Information processing device, information processing method, and program

Abstract

The information processing device includes: an acquisition unit that acquires an inspection image in which an inspection object is captured; and a detection unit that detects the presence of a gas in the vicinity of the test object using a prediction model that uses the temporal and spatial change rates of the test image as input parameters and information that can detect the presence or absence of a gas in the test image as output parameters.

Description

Technical Field

[0001] The present invention relates to an information processing device, an information processing method and a program.

[0002] This application claims priority based on patent application No. 2022-136947 filed in Japan on August 30, 2022, and the contents are incorporated herein by reference. Background Art

[0003] As a technique for visualizing a fluid such as a gas, it is considered to visualize the flow of a gas flow as a measurement object by measuring the distribution of temperature changes and displaying it in a time series. For example, Patent Document 1 describes a technique for visualizing the flow of a fluid by the following process: obtaining temperature image data of a fluid captured by an infrared camera for a specified period, generating partial temperature data by shifting and cutting out the temperature data for the specified period in the temperature image data in the time direction according to a specified analysis step, performing frequency analysis (Fourier transform processing) on ​​each of the cut-out partial temperature data, detecting the distribution of temperature changes every specified period, and displaying the distribution of temperature changes in a time series.

[0004] Previous technical literature

[0005] Patent Literature

[0006] Patent Document 1: Japanese Patent Application Publication No. 2020-52036 Summary of the invention

[0007] Technical issues to be solved by the invention

[0008] In the previous technology, the premise is that the infrared camera is fixedly installed and the background of the gas is a uniform object such as a wall. On the other hand, when detecting and visualizing gas leaks in petrochemical equipment, pipelines, storage sites, etc., erroneous detection of gas leaks may occur due to fluctuations in the background of the gas. As background fluctuations, for example, fluctuations caused by sunlight, clouds, wind, dust, and temperature (mist) can be considered. These behaviors are non-steady-state phenomena with strong randomness at the moment, and it may be difficult to distinguish the gas flow from the background fluctuations only through frequency analysis of time series like previous technologies.

[0009] An object of the present invention is to provide an information processing device, an information processing method, and a program capable of detecting the presence of a gas with high accuracy.

[0010] Means for solving technical problems

[0011] According to one embodiment of the present invention, an information processing device comprises: an acquisition unit, which acquires an inspection image that takes a picture of an inspection object; and a detection unit, which detects the presence of gas around the inspection object using a prediction model that uses the temporal and spatial change rates of the inspection image as input parameters and uses information that can detect whether gas exists in the inspection image as an output parameter.

[0012] According to one embodiment of the present invention, an information processing method has the following steps: acquiring an inspection image of an inspection object; and detecting the presence of gas around the inspection object using a prediction model that uses the temporal and spatial change rates of the inspection image as input parameters and uses information that can detect whether gas exists in the inspection image as output parameters.

[0013] According to one embodiment of the present invention, a program causes an information processing device to perform the following steps: acquiring an inspection image of an inspection object; and detecting the presence of gas around the inspection object using a prediction model that uses the temporal and spatial change rates of the inspection image as input parameters and uses information that can detect whether gas is present in the inspection image as an output parameter.

[0014] Effects of the Invention

[0015] According to the above aspect, the presence of gas can be detected with high accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a schematic diagram showing the overall configuration of an information processing system according to one embodiment.

[0017] Figure 2 This is a block diagram showing a functional structure of an information processing device according to one embodiment.

[0018] Figure 3 This is a flowchart showing an example of a learning process of a prediction model according to one embodiment.

[0019] Figure 4 This is a flowchart showing an example of a gas detection process according to one embodiment. DETAILED DESCRIPTION

[0020] (Overall structure of information processing system)

[0021] Hereinafter, embodiments will be described in detail with reference to the accompanying drawings.

[0022] Figure 1 It is a schematic diagram showing the overall configuration of an information processing system according to one embodiment.

[0023] The information processing system 1 is a system for detecting the presence (leakage) of gas G (spray F of gas G) around a gas pipe 9. The gas pipe 9 is provided at a gas utilization site, etc., and is a pipe for circulating gas G inside. The gas pipe 9 is an example of an inspection object involved in this embodiment. In other embodiments, for example, a storage tank of gas G may be set as an inspection object.

[0024] like Figure 1 As shown, the information processing system 1 includes an information processing device 10 , a moving object 20 , and a camera 30 .

