image is used as input, and the light absorption rate is estimated based on its features.

[0087] Next, the estimated light absorption rate value is converted into a concentration-thickness product value based on the relationship characteristics between the light absorption rate of the gas species and the concentration-thickness product (step S260).

[0088] This completes the estimation of the gas concentration feature amount.

[0089] (Evaluation test) A performance evaluation test was conducted using the gas concentration feature estimation device according to the embodiment. The results will be described below with reference to images.

[0090] Figure 12(a) shows an example of a moving image of gas distribution in an inspection target. As shown in Figure 12(a), the moving image of gas distribution in an inspection target is one frame image from a time-series inspection data set in which a gas image corresponding to the optical absorptance image in Figure 12(b) is superimposed on a background image of a vertically extending rod-shaped object. In this example, the inspection image has 100 frames, measuring 112 x 64 pixels. For each frame, a high-resolution frame image was generated with an integer shift amount. After the shift, the frame image was reduced to 112 x 64 pixels to impart subpixel vibration shift. Specifically, vibration noise with a normal distribution (unit pixel) was applied, with an average amplitude of 0.473 in the X direction, 0.362 in the Y direction, and a standard deviation of 0.486 in X and 0.378 in Y. Figure 12(a) was created by first creating a gas-free background image with this vibration shift and then superimposing a gas image corresponding to the optical absorptance image in Figure 12(b).

[0091] Figure 12(b) is an example of an inspection image (correct image) of a light absorption image, which is a gas light absorption rate image superimposed on the gas distribution moving image of Figure 12(a), and consists of data of 28 x 16 pixels per frame. From this inspection image, it is possible to extract regions of time-series pixel group inspection data, with 4 x 4 pixel regions as one pixel.

[0092] FIG. 12(c) is an example of the processing result of the light absorption image, in which the time-series pixel group inspection data of the light absorption rate image shown in FIG. 12(b) is divided into 28 × 16 regions (each region is 4 × 4 pixels), and a brightness value of 14387 (16-bit value) equivalent to the gas temperature is input to each region, and the gas concentration feature is calculated. This is an output image output as a single gas concentration feature image, with one frame of data being 28 × 16 pixels.

[0093] As shown in Figure 12(c), it can be seen that vibration noise and gas can be distinguished well.

[0094] (Small summary) As described above, in this embodiment, by using time-series information consisting of multiple pixel groups, it is possible to distinguish between brightness changes due to mechanical vibration noise and brightness changes due to gas flow. As a result, the influence of mechanical vibration noise can be reduced by performing machine learning using information from a sufficient number of pixels to determine whether the timing is aligned due to mechanical vibration noise or the phase is not aligned due to the influence of gas flow.

[0095] That is, in the gas concentration feature estimation device 20, by using time-series pixel group inspection data and teacher data of the gas distribution moving image as time-series pixel group inspection data having two or more vertical and horizontal pixels each, it is possible to detect whether there is a timing (phase) shift in the brightness change in the time-series pixel group data, thereby constructing an inference model that can clearly distinguish whether the brightness change in the time-series pixel group data is due to mechanical vibration noise without a phase shift, or a brightness change due to gas flow with a phase shift, thereby improving the accuracy of the calculated gas concentration feature.

[0096] <Machine learning data generation device 30> Next, the configuration of the machine learning data generation device 30 will be described.

[0097] [Generation of density-thickness product image data] The training data used in the learning phase can also be prepared using simulation. First, a three-dimensional simulation of gas diffusion is performed using a fluid simulation to determine the time series changes in the three-dimensional gas concentration distribution data. Next, a viewpoint is set at a specified position, and the three-dimensional concentration distribution data at a certain time is scanned over an angular region that includes the space where the gas exists, to determine the gas concentration thickness product (the integral value of the concentration distribution along the line of sight in the distance direction from the viewpoint), thereby obtaining two-dimensional concentration thickness product distribution data for that time, i.e., a concentration thickness product image. By repeating the above process while changing the time, time series data consisting of multiple concentration thickness product image frames can be obtained.

[0098] Fig. 13 is a diagram showing the configuration of the machine learning data generation device 30. As shown in Fig. 13, the machine learning data generation device 30 includes a control unit (CPU) 31, a communication unit 32, a storage unit 33, a display unit 34, and an operation input unit 35, and is realized as a computer in which the control unit 31 executes a machine learning data generation program.

[0099] The memory unit 33 stores a program 331 and the like necessary for the machine-learning data generation device 30 to operate, and also functions as a temporary storage area for temporarily storing the calculation results of the control unit 31. The communication unit 32 transmits and receives information to and from the machine-learning data generation device 30 and the memory means 40. The display unit 34 is, for example, a liquid crystal panel, and displays the display screen generated by the CPU 31.

[0100] The control unit 31 executes the machine learning data generation program 331 in the storage unit 33 to realize the functions of the machine learning data generation device 30.

[0101] Fig. 14 is a functional block diagram of the machine learning data generation device 30. The condition parameters required for processing input to each functional block in Fig. 14(b) are as shown in the table below.

[0102] [Table 1] As shown in Figure 14, the machine learning data generation device 30 includes a three-dimensional structure modeling unit 311, a three-dimensional fluid simulation execution unit 312, a two-dimensional single-view gas distribution image conversion processing unit 313, a light absorptance image conversion unit 314, a background image generation unit 315, and a light intensity image generation unit 316.

[0103] The three-dimensional structure modeling unit 311 designs a three-dimensional structure model based on operation input from the operator to the operation input unit 35, performs three-dimensional structure modeling to lay out the structure in three-dimensional space, and outputs three-dimensional structure data DTstr to a subsequent stage. The three-dimensional structure data DTstr is shape data that represents the three-dimensional shapes of, for example, piping and other plant facilities. Commercially available three-dimensional CAD (Computer-Aided Design) software can be used for three-dimensional structure modeling.

[0104] FIG. 15(a) is a schematic diagram showing the data structure of the three-dimensional structure data DTstr. Here, in this specification, the X direction, Y direction, and Z direction in each figure are defined as the width direction, depth direction, and height direction, respectively. As shown in FIG. 15(a), the three-dimensional structure data DTstr is three-dimensional voxel data representing a three-dimensional space, and is composed of structure identification information Std arranged at coordinates in the X direction, Y direction, and Z direction. Because the structure identification information Std is expressed as three-dimensional shape data, it may be recorded as a binary image using 0 and 1, such as "structure present" or "structure absent."

[0105] The three-dimensional fluid simulation execution unit 312 receives the three-dimensional structure data DTstr as input, and further receives condition parameters CP1 required for the fluid simulation based on operation input from the operator to the operation input unit 35. The condition parameters CP1 are parameters that determine the setting conditions required for the fluid simulation, mainly related to gas leakage, such as the gas type, gas flow rate, gas flow velocity in three-dimensional space, and the shape, diameter, and position of the gas leakage source, as shown in Table 1. By generating images while varying these condition parameters in various ways, it is possible to generate a large amount of training data.

