Gas leakage quantitative calibration method and system based on multispectral feature fusion

Abstract

The invention relates to the technical field of infrared thermal imaging, in particular to a gas leakage quantitative calibration method and system based on multispectral feature fusion, and the method comprises the steps: configuring a multiband blackbody radiation source; building a simulated gas cloud cluster environment, obtaining configuration parameters in real time, and performing classified collection on the configuration parameters to construct a four-dimensional parameter space; acquiring infrared thermal imaging data corresponding to the simulated gas cloud cluster environment, and associating the infrared thermal imaging data with the four-dimensional parameter space to form a structured sample set; sharing the structured sample set by using a dual-channel 3D-CNN to extract radiation intensity features and cloud cluster morphological features, and splicing the radiation intensity features and the cloud cluster morphological features to obtain a fusion feature vector; non-linear mapping based on deep learning is carried out on the fusion feature vector to output a concentration prediction value; and introducing an atmospheric transmission correction factor to correct the concentration predicted value so as to obtain a concentration accurate value. The method has the effect of improving the accuracy of gas leakage concentration calibration.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of infrared thermal imaging, and in particular to a gas leakage quantitative calibration method and system based on multispectral feature fusion. BACKGROUND

[0002] In the industrial fields of petrochemical industry, energy pipeline, power equipment, etc., gas leakage monitoring is a key link to ensure production safety, prevent explosion accidents, and protect the ecological environment. Among them, methane (CH4), sulfur hexafluoride (SF6) and other gases have the characteristics of flammability, explosion and strong greenhouse effect, and the precise quantitative monitoring of their leakage is particularly urgent.

[0003] Infrared thermal imaging technology has become one of the mainstream technologies for gas leakage monitoring due to its advantages of non-contact measurement, strong real-time performance, and long-distance monitoring. Its core principle is to use the absorption characteristics of gas molecules to specific wavelength infrared radiation, capture the radiation difference between gas clouds and background environment by infrared thermal imager, and then inverse the gas concentration.

[0004] However, the existing infrared thermal imaging gas concentration calibration technology is difficult to achieve high-precision quantitative monitoring in actual industrial scene application due to the limitations of the calibration method and the interference of complex environment. The specific technical status is as follows:

[0005] 1. Single-point blackbody calibration method is generally used in the industry to calibrate temperature readings only, but the actual gas absorption process of infrared radiation follows Lambert-Beer's law, but its absorption effect will change with factors such as radiation transmission path length and gas molecule motion state, forming a nonlinear mapping relationship between concentration and radiation. Single-point blackbody calibration can only correct the temperature measurement deviation of the device itself, and cannot solve the concentration-radiation nonlinear mapping process of the gas absorption path.

[0006] 2. There are many interference items in the environment, such as water vapor / dust scattering, which will cause irregular drift of the radiation signal intensity received by the infrared thermal imager, and then cause distortion of the infrared image gray value;

[0007] 3. The shape of the gas cloud is significantly affected by factors such as wind speed, wind direction, and environmental temperature, resulting in high concentration inversion deviation. SUMMARY

[0008] In order to improve the accuracy of gas leakage concentration calibration, the present application provides a gas leakage quantitative calibration method and system based on multispectral feature fusion.

[0009] In the first aspect, the present application provides a gas leakage quantitative calibration method based on multispectral feature fusion, which adopts the following technical scheme:

[0010] ​​A gas leakage quantitative calibration method based on multispectral feature fusion, comprising the following steps:

[0011] Configuring a multi-band blackbody radiation source;

[0012] Building a simulated gas cloud environment and acquiring configuration parameters in real time, and classifying and collecting the configuration parameters to construct a four-dimensional parameter space;

[0013] Obtaining infrared thermal imaging data corresponding to the simulated gas cloud environment, and associating the four-dimensional parameter space to form a structured sample set;

[0014] Using a dual-channel 3D-CNN to share the structured sample set to extract radiation intensity features and cloud shape features and splice them to obtain a fusion feature vector;

[0015] Performing a deep learning-based nonlinear mapping on the fusion feature vector to output a concentration prediction value;

[0016] Introducing an atmospheric transmission correction factor to correct the concentration prediction value to obtain an accurate concentration value.