[0025] The mobile body 20 is, for example, an unmanned vehicle such as a drone, and moves within the installation area of ​​the gas pipe 9 according to a preset monitoring route or in accordance with an operation instruction of an operator.

[0026] The camera 30 is mounted on the mobile body 20 and photographs the gas pipe 9. The camera 30 is, for example, an infrared camera capable of detecting the gas absorption wavelength (specific wavelength) of the gas G. Furthermore, the camera 30 changes the orientation (photographing direction) of the photographing range R according to the positional relationship between the mobile body 20 and the gas pipe 9, thereby being able to photograph various parts of the gas pipe 9, such as the upper surface and the side surface.

[0027] When the moving body 20 moves into the installation area, the camera 30 always captures an image of the gas pipe 9. In addition, the moving body 20 transmits the image (moving image) captured by the camera 30 to the information processing device 10.

[0028] The information processing device 10 is connected to the mobile body 20 in a communicative manner. When there is no leakage of gas G, the information processing device 10 learns a prediction model for detecting the presence of gas G based on an image captured by the camera 30 (hereinafter, also referred to as a learning image). The details of the prediction model will be described later. In addition, when the information processing device 10 is actually used, it detects whether gas G exists (is leaking) around the gas pipe 9 based on the image captured by the camera 30 (hereinafter, also referred to as an inspection image) and the prediction model.

[0029] in addition, Figure 1 2 shows an example in which the information processing device 10 is installed at a monitoring base point separated from the installation area of ​​the gas pipe 9 , but the present invention is not limited to this. In other embodiments, the information processing device 10 may be built in the mobile body 20 .

[0030] (Functional Structure of Information Processing Device)

[0031] Figure 2 This is a block diagram showing a functional structure of an information processing device according to one embodiment.

[0032] like Figure 2 As shown, the information processing device 10 includes a processor 11 , a memory 12 , a storage device 13 , and a communication interface 14 .

[0033] The processor 11 operates according to a predetermined program to function as an acquisition unit 110 , an image processing unit 111 , a detection unit 112 , and a learning unit 113 .

[0034] The acquisition unit 110 acquires an inspection image in which the gas pipe 9 is photographed.

[0035] The image processing unit 111 performs a convolution operation on the inspection image to generate a texture image.

[0036] The detection unit 112 detects the presence of gas G around the gas pipe 9 using a prediction model that takes temporal and spatial changes in the brightness of the inspection image as input parameters and outputs information that can detect whether gas G exists in the inspection image.

[0037] The learning unit 113 learns the prediction model based on temporal changes and spatial changes in brightness of a learning image captured when the gas G is not present around the gas pipe 9 .

[0038] The memory 12 has a memory area necessary for the operation of the processor 11 .

[0039] The storage device 13 is a so-called auxiliary storage device, and may be, for example, a HDD (Hard Disk Drive) or an SSD (Solid State Drive).

[0040] The communication interface 14 is an interface for transmitting and receiving various information with an external device (moving object 20 ).

[0041] In addition, the prescribed program executed by the processor 11 of the information processing device 10 is stored in a computer-readable recording medium. In addition, the computer-readable recording medium refers to a magnetic disk, a magneto-optical disk, a CD-ROM, a DVD-ROM, a semiconductor memory, etc. In addition, the computer program can be transmitted to a computer via a communication line, and the program can be executed by the computer that receives the transmission. In addition, the program can also be used to implement a part of the above-mentioned functions. In addition, it can also be a so-called differential file (differential program) that can be implemented by combining with a program that has recorded the above-mentioned functions in a computer system.

[0042] (About the learning process of the prediction model)

[0043] Figure 3 This is a flowchart showing an example of a learning process of a prediction model according to one embodiment.

[0044] Here, reference Figure 3 , the processing flow of the information processing device 10 learning the prediction model is explained.

[0045] First, the acquisition unit 110 acquires a learning image captured by the camera 30 (step S1 ). The learning image is an image captured when the gas G does not exist around the gas pipe 9 .