[0106] Then, a three-dimensional fluid simulation is performed in the three-dimensional space where the three-dimensional structure modeling has been performed, and three-dimensional gas distribution image data DTgas is generated and output to a subsequent stage. The three-dimensional gas distribution image data DTgas is data that includes at least a three-dimensional gas concentration distribution. Calculations are performed using commercially available three-dimensional fluid simulation software, such as ANSYS Fluent, Flo EFD, or Femap / Flow.

[0107] Fig. 15(b) is a schematic diagram showing the data structure of the three-dimensional gas distribution image data DTgas. As shown in Fig. 15(b), the three-dimensional gas distribution image data DTgas is three-dimensional voxel data representing a three-dimensional space, and is composed of gas concentration data Dst (%) of voxels arranged at coordinates in the X, Y, and Z directions, and may further include gas flow velocity vector data Vc (Vx, Vy, Vz) of the voxels.

[0108] The 2D single-view gas distribution image conversion processor 313 inputs and acquires 3D gas distribution image data DTgas of gas leaking from a gas leakage source into a three-dimensional space, and further acquires condition parameters CP2 required for conversion processing into a 2D single-view gas distribution image based on operation input from the operator to the operation input unit 35. The condition parameters CP2 are parameters related to the imaging conditions of the gas visualization imaging device, such as the imaging device field of view, line of sight direction, distance, and image resolution, as shown in Table 1. The 2D single-view gas distribution image conversion processor 313 then converts the 3D gas distribution image data DTgas into 2D gas distribution image data observed from a predetermined viewpoint. This allows processing to convert the 2D gas distribution image data into concentration-thickness product image data DTdt.

[0109] The concentration thickness product image DTdt is an image equivalent to an inspection image of leaked gas captured by the gas visualization imaging device 10, and is an image that represents how the gas appears from a viewpoint. Furthermore, by taking into account the information of the structure three-dimensional data DTstr, it is possible to generate gas concentration thickness product image data DTdt that does not reflect gas images that are blocked by structures and cannot be observed from the viewpoint.

[0110] Figure 16 is a schematic diagram for explaining an overview of the concentration thickness product calculation method in the 3D single-view gas distribution image conversion process, and is a conceptual diagram of a method for generating gas concentration thickness product image data DTdt from 3D gas distribution image data DTgas of a gas that represents the behavior of the gas.

[0111] The two-dimensional single viewpoint gas distribution image conversion processing unit 313 generates multiple values ​​of the concentration thickness product Dst, which is obtained by spatially integrating the three-dimensional gas concentration image represented by the three-dimensional gas distribution image data DTgas in the line of sight direction from a preset viewpoint position (X, Y, Z), by changing the angles θ and σ of the line of sight direction, and arranges the obtained concentration thickness product Dst values ​​two-dimensionally to generate the concentration thickness product image data DTdt.

[0112] Specifically, as shown in FIG. 21 , an arbitrary viewpoint position SP(X, Y, Z) is set in three-dimensional space, and a virtual image plane VF is set at a position a predetermined distance from the viewpoint position SP(X, Y, Z) in the direction of the three-dimensional gas concentration image represented by the three-dimensional gas distribution image data DTgas. At this time, the virtual image plane VF is set so that its center O intersects with a line passing through the viewpoint position SP(X, Y, Z) and the central voxel of the three-dimensional gas distribution image data DTgas. The image frame of the virtual image plane VF is set according to the angle of view of the gas visualization imaging device 10. The line of sight direction DA from the viewpoint position SP(X, Y, Z) toward a pixel of interest A(x, y) on the virtual image plane VF is inclined by an angle θ in the X direction and an angle σ in the Y direction with respect to the line of sight direction DO toward the center pixel O, i.e., the line of sight direction DO of the gas visualization imaging device.

[0113] Along the line of sight DA corresponding to this pixel of interest A(x, y), gas concentration distribution data corresponding to voxels of the 3D gas distribution image that intersect with the line of sight is spatially integrated in the line of sight direction DA with respect to the voxels that intersect with the line of sight to calculate the value of the gas concentration thickness product for the pixel of interest A(x, y).Then, while changing the angles θ and σ according to the angle of view of the gas visualization imaging device 10, the position of the pixel of interest A(x, y) is gradually moved, and the calculation of the gas concentration thickness product value is repeated for all pixels on the virtual image plane VF as the pixel of interest A(x, y), thereby calculating the concentration thickness product image data DTdt.

[0114] Furthermore, by using the same three-dimensional gas distribution image data DTgas and varying the viewpoint position SP (X, Y, Z) to generate concentration-thickness product image data DTdt, multiple concentration-thickness product image data DTdt can be easily generated from a single fluid simulation.

[0115] FIG. 17 is a schematic diagram for explaining an outline of the concentration-thickness product calculation method in the 3D single-viewpoint gas distribution image conversion process when a structure is present, and shows a case where the structure blocks observation from the gas visualization imaging device 10.

[0116] In this case, taking into account the three-dimensional position of the structure, the value of the gas concentration thickness product for the pixel of interest A(x, y) is calculated using a method that does not perform spatial integration of the gas concentration distribution behind the structure as viewed from the viewpoint position SP(X, Y, Z).

[0117] Specifically, the value of the gas concentration thickness product for the pixel of interest A(x,y) is calculated by spatially integrating from the viewpoint position SP(X,Y,Z) to the line of sight DA corresponding to the pixel of interest A(x,y).

[0118] Then, by repeatedly calculating the gas concentration thickness product value for all pixels on the virtual image plane VF, taking into account the three-dimensional position of the structure, a gas concentration thickness product image DTdt is generated that does not reflect gas images that are blocked by structures and cannot be observed from the viewpoint.

[0119] [Generation of light absorption rate image data] Next, the time series data of the concentration-thickness product image is converted into time series data of the light absorptance image using the spectral absorption coefficient data of the gas species set during the simulation.

[0120] The light absorptance image converter 314 acquires the condition parameter CP3 based on the operation input, and further converts the gas concentration thickness product image data DTdt into light absorptance image data DTα, and outputs it to the subsequent stage.

[0121] Figure 18(a) is a schematic diagram showing an overview of concentration-thickness product image data, (b) is a schematic diagram for explaining an overview of the concentration-thickness product calculation method in the light absorptance image conversion process, and (c) is a schematic diagram showing an overview of the light absorptance image data.

[0122] The machine learning data generation device 30 calculates the light absorptance value α corresponding to the gas concentration thickness product at pixel (x, y) shown in FIG. 18(c) using the relationship between the gas concentration thickness product and the light absorptance α as shown in FIG. 18(b) for the gas concentration thickness product value Dt at pixel (x, y) in the gas concentration thickness product image data DTdt shown in FIG. 18(a). The light absorptance α varies depending on the gas species specified by the condition parameter CP3. Based on data relating the concentration thickness product value Dt and the light absorptance α, for example, the light absorptance α value corresponding to the gas concentration thickness product value Dt stored in a data table or a mathematical formula representing an approximation curve, the gas concentration thickness product image data DTdt is converted into light absorptance image data Dtα based on the concentration thickness product value and the light absorption characteristics of the gas. The relationship between the gas concentration thickness product value Dt and the light absorptance α for each gas species may be obtained in advance by actual measurement.