[0017] In some embodiments, building a simulated gas cloud environment and acquiring configuration parameters in real time, and classifying and collecting the configuration parameters to construct a four-dimensional parameter space, comprises the following steps:

[0018] Configuring a wind speed according to an adjustable turbulence generator and configuring a real-time temperature according to a temperature control air curtain;

[0019] Obtaining real concentration values in the simulated gas cloud environment and obtaining the current position of the thermal infrared imaging device;

[0020] Collecting each of the configuration parameters and classifying them into a four-dimensional parameter space containing gas concentration, observation distance, environmental temperature, and turbulence intensity, and configuring corresponding data ranges for each dimension parameter based on a preset classification data amount.

[0021] In some embodiments, obtaining infrared thermal imaging data corresponding to the simulated gas cloud environment, and associating the four-dimensional parameter space to form a structured sample set, comprises the following steps:

[0022] Collecting a plurality of infrared thermal imaging images corresponding to the configuration parameters;

[0023] Calculating the gray scale matrix of each infrared thermal imaging image and forming a three-dimensional tensor with the corresponding frame time sequence;

[0024] Associating a plurality of three-dimensional tensors with the four-dimensional parameter space to obtain the structured sample set.

[0025] In some embodiments, a double-channel 3D-CNN is used to share the structured sample set to extract radiation intensity features, including the following steps:

[0026] Distance attenuation correction is performed on the gray matrix based on the observation distance;

[0027] Radiation gray calibration is performed on the gray matrix based on the ambient temperature;

[0028] Feature extraction is performed on the corrected gray matrix based on a 3D convolution layer, specifically,

[0029] The first layer extracts the local radiation intensity difference corresponding to the three-dimensional tensor, the second layer extracts the time variation trend corresponding to the radiation intensity, and the third layer extracts the spatial distribution corresponding to the radiation intensity.

[0030] In some embodiments, a double-channel 3D-CNN is used to share the structured sample set to extract cloud cluster morphology features, including the following steps:

[0031] Binary processing is performed on the gray matrix to obtain a cloud cluster space mask, and a corresponding morphological parameter sequence is calculated based on the cloud cluster space mask;

[0032] Feature extraction is performed on the morphological parameter sequence based on a 3D convolution layer, specifically,

[0033] The first layer extracts local morphological changes, the second layer extracts spatial evolution rules of the morphology, and the third layer extracts the correlation between the morphological parameter sequence and the turbulence intensity.

[0034] In some embodiments, an atmospheric transmission correction factor is introduced to correct the concentration prediction value to obtain an accurate concentration value, including the following steps:

[0035] A preset calibration transmittance is obtained;

[0036] A real-time on-site measured transmittance is calculated;

[0037] The atmospheric transmission correction factor is calculated based on the calibration transmittance and the on-site measured transmittance, and the accurate concentration value is calculated by the following formula:

[0038] ,

[0039] wherein, the concentration prediction value is a nonlinear mapping output through deep learning, the atmospheric transmission correction factor, the calibration transmittance is at a wavelength of and a temperature of ​characterized by in-situ measured transmittance at wavelength , temperature .

[0040] In some embodiments, the real-time in-situ measured transmittance is calculated, comprising the steps of:

[0041] acquiring actual water vapor absorption band based on the thermal infrared imaging device and calculating a first gray value;

[0042] acquiring a second gray value of the multi-band blackbody radiation source at the same temperature and the same band;

[0043] calculating the in-situ measured transmittance based on the first gray value and the second gray value in combination with the following formula,

[0044] ,

[0045] wherein, the first gray value represents the first gray value, the second gray value represents the second gray value, and the water vapor transmittance calibration coefficient represents a laboratory-calibrated water vapor transmittance calibration coefficient.

[0046] In some embodiments, the method further comprises the steps of:

[0047] monitoring the wind speed in real time and determining whether it is greater than a wind speed threshold;

[0048] if not, loading a standard model;

[0049] if so, activating a turbulent flow enhancement mode;

[0050] under the standard model, loading a standard weight set in which the weight of radiation intensity is greater than or equal to the weight of cloud cluster morphology;

[0051] when the turbulent flow enhancement mode is activated, loading a high turbulent flow weight set in which the weight of radiation intensity is less than the weight of cloud cluster morphology.

[0052] In some embodiments, an atmospheric transmission correction factor is introduced to correct the concentration prediction value to obtain an accurate concentration value, and the method further comprises the steps of:

[0053] based on the fluctuation rate of the first gray value in the continuous frame number time sequence, dynamically adjusting the correction frequency of the atmospheric transmission correction factor to the concentration prediction value;

[0054] if the fluctuation rate is greater than a preset value, the correction frequency is increased;

[0055] if the fluctuation rate is less than a preset value, the correction frequency is decreased.