[0046] The image processing unit 111 performs convolution operation on the learning image as an array of grayscale levels to generate a texture image (step S2). Convolution operation is a method of using a filter of any size to generate a coarse pixel that integrates any pixel of the learning image and its surrounding adjacent pixels (for example, the surrounding 8 pixels). As the value of the coarse pixel, the sum, average or maximum value of the original values ​​of any pixel and adjacent pixels is used. Through this processing, it is possible to obtain a texture image with a reduced array size while maintaining the texture of the learning image, so that the operation speed in the subsequent processing (steps S3 to S4) can be accelerated. In addition, here, "texture" means "features" such as bumps caused by brightness, etc., and "texture image" means an image with bumps. In addition, since the changes in the pixel group within a certain specified range can be evaluated, the fluctuations caused by the noise in each pixel of the original image can be eliminated. In addition, the noise here refers to noise generated by the influence of surrounding radio waves mixed in during image communication, etc., rather than fluctuations in the jet and background.

[0047] Next, the learning unit 113 uses the texture image to learn the prediction model. The prediction model predicts whether the object presented in the image is gas G or background (object other than gas G) based on the feature quantity of the texture image, and outputs information that can detect whether gas G exists in each pixel. In this embodiment, the learning unit 113 uses LSTM (Long short-term memory) to learn the prediction model.

[0048] LSTM is one of the RNNs (Recurrent neural networks). There are accumulated data and newly acquired data in LSTM. When updating the prediction model, the network autonomously determines the feature quantity of the newly acquired data, and refers to the data close to the found feature quantity from the accumulated data, and updates the prediction model. Specifically, LSTM has a forget gate, an input gate, and an output gate. The forget gate makes the following selection of accumulated data (long-term storage): among the accumulated data, the data related to the last output (short-term storage) and the newly acquired data are retained, and the irrelevant data is forgotten. The input gate accumulates the last output (short-term storage) and the newly acquired data. The output gate outputs the prediction based on the last output (short-term storage), the newly acquired data, and the data remaining in the forget gate (long-term storage). Therefore, LSTM has the advantage of being able to make predictions in cyclic data with multiple periodic features.

[0049] Furthermore, for example, assume that for a certain image, the brightness of a part of the area decreases due to the influence of clouds, and gas G is present in another area separated from the area. At this time, in the prediction of whether it is gas G, the feature quantity of the area whose brightness decreases due to clouds is unnecessary information. In a conventional RNN, learning is performed on the entire image, that is, including areas that are not spatially connected to gas G (areas where the brightness decreases due to clouds), which may lead to a decrease in prediction accuracy. In addition, in a conventional LSTM, changes in time are usually processed. On the other hand, if it is assumed that the brightness of a region of a certain image changes due to the influence of clouds, when gas G is not present in the region, assuming that the change period of the brightness of the region caused by the influence of clouds is the same as the period caused by gas G, the influence of clouds and the changes caused by gas G will be confused for prediction, which may reduce the prediction accuracy.

[0050] Therefore, in this embodiment, not only the temporal variation but also the spatial variation is learned as a feature quantity for LSTM, thereby selecting and rejecting the spatial features and improving the accuracy of prediction of whether it is gas G. Spatial variation is also temporal variation, so spatial variation is added as one of the data (input parameters) used for learning.

[0051] Specifically, first, the learning unit 113 obtains input parameters that become learning data from the texture image of the learning image (step S3). The input parameters are evaluation items of fluctuations of the background included in the learning image, and in this embodiment, they are the temporal and spatial change rates of the texture image. Here, an example using the brightness change rate is described.

[0052] The temporal and spatial brightness change rate refers to information indicating the brightness change rate relative to time (frame) and the range (distance) to which the brightness change extends in each image space at each time for a certain element (pixel) of the texture image array. In addition, the distance can be a distance represented by two-dimensional coordinates (xy coordinates) or a distance represented by polar coordinates.

[0053] Next, the learning unit 113 uses the LSTM to learn and output a prediction model capable of detecting the presence or absence of the gas G (for example, the probability of the presence of the gas G) for each pixel using input parameters as learning data (step S4 ).

[0054] As described above, the learning image serving as the basis of the learning data is an image captured when there is no gas G around the gas pipe 9. Therefore, the prediction model according to the present embodiment learns the changes in the background other than the gas G in a series of images (animation) in which the scenery changes rapidly, and performs prediction by removing data related to the fluctuation of the background (e.g., the change in brightness value caused by the influence of clouds or the range in the image space where the brightness value changes). After removing the fluctuation of the background, the prediction model predicts that if there is a fluctuation in brightness in a certain pixel, the probability of the existence of gas G in the pixel is high.

[0055] If the information processing device 10 has sufficient computing power, step S2 may be omitted. In this case, in steps S3 and S4, the learning unit 113 acquires input parameters from the original learning image and performs learning.