[0123] [Generation of light intensity image data] Next, an image equivalent to the captured image is generated using the time-series data of the obtained light absorption rate image and a specified background image. The background image is composed of the same number of frames as the light absorption rate image, and is configured so that the image brightness is constant or changes over time. In addition, the pattern within one frame may be a full-screen fill, or different brightness may be set for each small area.

[0124] The light absorptance can be determined from the time-series data of the light absorptance image described above by extracting an area and processing average values ​​in the spatial and temporal directions.

[0125] The gas temperature used in generating the image equivalent to the captured image is used as is. This concludes the method for preparing training data using simulation.

[0126] The background image generating unit 315 acquires the background location data PTbk and the condition parameter CP5 based on the operation input, and generates background image data DTIback.

[0127] Figure 19(a) is a schematic diagram showing an overview of the three-dimensional structure data. As shown in Figure 19(a), the three-dimensional structure data DTstr is three-dimensional voxel data that represents three-dimensional space, and the structure identification information Std differs from the structure data stored as a binary image such as "structure present" or "structure absent" in that a value indicating the classification of the structure surface is assigned to each pixel and recorded as a multi-valued image such as 0, 1, 2, 3, etc.

[0128] As described above, the structure identification information Std consisting of this multi-valued image is subjected to two-dimensional single viewpoint processing using the virtual image plane VF observed from the viewpoint position SP (X, Y, Z) as the image frame, thereby generating background location data PTbk.

[0129] 19(b) is a schematic diagram showing an overview of the extracted background location data PTbk. The background location data PTbk is two-dimensional image data consisting of background classification Std (Std=0, 1, 2, 3, etc.) for pixel A(x, y) in the virtual image plane VF. Here, the background classification Std is a classification number determined, for example, based on the optical characteristics of the structure surface. For example, unpainted pipes may be set to 1, painted pipes to 2, and concrete to 3.

[0130] 19(c) is a conceptual diagram showing a method for generating background image data DTIback based on background location data PTbk. As shown in FIG. 19(c), background image data DTIback is generated based on background location data PTbk generated as a multi-valued image and condition parameters CP5 based on operational input.

[0131] The condition parameter CP5 is the lighting conditions for the structure, such as the background two-dimensional temperature distribution, background surface spectral emissivity, background surface spectral reflectance, illumination light wavelength distribution, spectral illuminance, and illumination angle, as well as the temperature conditions of the structure itself, as shown in Table 1. When the background location data PTbk is a multi-value image, different condition parameters CP5 are assigned for each background classification Std to generate background image data DTIback. When the background location data is a binary image, different conditions, whether there is a background or not, are assigned for each background classification Std to generate background image data DTIback.

[0132] The light intensity image generating unit 316 generates light intensity image data DTI based on the light absorptance image data DTα, background image data DTIback, and the gas temperature condition provided as the condition parameter CP4 based on an operational input.

[0133] Figure 20(a) is a schematic diagram showing an overview of the background image data DTIback, (b) is a schematic diagram showing an overview of the light absorptance image data DTα, and (c) is a schematic diagram showing an overview of the light intensity image data.

[0134] When the infrared intensity at the coordinates (x, y) of the background image is DTIback(x, y), the blackbody radiance equivalent to the gas temperature is Igas, and the light absorptance value at the coordinates (x, y) of the light absorptance image is DTα(x, y), the infrared intensity at the coordinates (x, y) of the light intensity image, DTI(x, y), is calculated using Equation 1.

[0135]

number

[0136] Then, the infrared intensities DTI(x, y) are calculated for all pixels A(x, y) on the virtual image plane VF, and light intensity image data DTI is generated.

[0137] (2D single-view gas distribution image conversion processing operation) Next, the two-dimensional single-viewpoint gas distribution image conversion processing operation in the machine learning data generation device 30 will be described with reference to the drawings.

[0138] 21 is a flowchart showing an outline of the two-dimensional single-view gas distribution image conversion process. This process is executed by the two-dimensional single-view gas distribution image conversion processing unit 313, the function of which is configured by the control unit 31.

[0139] First, the 2D single viewpoint gas distribution image conversion processing unit 313 acquires the structure 3D data DTstr (step S1), and then acquires the 3D gas distribution image data DTgas (step S2). Next, based on the operation input, it accepts input of information regarding, for example, the imaging device field of view, line of sight direction, distance, and image resolution as condition parameters CP2 (step S3). Furthermore, based on the operation input, it sets the viewpoint position SP (X, Y, Z) corresponding to the position of the imaging portion of the gas visualization imaging device 10 in the 3D space (step S4).

[0140] Next, a virtual image plane VF is set at a predetermined distance from the viewpoint position SP (X, Y, Z) in the direction of the three-dimensional gas concentration image, and the position of the image frame of the virtual image plane VF is calculated according to the angle of view of the gas visualization imaging device 10, as described above (step S5).

[0141] Next, the coordinates of the pixel of interest A(x,y) are set to initial values ​​(step S6), and the position LV on the line of sight from the viewpoint position SP(X,Y,Z) toward the pixel of interest A(x,y) on the virtual image plane VF is set to initial values ​​(step S7).

[0142] Next, it is determined whether the structure identification information Std of the voxel in the structure 3D data DTstr that intersects with the line of sight indicates "no structure" (Std=0) (step S8), and if it indicates "structure present", the position of the pixel of interest A(x,y) is gradually moved (step S9), and the process returns to step S7; if it does not indicate "structure present", the process proceeds to step S10.

[0143] In step S10, it is determined whether or not there are any voxels in the three-dimensional gas distribution image data DTgas that intersect with the line of sight. If there are no voxels in the three-dimensional structure data DTstr that intersect with the line of sight, the position LV on the line of sight is incremented by a unit length (for example, 1) (step S11), and the process returns to step S8. If there are any voxels that intersect with the line of sight, the concentration thickness value Dst of that voxel is read, and the sum of this with the integrated value Dsti stored in an addition register or the like is saved in an addition register or the like as a new integrated value Dsti (step S12).

[0144] Next, it is determined whether or not the calculation has been completed for the entire line of sight, which corresponds to the range where the line of sight intersects with the voxel (step S13). If not completed, the position LV on the line of sight is incremented by unit length (step S14), and the process returns to step S8. If completed, the gas concentration distribution data of the voxel of the three-dimensional gas distribution image that intersects with the line of sight is spatially integrated along the line of sight direction DA corresponding to the pixel of interest A(x, y), and the value of the gas concentration thickness product for the pixel of interest A(x, y) is calculated.