[0056] In a second aspect, the application provides a gas leakage quantitative calibration system based on multispectral feature fusion, which adopts the following technical scheme:

[0057] A gas leakage quantitative calibration system based on multispectral feature fusion, comprising:

[0058] A multi-band blackbody radiation source configured;

[0059] A gas cloud simulation cabin for building a simulated gas cloud environment and acquiring configuration parameters in real time;

[0060] An environmental sensor array for classifying and collecting the configuration parameters to construct a four-dimensional parameter space;

[0061] An infrared thermal imager for acquiring infrared thermal imaging data corresponding to the simulated gas cloud environment;

[0062] A feature extraction engine for associating the infrared thermal imaging data with the four-dimensional parameter space to form a structured sample set, and sharing the structured sample set by a dual-channel 3D-CNN to extract radiation intensity features and cloud shape features and splice them to obtain a fusion feature vector;

[0063] A concentration inversion model for performing deep learning-based nonlinear mapping on the fusion feature vector to output a concentration prediction value, and introducing an atmospheric transmission correction factor to correct the concentration prediction value to obtain an accurate concentration value.

[0064] The technical scheme provided by the embodiments of the application has the following technical effects:

[0065] The calibration method is realized by fusing a physical experiment model and data driving, and the nonlinear relationship between "concentration-radiation" is accurately modeled; by multi-channel feature extraction, the radiation intensity features and cloud shape features are fused into the concentration inversion process to realize adaptive calibration of different diffusion modes; by the atmospheric transmission correction factor, the influence of water vapor and dust on radiation transmission is dynamically compensated, and the environmental robustness in the calibration process is improved. BRIEF DESCRIPTION OF DRAWINGS

[0066] Figure 1 is a step schematic diagram of a gas leakage quantitative calibration method based on multispectral feature fusion provided by the embodiments of the application.

[0067] Figure 2 is a logic diagram for adjusting the turbulent flow mode in the gas leakage quantitative calibration method based on multispectral feature fusion provided by the embodiments of the application.

[0068] Figure 3 is a module schematic diagram of the gas leakage quantitative calibration system based on multispectral feature fusion provided by the embodiments of the application. DETAILED DESCRIPTION

[0069] In order to more clearly understand the objects, technical solutions and advantages of the present application, the present application will be described and explained in detail below in connection with the drawings and embodiments. However, it should be understood by those skilled in the art that the present application can be implemented without these details. In some cases, in order to avoid unnecessary description and make aspects of the present application obscure, well-known methods, processes, systems, components and / or circuits that have been described at a higher level will not be described in detail. It is obvious for those skilled in the art that various changes can be made to the embodiments disclosed in the present application, and the general principles defined in the present application can be applied to other embodiments and application scenarios without departing from the principles and scope of the present application. Therefore, the present application is not limited to the embodiments shown, but conforms to the broadest scope claimed by the present application.

[0070] It should be noted that the description of the embodiments is used to help understand the present application, but does not constitute a limitation of the present application. In addition, the technical features involved in each embodiment of the present application described below can be combined with each other as long as they do not conflict with each other.

[0071] In the description of the present application, the meaning of one or more is one or more, the meaning of multiple is two or more, greater than, less than, more than, etc. is understood as not including the number, above, below, etc. is understood as including the number. If it is described as first, second, it is only used to distinguish technical features for the purpose, and cannot be understood as indicating or implying the relative importance of the indicated technical features or implying the number of indicated technical features or implying the order of indicated technical features.

[0072] In the description of the present application, the description of the terms "one embodiment", "some embodiments", "exemplary embodiment", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the described specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a combined manner.

[0073] As Figure 1 shown, the present application discloses a gas leakage quantitative calibration method based on multi-spectral feature fusion, comprising the following steps:

[0074] S100, configuring a multi-band blackbody radiation source.

[0075] In the embodiments of the present application, a double-band independent temperature control design is adopted for the band characteristics of different gas absorption, solving the problem that single-band calibration cannot cover the multi-gas absorption requirement, specifically,

[0076] The band range includes 3-5 μm and 8-12 μm, 3-5 μm is mainly used for monitoring methane (CH4) ), propane (C3H8) ) and other hydrocarbon gases, which have a strong absorption peak at about 3.3 μm; 8-12 μm is mainly used for monitoring sulfur hexafluoride (SF6) ), carbon dioxide (CO2) ) and other gases.