[0056] Furthermore, in the present embodiment, an example of using the temporal and spatial brightness change rates of a learning image (texture image) as input parameters (change rates) is described, but the present invention is not limited thereto. In other embodiments, (1) the average and standard deviation of the brightness change rates of elements of a texture image, (2) the average and standard deviation of the movement of corner points of elements of a texture image, and (3) the change rates of the histogram shape (average and standard deviation) of the brightness of the texture image at each prescribed time interval may be used as input parameters. A prescribed time interval is set for each of (1) to (3). The corner points of the elements are feature points detected by a known corner point detection process. For example, the parameters of (1) can be used to evaluate fluctuations caused by sunlight or haze. The parameters of (2) can be used to evaluate fluctuations caused by wind or dust. The parameters of (3) can be used to evaluate fluctuations caused by clouds. Furthermore, the interval between two data for calculating the brightness change rate is determined, for example, by the interval at which the average value of the brightness change rates of all elements is the largest.

[0057] Furthermore, environmental conditions for each hour may be added as input parameters. Environmental conditions include data such as temperature, humidity, weather, and air pressure. Furthermore, when the above (1) to (3) are used as input parameters, the data of environmental conditions are obtained in accordance with the time intervals set for (1) to (3). Thus, a prediction model can be learned that can perform predictions under any environmental conditions based on a neural network with an LSTM forget gate.

[0058] Furthermore, in this embodiment, an example in which the learning unit 113 obtains input parameters from a learning image (texture image) is described, but the present invention is not limited thereto. In other embodiments, other computers or engineers may extract feature quantities as input parameters from an image and use them as learning data for a prediction model.

[0059] (About gas detection processing)

[0060] Figure 4 This is a flowchart showing an example of a gas detection process according to one embodiment.

[0061] Here, reference Figure 4 , the processing flow of the information processing device 10 detecting the gas G is described. In addition, in the present embodiment, it is assumed that the information processing device 10 acquires the inspection images captured by the camera 30 one by one, and detects the gas G in real time. In other embodiments, the information processing device 10 may detect the gas G based on the inspection images acquired together after the camera 30 captures a part or all of the gas piping 9.

[0062] First, the acquisition unit 110 acquires an inspection image captured by the camera 30 (step S11 ).

[0063] The image processing unit 111 performs convolution operation on the inspection image as an array of gray levels to generate a texture image (step S12). Figure 3 The same as step S2 of . Through this process, a texture image with a reduced array size can be obtained while maintaining the texture of the inspection image, so the operation speed in the subsequent processing (steps S13 to S14) can be accelerated. In addition, since the change of the pixel group within a certain specified range can be evaluated, the fluctuation caused by the noise in each pixel of the original image can be eliminated.

[0064] Next, the detection unit 112 detects the gas G based on the texture image and the learned prediction model. Specifically, first, the detection unit 112 obtains input parameters to the prediction model from the texture image of the inspection image (step S13). The input parameters are the temporal and spatial brightness change rates of the texture image.

[0065] Then, the detection unit 112 inputs the input parameters to the prediction model, and obtains as output the existence probability of the gas G in each pixel. When the existence probability of the gas G exceeds a predetermined threshold, the detection unit 112 detects the existence of the gas G in the pixel (step S14).

[0066] Each time the information processing device 10 acquires an inspection image from the mobile object 20 (camera 30), it executes Figure 4 A series of processes are performed and the gas G from the gas pipe 9 is monitored for leakage.

[0067] If the information processing device 10 has sufficient computing power, step S12 may be omitted. In this case, the detection unit 112 acquires input parameters from the original inspection image and detects the gas G in steps S13 and S14.

[0068] Furthermore, environmental conditions for each hour may be added as input parameters. Environmental conditions include data such as temperature, humidity, weather, and air pressure. Furthermore, when the above (1) to (3) are used as input parameters, the data of environmental conditions are obtained in accordance with the time intervals set for (1) to (3). Thus, the presence or absence of gas G can be detected under any environmental conditions based on a neural network with an LSTM forget gate.

[0069] Furthermore, in this embodiment, an example in which the detection unit 112 obtains input parameters from the inspection image (texture image) is described, but the present invention is not limited thereto. In other embodiments, other computers or engineers may extract feature quantities as input parameters from the image, and the detection unit 112 detects the gas G using the input parameters specified by the other computers or engineers.