[0145] Next, it is determined whether the calculation of the gas concentration thickness product values ​​for all pixels on the virtual image plane VF has been completed (step S15). If not, the position of the target pixel A(x, y) is gradually moved (step S16) and the process returns to step S7. If completed, the gas concentration thickness product values ​​are calculated for all pixels on the virtual image plane VF, and a gas concentration thickness product image DTdt is generated as two-dimensional gas distribution image data for the virtual image plane VF.

[0146] Next, it is determined whether the generation of the gas concentration thickness product image DTdt has been completed for all viewpoint positions SP (X, Y, Z) to be calculated (step S17). If it has not been completed, the process returns to step S4 and a gas concentration thickness product image DTdt is generated for the new viewpoint position SP (X, Y, Z) input by operation. If it has been completed, the process ends.

[0147] (Background image generation process) Next, the background image generation processing operation in the machine learning data generation device 30 will be described with reference to the drawings.

[0148] 22 is a flowchart showing an outline of the background image generation process, which is executed by the background image generation unit 315, the function of which is configured by the control unit 31.

[0149] First, the background image generating unit 315 acquires the three-dimensional structure data DTstr (step S1). The operations of steps S3 to S8 are the same as the operations of the respective steps in FIG.

[0150] In step 8, if the background location data PTbk of the voxel that intersects with the line of sight is "structure present," the background classification Std of the background location data PTbk is obtained (step S121A), the condition parameter CP5 corresponding to the background classification Std is input (step S122A), the background image data value of the target pixel A is determined, and then the position of the target pixel A(x, y) is gradually moved (step S9), and the process returns to step S7.

[0151] The condition parameter CP5 is, for example, a background two-dimensional temperature distribution, a background surface spectral emissivity, a background surface spectral reflectance, an illumination light wavelength distribution, a spectral illuminance, and an illumination angle.

[0152] On the other hand, if the result is not "structure present," it is determined whether or not calculation has been completed for the entire line of sight, which corresponds to the range where the line of sight intersects with the voxel (step S13). If not, the line of sight position LV is incremented by a unit length (step S14), and the process returns to step S8. On the other hand, if calculation has been completed, the standard value set when there is no structure is determined as the background image data value for the target pixel A, and it is determined whether or not calculation has been completed for all pixels on the virtual image plane VF (step S15). If not, the position of the target pixel A(x, y) is gradually moved (step S16), and the process returns to step S7. If calculation has been completed, the process ends. Here, the standard value set when there is no structure is, for example, the background image data value corresponding to the ground or sky in real space. The standard value can be obtained by appropriately setting the condition indicated by the condition parameter CP5.

[0153] As a result of the above, the background classification Std of the background location data PTbk is acquired for all pixels on the virtual image plane VF, and background image data DTIback relating to the virtual image plane VF is generated.

[0154] (Light intensity image data generation process) Next, the light intensity image data generation processing operation in the machine learning data generation device 30 will be described with reference to the drawings.

[0155] FIG. 23 is a flowchart showing an outline of the light intensity image data generation process.

[0156] First, background image data DTIback(x,y) and light absorptance image data DTα(x,y) on the virtual image plane VF are obtained (steps S101 and S102), and condition parameters CP4 related to gas temperature conditions are input (step S103).

[0157] Next, the blackbody radiance Igas corresponding to the gas temperature is acquired (step S104), the infrared intensity DTI(x, y) of the light intensity image is calculated using (Equation 1) (step S105), and the calculated value is output as light intensity image data DTI (step S106), and the process ends.

[0158] (Small summary) As a result of the above, the light intensity image data DTI generated by the machine learning data generation device 30 can be used as a gas distribution moving image, and the gas concentration features calculated from the light absorptance image data Dtα or the gas concentration thickness product image DTdt can be used as machine learning training data for the gas concentration feature estimation device 20.

[0159] This makes it possible to efficiently generate tens of thousands of sets of training data consisting of inputs and correct outputs, such as gas leak images and gas leak source location coordinate information in this example, for training the machine learning model in the gas concentration feature estimation device 20 of this embodiment, thereby contributing to improving the training accuracy.

[0160] Furthermore, by generating training data of light absorptance image data DTα that is closer to the gas distribution image obtained by the gas visualization imaging device 10, training data that is closer to the inspection image can be generated, which can contribute to further improving the learning accuracy of the machine learning model.

[0161] Furthermore, by generating training data that is light intensity image data DTI that is more similar to the gas distribution image obtained by the gas visualization imaging device 10, training data that is more similar to the inspection image can be generated, making it easier to extract the target gas portion. This contributes to further improving the learning accuracy of the machine learning model in the gas leak detection device.

[0162] <Summary> A gas concentration feature estimation device according to one aspect of the present disclosure includes an inspection data acquisition unit that acquires time-series pixel group inspection data of a gas distribution moving image, each of which has two or more vertical and horizontal pixels, extracted from inspection data of a gas distribution moving image that represents a gas presence region in space, and a temperature value of the gas, and an estimation unit that calculates the gas concentration feature corresponding to the time-series pixel group inspection data acquired by the inspection data acquisition unit using an inference model trained by machine learning using time-series pixel group teacher data of the gas distribution moving image that is the same size as the time-series pixel group inspection data and values ​​of the gas temperature and gas concentration feature corresponding to the time-series pixel group teacher data as teacher data. Furthermore, the time-series pixel group inspection data may have a number of frame pixels that is smaller than the number of pixels in the frames of the gas distribution moving image.

[0163] This configuration reduces the influence of mechanical vibration noise on measurements, and enables accurate detection of feature quantities representing the gas concentration in space from the infrared gas distribution dynamic image.

[0164] In another aspect, in any of the above aspects, the time-series pixel group inspection data may have a frame pixel count of 3 to 7 pixels in both the vertical and horizontal directions.

[0165] In another aspect, in any of the above aspects, the time-series pixel group teacher data may be configured to be a moving image including vibration noise. In another aspect, in any of the above aspects, the gas concentration feature amount may be an optical absorptance, and the system may further include a conversion unit that converts the optical absorptance corresponding to the time-series pixel group inspection data into a concentration-thickness product value related to the gas species based on a relationship characteristic between the optical absorptance and a concentration-thickness product of the gas species.

[0166] With this configuration, it is possible to calculate the light absorptance independent of the gas type as the gas concentration characteristic amount, and to obtain the concentration-thickness product for a specific gas type based on the calculated light absorptance.

[0167] In another aspect, in any of the above aspects, the gas concentration feature amount may be a gas concentration thickness product.

[0168] With this configuration, it is possible to directly obtain a concentration-thickness product that specifies a specific gas species as a gas concentration feature, thereby improving accuracy and simplifying calculations when the gas species is fixed.

[0169] In another aspect, in any of the above aspects, the gas distribution moving image may be an image captured by an imaging device.

[0170] In another aspect, in any of the above aspects, the training data for the gas distribution dynamic image may be generated by simulation.

[0171] This configuration makes it possible to efficiently generate tens of thousands of sets of training data consisting of inputs and correct outputs, which contributes to improving the accuracy of training.