[0077] At the same time, platinum resistance temperature sensors and PID temperature control algorithms are used, and the independent temperature control accuracy of the double-band can reach ±0.1℃, which can output stable blackbody radiation signals, providing accurate reference sources for subsequent calibration of radiation transfer models.

[0078] S200, building a simulated gas cloud environment and acquiring configuration parameters in real time, and classifying and collecting the configuration parameters to build a four-dimensional parameter space.

[0079] The simulated gas cloud environment is built by a gas cloud simulation cabin containing an adjustable turbulence generator and a temperature control air curtain, which simulates the complex environment in the industrial field and provides real and controllable data sets for subsequent model training.

[0080] After the simulation environment is configured, the corresponding configuration parameters are acquired in real time, and the configuration parameters are classified to build a data space containing four-dimensional parameters. The radiation transfer matrix is established through the four-dimensional parameter space, and sample support is provided for the subsequent configuration of the training data set in the full scene.

[0081] S300, acquiring infrared thermal imaging data corresponding to the simulated gas cloud environment, and associating with the four-dimensional parameter space to form a structured sample set.

[0082] The infrared thermal imaging data corresponding to the simulated gas cloud is acquired, and is associated with the corresponding four-dimensional parameter space configuration, wherein a plurality of frames of infrared thermal imaging images are collected under each group of parameter combination, and an association relationship with the actual real concentration value is formed, and finally a structured data sample set is constructed, which is used to extract different feature information according to a plurality of sample data.

[0083] S400, using a double-channel 3D-CNN shared structured sample set to extract radiation intensity features and cloud shape features and to splice to obtain a fusion feature vector.

[0084] According to the structured sample set, the 3D-CNN of two channels is used to extract the radiation intensity features and the cloud shape features respectively. The infrared thermal imaging signal of gas leakage contains the "radiation energy attenuation change" based on the distance and the ambient temperature, and the "spatial shape distribution" affected by the wind speed and the gas characteristics, which are two dimensions that cannot be ignored. The current technology does not consider the single feature extraction under the interference of wind speed, wind direction, environment, etc. Therefore, it cannot fully represent the gas concentration information.

[0085] The absorption ability of the gas to the infrared radiation, and the distribution and diffusion state of the gas in space are combined to extract the features on both sides and merge the vectors, so that the learning and reasoning ability of the model is more comprehensive and accurate.

[0086] S500, the fusion feature vector is subjected to nonlinear mapping based on deep learning to output a concentration prediction value.

[0087] The deep learning module learns the mapping relationship between the four-dimensional parameter space composed of the real configuration parameters of the laboratory and the feature vectors corresponding to the actually measured gas concentration, establishes a nonlinear model from "original gray data" to "preliminary concentration estimation", and finally outputs the preliminary estimated concentration value without atmospheric transmission correction.

[0088] S600, introducing an atmospheric transmission correction factor to correct the concentration prediction value to obtain an accurate concentration value.

[0089] During the transmission of infrared radiation from the gas cloud to the thermal imaging device, it will be absorbed or scattered by water vapor, dust, carbon dioxide and other media in the atmosphere (i.e. atmospheric attenuation). The core logic of concentration inversion is to "reverse the gas concentration through the radiation intensity attenuation", so atmospheric attenuation will cause the radiation intensity of the thermal infrared imaging device to deviate from the true absorption level of the gas. Therefore, in order to make the final result more accurate, it is necessary to quantify and eliminate the radiation attenuation caused by non-target gases through atmospheric transmission correction.

[0090] Through the above method, a calibration method combining physical experiment model and data-driven is realized, and a precise modeling of the nonlinear relationship between "concentration-radiation" is realized. Through multi-channel feature extraction, the radiation intensity features and cloud shape features are fused into the concentration inversion process to realize adaptive calibration of different diffusion shapes. Through the atmospheric transmission correction factor, the influence of water vapor and dust on radiation transmission is dynamically compensated, and the environmental robustness in the calibration process is improved.

[0091] In some other embodiments, a simulated gas cloud environment is built and real-time configuration parameters are obtained, and the configuration parameters are classified and collected to build a four-dimensional parameter space, including the following steps:

[0092] S210, configure the wind speed according to the adjustable turbulence generator, and configure the real-time temperature according to the temperature control air curtain.