[0070] Furthermore, in this embodiment, an example in which the prediction model predicts (outputs the probability of existence) the presence or absence of gas G in each pixel is described, but it is not limited to this. In other embodiments, the prediction model can output (1) the rate of change of brightness, (2) the rate of change of the movement of the corner points, and (3) the rate of change of the histogram shape of the brightness of the texture image within a specified time interval for each element of an array of images (texture images) acquired within a specified period. The specified time intervals are set for (1) to (3), respectively. In this case, Figure 4 In step S14, the detection unit 112 detects the presence of gas G in the element, for example, when the brightness change rate exceeds ±3 times (±3σ) the standard deviation of the brightness change rate obtained from the data when there is no gas G. In addition, the detection unit 112 may apply a background difference method to the pixels corresponding to the element in which the gas G is detected, thereby extracting only the movement of the spray F portion.

[0071] (Function, Effect)

[0072] As described above, the information processing device 10 involved in this embodiment includes: an acquisition unit 110, which acquires an inspection image of the gas piping 9; and a detection unit 112, which detects the presence of gas G around the gas piping 9 using a prediction model that uses the temporal and spatial brightness change rate of the inspection image as an input parameter and uses information that can detect whether gas G exists in the inspection image as an output parameter.

[0073] Thus, the information processing device 10 can distinguish whether an object appearing in each part of the inspection image is the gas G or the background other than the gas G by considering not only the temporal change of the background other than the gas G in the inspection image but also the spatial change. Thus, the information processing device 10 can detect the presence (leakage) of the gas G around the gas pipe 9 with high accuracy.

[0074] And, the prediction model is a machine learning model using LSTM learning.

[0075] The evaluation item (brightness change rate in this embodiment) has a long-term frequency due to seasonal changes, climate changes, etc., but such long-term frequencies cannot be learned in conventional machine learning methods (such as RNN). In contrast, the prediction model involved in this embodiment can retain data similar to the trend of the evaluation item in the accumulated past data through the forget gate of LSTM and learn. Therefore, by using such a prediction model, the information processing device 10 can detect the gas G with good accuracy without being affected by seasonal changes or climate changes.

[0076] Furthermore, the input parameters include the average and standard deviation of the brightness change rate, the average and standard deviation of the movement amount of the corner points of the elements included in the inspection image, and the change rate of the histogram shape of the brightness of the inspection image.

[0077] Thus, the information processing device 10 can distinguish between changes in brightness caused by various factors such as sunlight, mist, wind, dust, and clouds, and changes in brightness caused by the presence of the gas G.

[0078] Furthermore, the information processing device 10 further includes an image processing unit 111 that performs a convolution operation on the inspection image to generate a texture image. The detection unit 112 acquires an input parameter from the texture image.

[0079] Thus, the information processing device 10 can obtain a texture image with a reduced array size while maintaining the texture of the inspection image, thereby increasing the calculation speed in the detection unit 112. In addition, since the change of the pixel group within a certain specified range can be evaluated, the fluctuation caused by the noise in each pixel of the original image can be eliminated. Thus, the detection accuracy of gas G can be improved.

[0080] In the above, one embodiment is described in detail with reference to the drawings, but the specific structure is not limited to the above description, and various design changes can be made. That is, in other embodiments, the order of the above processes can be appropriately changed. In addition, some processes can also be executed simultaneously.

[0081] The information processing device 10 according to the above embodiment may be composed of a single computer, or the structure of the information processing device 10 may be divided and configured in a plurality of computers, and the plurality of computers may cooperate with each other to function as the information processing device 10. In this case, a part of the computer constituting the information processing device 10 may be mounted inside the mobile body 20, and another part of the computer may be provided outside the mobile body 20.

[0082] Furthermore, in the above embodiment, the camera 30 is mounted on the moving body 20, but the present invention is not limited thereto. For example, in another embodiment, the camera 30 may be a fixed camera installed at predetermined intervals in the installation area of ​​the gas pipe 9.

[0083] <Notes>

[0084] The information processing device, the information processing method, and the program described in the above-mentioned embodiments can be understood, for example, as follows.

[0085] (1) According to the first method, the information processing device 10 includes: an acquisition unit 110 for acquiring an inspection image of the inspection object 9; and a detection unit 112 for detecting the presence of gas G around the inspection object 9 using a prediction model that uses the temporal and spatial change rates of the inspection image as input parameters and uses information that can detect whether gas G exists in the inspection image as an output parameter.