[0172] In another aspect, in any of the above aspects, the training data for the gas distribution dynamic image may be generated from a background image and light absorptance.

[0173] With this configuration, it is possible to generate training data that is closer to the inspection image by using an optical absorptance image that is closer to the gas distribution image obtained by the gas visualization imaging device, which can contribute to further improving the learning accuracy of the machine learning model.

[0174] In another aspect, in any of the above aspects, the number of frames in the time-series pixel group inspection data or the time-series pixel group teacher data is greater than the number of vertical or horizontal pixels in each frame.

[0175] With this configuration, by using time-series information consisting of multiple pixel groups, it is possible to distinguish between brightness changes due to mechanical vibration noise and brightness changes due to gas flow. As a result, the influence of mechanical vibration noise can be reduced by performing machine learning using information from a sufficient number of pixels to determine whether the timing is aligned due to mechanical vibration noise or the phase is not aligned due to the influence of gas flow.

[0176] In another aspect, in any of the above aspects, the gas concentration feature corresponding to the time-series pixel group inspection data may be configured as a sequence of values ​​calculated for each frame of the time-series pixel group inspection data.

[0177] In another aspect, in any of the above aspects, the gas concentration feature corresponding to the time-series pixel group inspection data may be configured to be an average value of values ​​calculated for each frame of the time-series pixel group inspection data.

[0178] This configuration simplifies the calculation for estimating the gas concentration feature amount.

[0179] In another aspect, in any of the above aspects, the imaging device may be an infrared camera.

[0180] Furthermore, a gas concentration feature estimation method according to one aspect of the present disclosure may be configured to acquire time-series pixel group inspection data of a gas distribution moving image having two or more vertical and horizontal pixels each, which is extracted from inspection data of a gas distribution moving image representing a gas presence region in space, and a temperature value of the gas, and to calculate an estimated value of the gas concentration feature corresponding to the acquired time-series pixel group inspection data using an inference model machine-learned using as training data time-series pixel group teacher data of a gas distribution moving image having the same size as the time-series pixel group inspection data and gas temperature values ​​and gas concentration feature values ​​corresponding to the time-series pixel group teacher data.

[0181] This configuration makes it possible to realize a gas concentration feature estimation method that reduces the influence of mechanical vibration noise on measurements and can accurately detect feature quantities that represent gas concentrations in space from infrared gas distribution dynamic images.

[0182] A program according to an aspect of the present disclosure is a program for causing a computer to perform gas concentration feature estimation processing, the gas concentration feature estimation processing including: extracting a region from inspection data of a gas distribution moving image that represents a gas presence region in a space, the region having two or more vertical and horizontal pixels, acquiring time-series pixel group inspection data of the gas distribution moving image and a temperature value of the gas; and extracting time-series pixel group inspection data of the gas distribution moving image that is the same size as the time-series pixel group inspection data. The configuration may also be such that an estimated value of the gas concentration feature corresponding to the acquired time-series pixel group inspection data is calculated using an inference model that has been machine-learned using group teacher data and the gas temperature value and gas concentration feature value corresponding to the time-series pixel group teacher data as teacher data.

[0183] Furthermore, a gas concentration feature inference model generation device according to one aspect of the present disclosure may be configured to include a teacher data acquisition unit that acquires, as teacher data, time-series pixel group teacher data of the gas distribution moving image and gas temperature values ​​and gas concentration feature values ​​corresponding to the time-series pixel group teacher data, where the number of vertical and horizontal pixels is two or more and an area is extracted from teacher data of a gas distribution moving image that represents a gas presence area in space; and a machine learning unit that constructs an inference model that calculates, based on the teacher data, time-series pixel group inspection data of the same size as the time-series pixel group teacher data and an estimated value of the gas concentration feature corresponding to the gas temperature value corresponding to the time-series pixel group inspection data.

[0184] This configuration reduces the impact of mechanical vibration noise on measurements and allows the construction of a gas concentration feature inference model that can accurately detect features representing spatial gas concentrations from infrared gas distribution dynamic images.

[0185] <<Variations>> The gas concentration feature estimation system 1 according to the embodiment has been described above, but the present disclosure is not limited to the above embodiment except for its essential characteristic components. For example, the present disclosure also includes various modifications that can be made to the embodiment by a person skilled in the art, and modifications realized by arbitrarily combining the components and functions of each embodiment without departing from the spirit of the present invention. Below, modifications of the above embodiment will be described as examples of such modifications.

[0186] (1) In the above embodiment, the gas concentration feature estimation device 20 acquires gas dynamic image training data that is generated by simulation using the machine learning data generation device 30. However, the gas dynamic image training data may be an actual image captured by the gas visualization imaging device 10. In this case, it is preferable to capture the gas dynamic image training image against a black body plate as the background to prevent the background image from affecting the concentration-thickness product or light absorption rate of the gas distribution. In this case, the temperature value may be the result of measuring the temperature of the captured atmosphere, and the gas concentration feature training data may be a value calculated based on the brightness of the gas dynamic image training image.

[0187] (2) In the above embodiment, the time-series pixel group data has 100 frames of time elements and 4 × 4 pixel elements, and machine learning is performed with a relatively increased weight on the time elements. However, a configuration in which the pixel elements are relatively increased compared to the time elements may also be used.

[0188] (3) In the above-described embodiment, a gas plant is used as an example of a gas facility as an inspection image. However, the present disclosure is not limited to this and may be applied to the generation of display images for gas-using equipment, devices, laboratories, research labs, factories, and offices.

[0189] (4) Although the present disclosure has been described based on the above-described embodiment, the present disclosure is not limited to the above-described embodiment, and the following cases are also included in the present invention.

[0190] For example, the present invention may be a computer system having a microprocessor and a memory, the memory storing the computer program, and the microprocessor operating in accordance with the computer program. For example, the present invention may be a computer system having a computer program for processing the system or components of the present disclosure, and operating in accordance with this program (or instructing each connected component to operate).

[0191] The present invention also includes a computer system that implements all or part of the processing in the above system or its components, including a microprocessor, storage media such as ROM and RAM, a hard disk unit, etc. The RAM or hard disk unit stores a computer program that performs the same operations as each of the above devices. The microprocessor operates in accordance with the computer program, causing each device to perform its function.

[0192] Furthermore, some or all of the components constituting each of the above devices may be configured from a single system LSI (Large Scale Integration). A system LSI is an ultra-multifunctional LSI manufactured by integrating multiple components on a single chip, and specifically, is a computer system configured to include a microprocessor, ROM, RAM, etc. These may be individually integrated into a single chip, or a single chip may include some or all of them. The RAM stores a computer program that achieves the same operations as each of the above devices. The system LSI achieves its functions when the microprocessor operates in accordance with the computer program. For example, the present invention also includes a case where the processing of a system or its components is stored as an LSI program, and this LSI is inserted into a computer to execute a predetermined program.