[0093] The adjustable turbulence generator is designed by combining a multi-fan array with an air flow guide plate, which can achieve continuous adjustment of wind speed and direction from 0 to 15 m / s, covering common industrial wind conditions from no wind (0 m / s) to strong wind (15 m / s). At the same time, by controlling the fluctuation amplitude of fan speed, gear changes between different turbulence intensities are realized, accurately simulating the diffusion form of gas clouds under different turbulence states.

[0094] The temperature-controlled air curtain uses semiconductor temperature control technology to set a temperature-controlled air curtain at the inlet and outlet of the simulation cabin, achieving precise control of the cabin environment temperature from -30 to 60°C. This setting can simulate daily environments (pipeline environments of -5-30°C), winter cold (northern pipeline environments of -30°C), and summer heat (exposure environments of 60°C in the tank area), ensuring the environmental adaptability of the calibration data.

[0095] S220, obtaining the real concentration value in the simulated gas cloud environment and obtaining the current position of the thermal infrared imaging device.

[0096] The real concentration value is accurately controlled by a high-precision gas command controller, which accurately controls the gas injection rate and collects the corresponding real concentration.

[0097] In the cloud simulation cabin, an infrared thermal imager can be configured to move towards or away from the gas outlet of the simulation cabin, and the current position is collected in real time based on the position of the infrared thermal imaging device on the guide rail. At the same time, it is necessary to ensure that the observation angle remains perpendicular to the gas diffusion direction.

[0098] S230, respectively collecting and classifying each configuration parameter into a four-dimensional parameter space containing gas concentration, observation distance, environmental temperature, and turbulence intensity, and configuring corresponding data ranges for each dimension parameter based on a preset classification data amount.

[0099] According to the real concentration value, the "gas concentration" dimension is configured, and the setting basis is to cover the full range from trace leakage (0.1% LEL, below the safety threshold) to high-concentration leakage (100% LEL, close to the risk of explosion), and the gradient point interval increases with the increase of concentration, ensuring the monitoring accuracy at low concentration leakage. The data amount in this dimension is >20 gradient points, and the data range is 0.1%-100% LEL.

[0100] According to the current position, the "observation distance" dimension is configured, and the setting basis is to cover common monitoring distances in industrial sites (such as 1m for close-range inspection in tank areas and 20m for long-distance monitoring of pipelines), avoiding the problem of decreased accuracy at long distances caused by single-distance calibration. The data amount in this dimension is 4 points, and the data range is 1m, 5m, 10m, and 20m, respectively.

[0101] According to the real-time temperature configuration "environmental temperature" dimension, the setting basis covers extreme temperatures (extremely high temperature, extremely low temperature) in industrial sites, and at the same time covers normal temperature benchmarks and freezing point critical temperatures. The data quantity under this dimension is 4 points, and the data ranges are -20℃, 0℃, 25℃, and 50℃, respectively.

[0102] According to the wind speed configuration "turbulence intensity" dimension, the setting basis covers common main wind conditions in industrial sites, with low turbulence corresponding to a breeze environment, medium turbulence corresponding to a moderate wind environment, and high turbulence corresponding to a strong wind environment. The data quantity under this dimension is 3 points, and the data ranges are low (δ = 0.1, corresponding to a wind speed less than 2 m / s), medium (δ = 0.3, corresponding to a wind speed between 2-8 m / s), and high (δ = 0.5, corresponding to a wind speed greater than 8 m / s).

[0103] The core function of the gas cloud simulation cabin is to reproduce a controllable environment of industrial scenes. Therefore, through the turbulence generator, temperature-controlled air curtain, guide rail, and other devices, the observation distance, temperature, wind speed, and other parameters have been accurately controlled. Finally, the accurate control of "real concentration" is the key link. Only when the concentration, distance, temperature, and turbulence four-dimensional parameters are controllable real values, can the infrared thermal imaging data collected accurately correspond to the influence of a single variable, providing pure samples for the subsequent extraction of double-channel features by the model.

[0104] In other embodiments, infrared thermal imaging data corresponding to the simulated gas cloud environment is acquired and associated with the four-dimensional parameter space to form a structured sample set, including the following steps:

[0105] S310, a plurality of frames of infrared thermal imaging images corresponding to the configuration parameters are collected.

[0106] S320, the gray matrix of each infrared thermal imaging image is calculated and associated with the corresponding frame time sequence to form a three-dimensional tensor.

[0107] S330, a plurality of three-dimensional tensors are associated with the thinking parameter space to obtain a structured sample set.