[0086] Thus, the information processing device 10 can distinguish whether an object appearing in each part of the inspection image is the gas G or the background other than the gas G by considering not only the temporal change of the background other than the gas G in the inspection image but also the spatial change. Thus, the information processing device 10 can detect the presence (leakage) of the gas G around the gas pipe 9 with high accuracy.

[0087] (2) According to the second aspect, in the information processing device 10 according to the first aspect, the rate of change is a rate of change of brightness in the inspection image.

[0088] (3) According to a third aspect, in the information processing device 10 according to the first or second aspect, the prediction model is a machine learning model using LSTM (Long short-term memory) learning.

[0089] By using such a prediction model, the information processing device 10 can detect the gas G with high accuracy without being affected by seasonal changes, climate changes, and the like.

[0090] (4) According to the fourth method, in the information processing device 10 involved in any one of the first to third methods, the input parameters include the average and standard deviation of the brightness change rate, the average and standard deviation of the movement amount of the corner points of the elements contained in the inspection image, and the change rate of the histogram shape of the brightness of the inspection image.

[0091] Thus, the information processing device 10 can distinguish between changes in brightness caused by various factors such as sunlight, mist, wind, dust, and clouds, and changes in brightness caused by the presence of the gas G.

[0092] (5) According to the fifth aspect, the information processing device 10 according to any one of the first to fourth aspects further includes an image processing unit 111 that performs a convolution operation on the inspection image to generate a texture image, and the detection unit 112 acquires an input parameter from the texture image.

[0093] Thus, the information processing device 10 can obtain a texture image with a reduced array size while maintaining the texture of the inspection image, thereby increasing the calculation speed in the detection unit 112. In addition, since the change of the pixel group within a certain specified range can be evaluated, the fluctuation caused by the noise in each pixel of the original image can be eliminated. Thus, the detection accuracy of gas G can be improved.

[0094] (6) According to the sixth embodiment, the information processing method has the following steps: obtaining an inspection image of the inspection object 9; and detecting the presence of gas G around the inspection object 9 using a prediction model that uses the temporal and spatial change rates of the inspection image as input parameters and uses information that can detect whether gas G exists in the inspection image as an output parameter.

[0095] (7) According to the seventh method, the program causes the information processing device 10 to perform the following steps: obtaining an inspection image of the inspection object 9; and detecting the presence of gas G around the inspection object 9 using a prediction model that uses the temporal and spatial change rates of the inspection image as input parameters and uses information that can detect whether gas G exists in the inspection image as an output parameter.

[0096] Industrial Applicability

[0097] According to the above aspect, the presence of gas can be detected with high accuracy.

[0098] Explanation of symbols:

[0099] 1-information processing system, 9-gas piping (inspection object), 10-information processing device, 11-processor, 110-acquisition unit, 111-image processing unit, 112-detection unit, 113-learning unit, 12-memory, 13-storage device, 14-communication interface, 20-moving body, 30-camera.

Claims

1. An information processing device comprising: an acquisition unit that acquires an inspection image that captures the inspection object; and The detection unit detects the presence of gas around the inspection object using a prediction model that takes the temporal and spatial change rates of the inspection image as input parameters and takes information that can detect whether gas exists in the inspection image as an output parameter.

2. The information processing device according to claim 1, wherein: The change rate is a brightness change rate in the inspection image.

3. The information processing device according to claim 1, wherein: The prediction model is a machine learning model using LSTM learning.

4. The information processing device according to claim 1, wherein: The input parameters include an average and a standard deviation of a brightness change rate, an average and a standard deviation of movement amounts of corner points of elements included in the inspection image, and a change rate of a histogram shape of brightness of the inspection image.

5. The information processing device according to any one of claims 1 to 4, further comprising an image processing unit that performs a convolution operation on the inspection image to generate a texture image, The detection unit obtains the input parameter from the texture image.

6. An information processing method comprising the following steps: acquiring an inspection image capturing the inspection object; and The presence of gas around the inspection object is detected using a prediction model that uses the temporal and spatial change rates of the inspection image as input parameters and uses information that can detect whether gas exists in the inspection image as an output parameter.

7. A program causing an information processing device to execute the following steps: acquiring an inspection image capturing the inspection object; and The presence of gas around the inspection object is detected using a prediction model that uses the temporal and spatial change rates of the inspection image as input parameters and uses information that can detect whether gas exists in the inspection image as an output parameter.

Images