[0193] The method of integration is not limited to LSI, but may be realized by dedicated circuits or general-purpose processors. It is also possible to use FPGAs (Field Programmable Gate Arrays), which can be programmed after the LSI is manufactured, or reconfigurable processors, which allow the connections and settings of circuit cells inside the LSI to be reconfigured.

[0194] Furthermore, if an integrated circuit technology that can replace LSI emerges due to advances in semiconductor technology or other derived technologies, it is of course possible to use that technology to integrate functional blocks.

[0195] Furthermore, some or all of the functions of the system or its components according to each embodiment may be realized by a processor such as a CPU executing a program. A non-transitory computer-readable recording medium may also be used, on which a program for implementing the operations of the system or its components is recorded. The program may be executed by another independent computer system by recording a program or signal on a recording medium and transferring it. It goes without saying that the program can be distributed via a transmission medium such as the Internet.

[0196] Furthermore, the system according to the above embodiment or each of its components may be configured to be realized by a programmable device such as a CPU, a GPU (Graphics Processing Unit), or a processor, and software. These components may be a single circuit component, or may be a collection of multiple circuit components. Furthermore, multiple components may be combined to form a single circuit component, or may be a collection of multiple circuit components.

[0197] (5) The division of functional blocks is an example, and multiple functional blocks may be realized as a single functional block, one functional block may be divided into multiple blocks, or some functions may be moved to another functional block. Furthermore, the functions of multiple functional blocks with similar functions may be processed in parallel or time-shared by a single piece of hardware or software.

[0198] The order in which the steps are performed is merely an example for specifically explaining the present invention, and other orders may be used. Some of the steps may be performed simultaneously (in parallel) with other steps.

[0199] Furthermore, at least some of the functions of the respective embodiments and their modified examples may be combined. Furthermore, the numbers used above are all examples for specifically explaining the present invention, and the present invention is not limited to the exemplified numbers.

[0200] <<Additional Information>> The embodiments described above each illustrate a preferred specific example of the present invention. The numerical values, shapes, materials, components, component placement and connection configurations, steps, and step sequences shown in the embodiments are merely examples and are not intended to limit the present invention. Furthermore, among the components in the embodiments, steps that are not recited in the independent claims that represent the highest concept of the present invention are described as optional components that constitute more preferred embodiments.

[0201] In order to facilitate understanding of the invention, the scale of the components in the drawings of the above embodiments may differ from the actual scale. Furthermore, the present invention is not limited to the description of the above embodiments, and can be modified as appropriate within the scope of the gist of the present invention.

[0202] Furthermore, although there are circuit components, lead wires, and other components on the substrate, various embodiments of electrical wiring and electrical circuits can be implemented based on common knowledge in the technical field, and since they are not directly related to the explanation of the present invention, their explanation is omitted. Note that the figures shown above are schematic diagrams and are not necessarily strict illustrations. [Industrial Applicability]

[0203] The gas concentration feature estimation device, gas concentration feature estimation method, program, and gas concentration feature inference model generation device according to the present disclosure are widely applicable to the estimation of gas concentration feature values ​​using an infrared imaging device. [Explanation of symbols]

[0204] 1. Gas concentration feature estimation system 10 Gas visualization imaging device 20 Gas concentration feature estimation device 21 Control section 210 Gas leak location identification device 211 Inspection image acquisition unit 212 Gas temperature value inspection data acquisition unit 213 Teacher image acquisition unit 214 Gas feature training data acquisition unit 215 Gas concentration feature estimation unit 2151 Machine Learning Department 2152 Learning Model Maintenance and Inference Unit 216 Gas concentration feature output unit 22 Communication Circuit 23 Storage device 24 Display 25 Operation input section 30 Machine learning data generation device 31 Control unit (CPU) 311 3D Structure Modeling Department 312 3D Fluid Simulation Execution Unit 313 2D single-view gas distribution image conversion processing unit 314 Light absorption rate image conversion unit 315 Background Image Generation Unit 316 Light Intensity Image Generation Unit 32 Communications Department 33 Storage section 34 Display section 35 Operation input section 40 Memory means

Claims

1. An inspection data acquisition unit that acquires time-series pixel group inspection data, each having two or more vertical and horizontal pixels, extracted from inspection data of a gas distribution dynamic image that represents the area where gas exists in space, and the temperature value of the gas; an estimation unit that calculates the gas concentration feature amount corresponding to the time-series pixel group inspection data acquired by the inspection data acquisition unit using an inference model that has been machine-learned using time-series pixel group teacher data of a gas distribution moving image of the same size as the time-series pixel group inspection data and gas temperature values ​​and gas concentration feature amount values ​​corresponding to the time-series pixel group teacher data as teacher data, the gas concentration feature quantity is a light absorptance, Further, a conversion unit converts the optical absorptance corresponding to the time-series pixel group inspection data into a concentration-thickness product value related to the gas species based on a relationship characteristic between the optical absorptance of the gas species and the concentration-thickness product. Gas concentration feature estimation device.

2. An inspection data acquisition unit that acquires time-series pixel group inspection data, each having two or more vertical and horizontal pixels, extracted from inspection data of a gas distribution dynamic image that represents the area where gas exists in space, and the temperature value of the gas; an estimation unit that calculates the gas concentration feature amount corresponding to the time-series pixel group inspection data acquired by the inspection data acquisition unit using an inference model that has been machine-learned using time-series pixel group teacher data of a gas distribution moving image of the same size as the time-series pixel group inspection data and gas temperature values ​​and gas concentration feature amount values ​​corresponding to the time-series pixel group teacher data as teacher data, The training data for the gas distribution dynamic image is generated from a background image and light absorption rate. Gas concentration feature estimation device.

3. An inspection data acquisition unit that acquires time-series pixel group inspection data, each having two or more vertical and horizontal pixels, extracted from inspection data of a gas distribution dynamic image that represents the area where gas exists in space, and the temperature value of the gas; an estimation unit that calculates the gas concentration feature amount corresponding to the time-series pixel group inspection data acquired by the inspection data acquisition unit using an inference model that has been machine-learned using time-series pixel group teacher data of a gas distribution moving image of the same size as the time-series pixel group inspection data and gas temperature values ​​and gas concentration feature amount values ​​corresponding to the time-series pixel group teacher data as teacher data, The number of frames in the time-series pixel group inspection data or the time-series pixel group teacher data is greater than the number of vertical or horizontal pixels in each frame. Gas concentration feature estimation device.

4. An inspection data acquisition unit that acquires time-series pixel group inspection data, each having two or more vertical and horizontal pixels, extracted from inspection data of a gas distribution dynamic image that represents the area where gas exists in space, and the temperature value of the gas; an estimation unit that calculates the gas concentration feature amount corresponding to the time-series pixel group inspection data acquired by the inspection data acquisition unit using an inference model that has been machine-learned using time-series pixel group teacher data of a gas distribution moving image of the same size as the time-series pixel group inspection data and gas temperature values ​​and gas concentration feature amount values ​​corresponding to the time-series pixel group teacher data as teacher data, The gas concentration feature quantity corresponding to the time-series pixel group inspection data is a sequence of values ​​calculated for each frame of the time-series pixel group inspection data. Gas concentration feature estimation device.