[0108] In the structured sample set, the three-dimensional tensor [H, W, T] (H is the image height, W is the image width, and T is the frame number of the time sequence) corresponding to the infrared thermal imaging contains the spatial gray distribution and time variation information of the gas cloud layer. The observation distance is used to correct the attenuation effect of distance on radiation intensity (radiation intensity is inversely proportional to the square of distance). The environmental temperature is used to correct the interference of the environmental background radiation on the radiation signal of the gas cloud (the higher the environmental temperature, the stronger the background radiation, and the lower the radiation contrast of the gas cloud). The turbulence intensity is used to guide the extraction weight of the shape feature (the higher the turbulence intensity, the greater the weight of the shape feature, to highlight the influence of shape change on concentration).

[0109] In some other embodiments, using a dual-channel 3D-CNN to share a structured sample set to extract radiation intensity features includes the following steps:

[0110] S410, performing distance attenuation correction on the grayscale matrix based on the observation distance.

[0111] S411, performing radiometric grayscale calibration on the grayscale matrix based on the ambient temperature.

[0112] S412: extract features from the corrected grayscale matrix based on a 3D convolutional layer.

[0113] Specifically,

[0114] S413, the first layer extracts the local radiation intensity difference corresponding to the three-dimensional tensor, the second layer extracts the time variation trend corresponding to the radiation intensity, and the third layer extracts the spatial distribution corresponding to the radiation intensity.

[0115] First, the radiation intensity characteristics are based on the original grayscale matrix value size and time changes, combined with observation distance, ambient temperature and other parameters to extract and analyze the gas's absorption intensity of infrared radiation (for example, the larger the grayscale difference, the stronger the absorption, and the higher the indirect reaction concentration).

[0116] Specifically, the grayscale value is first corrected for distance attenuation by observing the distance (the formula is: grayscale value after correction = original grayscale value * ), to compensate for the natural attenuation of infrared radiation with distance;

[0117] Then, according to the ambient temperature, subtract the grayscale value corresponding to the ambient background radiation at the same temperature (calibrated by the grayscale value of the radiation of a multi-spectral blackbody at the same ambient temperature);

[0118] Finally, after distance correction and temperature calibration, the radiation grayscale matrix contributed only by the gas cloud is obtained.

[0119] At the same time, the corrected radiation grayscale matrix is ​​subjected to feature extraction through three layers of 3D convolutional layers (convolution kernel size is 3*3*3, step size is 1*1*1). The first layer extracts local radiation intensity differences, such as the grayscale difference between the center and edge of the gas cloud; the second layer extracts the temporal variation trend of the radiation intensity, such as the decay rate of the grayscale value during the diffusion of the cloud, and also includes the fluctuation amplitude over time; the third layer extracts the spatial distribution pattern of the radiation intensity, such as the morphology of the high grayscale area in the center of the cloud, and finally outputs a 256-dimensional radiation intensity feature vector.

[0120] In some other embodiments, using a dual-channel 3D-CNN to share a structured sample set to extract cloud morphological features includes the following steps:

[0121] S420, binarize the gray matrix to obtain a cloud cluster space mask and calculate a corresponding morphological parameter sequence based on the cloud cluster space mask.

[0122] S421, perform feature extraction on the morphological parameter sequence based on a 3D convolution layer.

[0123] Specifically,

[0124] S422, the first layer extracts local morphological changes, the second layer extracts spatial evolution rules of the morphology, and the third layer extracts the correlation between the morphological parameter sequence and the turbulence intensity.

[0125] The cloud cluster morphological feature is based on the spatial distribution of the original gray matrix to calculate the morphological parameters of the cloud cluster, and corrects the influence of the diffusion morphology on the concentration inversion.

[0126] First, morphological preprocessing is performed, the gray matrix is binarized (based on adaptive threshold segmentation, distinguishing between gas clusters and background), and a cloud cluster space mask matrix is obtained, wherein 1 represents gas and 0 represents background; Further, the space mask matrix can be subjected to an open operation (first erosion and then expansion) to remove isolated small pixel points caused by dust and noise and retain the complete cloud cluster outline.

[0127] Based on the space mask matrix, the area (pixel count, reflecting the scale of the leakage), circularity, and eccentricity of the cloud cluster are calculated.

[0128] Based on the Sobel operator, the gray scale gradient of the cloud cluster edge is calculated, and the mean value (reflecting the edge sharpness, high turbulence under the coating gradient) and the variance (reflecting the irregularity of the edge) of the edge gradient are calculated.

[0129] Finally, the area expansion rate and center offset of the cloud cluster are calculated based on the time series mask to form the final morphological parameter sequence.

[0130] At the same time, the same 3D convolution structure as the radiation intensity feature channel is used to perform feature extraction on the morphological parameter sequence, the first layer extracts local morphological changes, such as the concave-convex degree of the cloud cluster edge; the second layer extracts the time evolution rule of the morphology, such as the fragmentation speed of the cloud cluster in a turbulent environment; the third layer extracts the correlation feature between the morphology and the turbulence intensity, such as the irregularity degree of the cloud cluster under high turbulence. Finally, a 256-dimensional cloud cluster morphological feature vector is output.

[0131] When the feature extraction of the two channels is completed, the two need to be normalized by batch normalization to eliminate the dimensional difference, and the normalized radiation intensity features and cloud shape features are spliced and fused by the Concatenate() splicing function to obtain a full-dimensional feature vector that takes into account the "radiation-morphology-environment". After splicing, a full connection layer and a ReLU activation function are added to perform nonlinear transformation on the fused features, and finally the optimized fused features are output as the input of the subsequent deep learning estimation.

[0132] In some other embodiments, the atmospheric transmission correction factor is introduced to modify the concentration prediction value to obtain the concentration accurate value, including the following steps:

[0133] S610, obtaining a preset calibration transmittance.

[0134] S620, calculating a real-time on-site measured transmittance.

[0135] S630, calculating an atmospheric transmission correction factor based on the calibration transmittance and the on-site measured transmittance, and calculating the concentration accurate value through the following formula:

[0136] .

[0137] Wherein, the concentration prediction value is output by the deep learning nonlinear mapping, the atmospheric transmission correction factor, the calibration transmittance is at a wavelength of and a temperature of . the on-site measured transmittance is at a wavelength of and a temperature of .

[0138] By introducing the atmospheric transmission correction factor, the deep learning concentration estimation result is combined with the physical mechanism to solve the difference between the on-site environment and the laboratory calibration environment.

[0139] Specifically, first, the deep learning concentration prediction value is obtained, which is obtained by inputting the 512-dimensional fused feature vector extracted by multiple features into a full connection layer, and outputting a preliminary concentration estimation value. This value is trained based on a laboratory calibrated data set and reflects the nonlinear mapping relationship between the fused features and the concentration, but does not take into account the transmission loss of the on-site environment.

[0140] The calibration transmittance is calculated by combining the Lambert-Bill law with the radiation signal of a multi-spectral black body at different wavelengths and different temperatures in the laboratory calibration link, ensuring that there is no additional environmental interference. The specific calculation process is: wherein, is the absorption coefficient of the gas at wavelength , temperature , is the true concentration in the calibration chamber, is the radiation transmission path length. The calibration transmissivity is a fixed value, which is stored in the system database as a reference for transmissivity.

[0141] characterized as the on-site measured transmissivity, reflecting the degree of attenuation of water vapor and dust on infrared radiation on site, which is obtained by collecting the water vapor absorption band on site by an infrared thermal imaging device and inversion, and the value is the object of real-time calculation, reflecting the influence of water vapor and dust on radiation transmission.

[0142] The final concentration C is the conversion result from the "laboratory calibration result" to the "actual concentration on site" by multiplying the deep learning estimated value and the atmospheric transmission correction factor, which corrects the calculation error caused by the interference of the on-site environment.

[0143] The logic when corrected by the atmospheric transmission correction factor is exemplified as follows:

[0144] If the on-site environmental interference is strong (such as more water vapor and more dust), then small (serious radiation attenuation), The ratio is greater than 1, at this time the measured value will be greater than the predicted value, which means that the preliminary estimated value is underestimated due to radiation attenuation, and needs to be enlarged to the true concentration by the correction factor;

[0145] If the on-site environment is clean (dry and windless), then The value of is close to the value of is close to 1, so it is characterized that the preliminary estimated value is close to the true concentration, and the correction amplitude is small.

[0146] In other embodiments, the real-time on-site measured transmissivity is calculated, including the following steps:

[0147] S621, based on the thermal infrared imaging device to collect the actual water vapor absorption band and calculate the first gray value.

[0148] S622, obtaining the second gray value of the preset multi-band blackbody radiation source at the same temperature and the same wave band.