5. An inspection data acquisition unit that acquires time-series pixel group inspection data, each having two or more vertical and horizontal pixels, extracted from inspection data of a gas distribution dynamic image that represents the area where gas exists in space, and the temperature value of the gas; an estimation unit that calculates the gas concentration feature amount corresponding to the time-series pixel group inspection data acquired by the inspection data acquisition unit using an inference model that has been machine-learned using time-series pixel group teacher data of a gas distribution moving image of the same size as the time-series pixel group inspection data and gas temperature values ​​and gas concentration feature amount values ​​corresponding to the time-series pixel group teacher data as teacher data, The gas concentration feature amount corresponding to the time-series pixel group inspection data is an average value of values ​​calculated for each frame of the time-series pixel group inspection data. Gas concentration feature estimation device.

6. the time-series pixel group inspection data has a number of frame pixels that is less than the number of pixels of a frame of the gas distribution moving image; The gas concentration characteristic quantity estimation device according to claim 1 .

7. the time-series pixel group inspection data has a frame pixel count of 3 to 7 in both vertical and horizontal directions, The gas concentration characteristic quantity estimation device according to claim 6 .

8. The time-series pixel group teacher data is a moving image including vibration noise. The gas concentration characteristic quantity estimation device according to claim 1 .

9. The gas concentration feature quantity is a gas concentration thickness product The gas concentration characteristic quantity estimation device according to claim 1 .

10. The gas distribution moving image is an image captured by an imaging device. The gas concentration characteristic quantity estimation device according to claim 1 .

11. The training data of the gas distribution dynamic image is generated by simulation. The gas concentration characteristic quantity estimation device according to claim 1 .

12. The imaging device is an infrared camera. The gas concentration characteristic quantity estimation device according to claim 10 .

13. acquiring time-series pixel group inspection data of a gas distribution moving image, which is an area extracted from inspection data of a gas distribution moving image showing a gas presence area in a space, and which has two or more pixels in each of the vertical and horizontal directions, and a temperature value of the gas; calculating an estimated value of the gas concentration feature corresponding to the acquired time-series pixel group inspection data using an inference model machine-learned using time-series pixel group teacher data of a gas distribution moving image of the same size as the time-series pixel group inspection data and the gas temperature value and gas concentration feature value value corresponding to the time-series pixel group teacher data as teacher data; the gas concentration feature quantity is a light absorptance, Furthermore, based on the relationship characteristics between the light absorptance of the gas species and the concentration-thickness product, the light absorptance corresponding to the time-series pixel group inspection data is converted into a concentration-thickness product value for the gas species. Gas concentration feature estimation method.

14. Obtaining time-series pixel group inspection data of a gas distribution moving image, which is an area extracted from inspection data of a gas distribution moving image showing a gas existence area in space, and which has two or more vertical and horizontal pixels, and a temperature value of the gas; calculating an estimated value of the gas concentration feature corresponding to the acquired time-series pixel group inspection data using an inference model machine-learned using time-series pixel group teacher data of a gas distribution moving image of the same size as the time-series pixel group inspection data and the gas temperature value and gas concentration feature value value corresponding to the time-series pixel group teacher data as teacher data; The training data for the gas distribution dynamic image is generated from a background image and light absorption rate. Gas concentration feature estimation method.

15. A method for detecting a gas distribution motion image, the method comprising: obtaining time-series pixel group inspection data of a gas distribution motion image, the time-series pixel group inspection data of which is an area extracted from the inspection data of the gas distribution motion image showing the area where the gas exists in space, and the temperature value of the gas; and calculating an estimated value of the gas concentration feature corresponding to the acquired time-series pixel group inspection data using an inference model machine-learned using time-series pixel group teacher data of a gas distribution moving image of the same size as the time-series pixel group inspection data and the gas temperature value and gas concentration feature value value corresponding to the time-series pixel group teacher data as teacher data; The number of frames in the time-series pixel group inspection data or the time-series pixel group teacher data is greater than the number of vertical or horizontal pixels in each frame. Gas concentration feature estimation method.

16. A method for detecting a gas distribution motion image, the method comprising: obtaining time-series pixel group inspection data of a gas distribution motion image, the time-series pixel group inspection data of which is an area extracted from inspection data of a gas distribution motion image showing an area where gas exists in space, and the temperature value of the gas, the time-series pixel group inspection data of which is an area extracted from inspection data of a gas distribution motion image showing an area where gas exists in space ... calculating an estimated value of the gas concentration feature corresponding to the acquired time-series pixel group inspection data using an inference model machine-learned using time-series pixel group teacher data of a gas distribution moving image of the same size as the time-series pixel group inspection data and the gas temperature value and gas concentration feature value value corresponding to the time-series pixel group teacher data as teacher data; The gas concentration feature quantity corresponding to the time-series pixel group inspection data is a sequence of values ​​calculated for each frame of the time-series pixel group inspection data. Gas concentration feature estimation method.

17. A method for detecting a gas distribution motion image, the method comprising: obtaining time-series pixel group inspection data of a gas distribution motion image, the time-series pixel group inspection data of which is an area extracted from inspection data of the gas distribution motion image showing a gas existence area in space, the vertical and horizontal pixel numbers of which are 2 or more; and a temperature value of the gas; calculating an estimated value of the gas concentration feature corresponding to the acquired time-series pixel group inspection data using an inference model machine-learned using time-series pixel group teacher data of a gas distribution moving image of the same size as the time-series pixel group inspection data and the gas temperature value and gas concentration feature value value corresponding to the time-series pixel group teacher data as teacher data; The gas concentration feature amount corresponding to the time-series pixel group inspection data is an average value of values ​​calculated for each frame of the time-series pixel group inspection data. Gas concentration feature estimation method.

18. A program for causing a computer to perform gas concentration feature estimation processing, The gas concentration feature amount estimation process includes: acquiring time-series pixel group inspection data of a gas distribution moving image, which is an area extracted from inspection data of a gas distribution moving image showing a gas presence area in a space, and which has two or more pixels in each of the vertical and horizontal directions, and a temperature value of the gas; calculating an estimated value of the gas concentration feature corresponding to the acquired time-series pixel group inspection data using an inference model machine-learned using time-series pixel group teacher data of a gas distribution moving image of the same size as the time-series pixel group inspection data and the gas temperature value and gas concentration feature value value corresponding to the time-series pixel group teacher data as teacher data; the gas concentration feature quantity is a light absorptance, Furthermore, based on the relationship characteristics between the light absorptance of the gas species and the concentration-thickness product, the light absorptance corresponding to the time-series pixel group inspection data is converted into a concentration-thickness product value for the gas species. program.

19. A program for causing a computer to perform gas concentration feature estimation processing, comprising: The gas concentration feature amount estimation process includes: acquiring time-series pixel group inspection data of a gas distribution moving image, which is an area extracted from inspection data of a gas distribution moving image showing a gas presence area in a space, and which has two or more pixels in each of the vertical and horizontal directions, and a temperature value of the gas; calculating an estimated value of the gas concentration feature corresponding to the acquired time-series pixel group inspection data using an inference model machine-learned using time-series pixel group teacher data of a gas distribution moving image of the same size as the time-series pixel group inspection data and the gas temperature value and gas concentration feature value value corresponding to the time-series pixel group teacher data as teacher data; The training data for the gas distribution dynamic image is generated from a background image and light absorption rate. program.

20. A program for causing a computer to perform gas concentration feature estimation processing, comprising: The gas concentration feature amount estimation process includes: acquiring time-series pixel group inspection data of a gas distribution moving image, which is an area extracted from inspection data of a gas distribution moving image showing a gas presence area in a space, and which has two or more pixels in each of the vertical and horizontal directions, and a temperature value of the gas; calculating an estimated value of the gas concentration feature corresponding to the acquired time-series pixel group inspection data using an inference model machine-learned using time-series pixel group teacher data of a gas distribution moving image of the same size as the time-series pixel group inspection data and the gas temperature value and gas concentration feature value value corresponding to the time-series pixel group teacher data as teacher data; The number of frames in the time-series pixel group inspection data or the time-series pixel group teacher data is greater than the number of vertical or horizontal pixels in each frame. program.

21. A program for causing a computer to perform gas concentration feature estimation processing, comprising: The gas concentration feature amount estimation process includes: acquiring time-series pixel group inspection data of a gas distribution moving image, which is an area extracted from inspection data of a gas distribution moving image showing a gas presence area in a space, and which has two or more pixels in each of the vertical and horizontal directions, and a temperature value of the gas; calculating an estimated value of the gas concentration feature corresponding to the acquired time-series pixel group inspection data using an inference model machine-learned using time-series pixel group teacher data of a gas distribution moving image of the same size as the time-series pixel group inspection data and the gas temperature value and gas concentration feature value value corresponding to the time-series pixel group teacher data as teacher data; The gas concentration feature quantity corresponding to the time-series pixel group inspection data is a sequence of values ​​calculated for each frame of the time-series pixel group inspection data. program.

22. A program for causing a computer to perform gas concentration feature estimation processing, comprising: The gas concentration feature amount estimation process includes: acquiring time-series pixel group inspection data of a gas distribution moving image, which is an area extracted from inspection data of a gas distribution moving image showing a gas presence area in a space, and which has two or more pixels in each of the vertical and horizontal directions, and a temperature value of the gas; calculating an estimated value of the gas concentration feature corresponding to the acquired time-series pixel group inspection data using an inference model machine-learned using time-series pixel group teacher data of a gas distribution moving image of the same size as the time-series pixel group inspection data and the gas temperature value and gas concentration feature value value corresponding to the time-series pixel group teacher data as teacher data; The gas concentration feature amount corresponding to the time-series pixel group inspection data is an average value of values ​​calculated for each frame of the time-series pixel group inspection data. program.

23. a training data acquisition unit that acquires, as training data, time-series pixel group training data of a gas distribution moving image having two or more vertical and horizontal pixels each representing a gas presence region in space, and gas temperature values ​​and gas concentration feature values ​​corresponding to the time-series pixel group training data; a machine learning unit that constructs an inference model that calculates, based on the teacher data, time-series pixel group inspection data of the same size as the time-series pixel group teacher data, which is extracted from inspection data of a gas distribution moving image, and an estimated value of the gas concentration feature amount corresponding to the gas temperature value corresponding to the time-series pixel group inspection data; the gas concentration feature quantity is a light absorptance, Further, a conversion unit converts the optical absorptance corresponding to the time-series pixel group inspection data into a concentration-thickness product value related to the gas species based on a relationship characteristic between the optical absorptance of the gas species and the concentration-thickness product. Gas concentration feature inference model generation device.

24. A teacher data acquisition unit that acquires, as teacher data, time-series pixel group teacher data of a gas distribution moving image having two or more vertical and horizontal pixels each representing a gas presence area in space, and gas temperature values ​​and gas concentration feature values ​​corresponding to the time-series pixel group teacher data; a machine learning unit that constructs an inference model that calculates, based on the teacher data, time-series pixel group inspection data of the same size as the time-series pixel group teacher data, which is extracted from inspection data of a gas distribution moving image, and an estimated value of the gas concentration feature amount corresponding to the gas temperature value corresponding to the time-series pixel group inspection data; The training data for the gas distribution dynamic image is generated from a background image and light absorption rate. Gas concentration feature inference model generation device.

25. A teacher data acquisition unit that acquires, as teacher data, time-series pixel group teacher data of a gas distribution moving image having two or more vertical and horizontal pixels each representing a gas presence region in space, and gas temperature values ​​and gas concentration feature values ​​corresponding to the time-series pixel group teacher data; a machine learning unit that constructs an inference model that calculates, based on the teacher data, time-series pixel group inspection data of the same size as the time-series pixel group teacher data, which is extracted from inspection data of a gas distribution moving image, and an estimated value of the gas concentration feature amount corresponding to the gas temperature value corresponding to the time-series pixel group inspection data; The number of frames in the time-series pixel group inspection data or the time-series pixel group teacher data is greater than the number of vertical or horizontal pixels in each frame. Gas concentration feature inference model generation device.

26. A teacher data acquisition unit that acquires, as teacher data, time-series pixel group teacher data of a gas distribution moving image having two or more vertical and horizontal pixels each representing a gas presence area in space, and gas temperature values ​​and gas concentration feature values ​​corresponding to the time-series pixel group teacher data; a machine learning unit that constructs an inference model that calculates, based on the teacher data, time-series pixel group inspection data of the same size as the time-series pixel group teacher data, which is extracted from inspection data of a gas distribution moving image, and an estimated value of the gas concentration feature amount corresponding to the gas temperature value corresponding to the time-series pixel group inspection data; The gas concentration feature quantity corresponding to the time-series pixel group inspection data is a sequence of values ​​calculated for each frame of the time-series pixel group inspection data. Gas concentration feature inference model generation device.

27. ​​A teacher data acquisition unit that acquires, as teacher data, time-series pixel group teacher data of a gas distribution moving image having two or more vertical and horizontal pixels each representing a gas presence area in space, and gas temperature values ​​and gas concentration feature values ​​corresponding to the time-series pixel group teacher data; a machine learning unit that constructs an inference model that calculates, based on the teacher data, time-series pixel group inspection data of the same size as the time-series pixel group teacher data, which is extracted from inspection data of a gas distribution moving image, and an estimated value of the gas concentration feature amount corresponding to the gas temperature value corresponding to the time-series pixel group inspection data; The gas concentration feature amount corresponding to the time-series pixel group inspection data is an average value of values ​​calculated for each frame of the time-series pixel group inspection data. Gas concentration feature inference model generation device.

Images