[0149] S623, based on the first gray value and the second gray value, the on-site measured transmissivity is calculated according to the following formula,

[0150] ,

[0151] wherein, characterized as a first gray value, characterized as a second gray value, characterized as a water vapor transmittance calibration coefficient calibrated in the laboratory.

[0152] When calculating the on-site measured transmittance, the signal in the water vapor absorption band (6.5 μm, which has strong water vapor absorption and no overlap with the absorption band of most monitoring gases) is obtained by an infrared thermal imaging device. The first gray value is characterized as the gray value of the on-site water vapor at 6.5 μm, and the second gray value is characterized as the gray value of the multi-spectral blackbody at the same temperature and the same 6.5 μm band.

[0153] In some other embodiments, the following steps are further included:

[0154] S700, real-time monitoring of wind speed and determining whether it is greater than a wind speed threshold.

[0155] S710, if not, load the standard model.

[0156] S720, if greater, activate the turbulence enhancement mode.

[0157] S730, under the standard model, load the standard weight set whose radiation intensity weight is greater than or equal to the cloud cluster morphology weight.

[0158] S740, when the turbulence enhancement mode is activated, load the high turbulence weight set whose radiation intensity is less than the cloud cluster morphology weight.

[0159] In order to realize the accurate adaptation of gas concentration inversion under different turbulence intensities, and solve the problems of morphology interference and concentration deviation caused by turbulence in the prior art, different processing modes under different turbulence states are further realized in the embodiments of the present application:

[0160] First, the core index of turbulence intensity is wind speed, so the wind speed data of the calibration environment is first collected in real time as the basis for judging the turbulence intensity.

[0161] Second, compare the collected wind speed with the preset wind speed (2 m / s in this application). When the wind speed is less than the preset value, it corresponds to a low turbulence environment. At this time, the gas cloud diffusion state is stable, and the morphology has less interference on the concentration inversion. At this time, the standard model is loaded, and standardized processing is performed during subsequent feature extraction and concentration inversion. When the wind speed exceeds the preset value, it corresponds to a medium / high turbulence environment. At this time, the gas cloud is significantly disturbed by the airflow, and the morphology presents a flat, broken or rapid diffusion state, and the morphology has a large interference on the concentration inversion. The turbulence enhancement mode needs to be started, and customized correction needs to be performed during subsequent feature extraction and concentration inversion in this mode.

[0162] Finally, in different modes, the difference lies in the weight ratio between multiple features when subsequent feature extraction is performed.

[0163] When the standard model is used to learn the mapping relationship between "radiance intensity-concentration" under "low turbulence + stable morphology", the model parameters focus more on the absolute value and spatial distribution of the gray value. When feature extraction is performed, the weight corresponding to the morphology feature is relatively small, reducing unnecessary morphology correction, or the weight of the two features is the same.

[0164] When the turbulence enhancement mode is activated, the correlation between "morphology change rate-concentration" under "medium / high turbulence + dynamic morphology" is learned, and the dynamic fragmentation features of the cloud cluster are extracted. When feature extraction is performed, the weight ratio of the morphology feature is increased, focusing on capturing dynamic morphology information such as cloud cluster edge fragmentation degree and diffusion speed, while the weight ratio of the radiation intensity feature is appropriately reduced.

[0165] During calibration, the system continuously monitors the wind speed change (rejudges the threshold every 0.5 seconds), and if the wind speed decreases from >2m / s to ≤2m / s (such as the end of the gust), it automatically switches from "turbulence enhancement mode" to "standard mode", and synchronously updates the model weight; At the same time, the "gray data-real concentration" sample under the current turbulence state is returned to the database, and the weight parameters of the two modes are fine-tuned regularly (such as every month), to improve the smoothness of mode switching and the accuracy of calibration.

[0166] Through the simple and efficient logic of "wind speed monitoring-threshold judgment-mode switching", "complex turbulence interference" is converted into "quantifiable and dynamically adaptable" calibration strategy, solving the pain point of "fixed model cannot adapt to variable turbulence" in the prior art. Its actual effect not only lies in the significant improvement of concentration inversion accuracy, but also through lightweight design and rapid response, enhances the engineering practicability of the technical solution in industrial complex scenes (such as oil storage tank area, open pipeline), which is the key technical link of the present application "breakthrough turbulence adaptability bottleneck".

[0167] In other embodiments, the atmospheric transmission correction factor is introduced to correct the concentration prediction value to obtain the concentration accurate value, which further includes the following steps:

[0168] S640, dynamically adjusting the correction frequency of the atmospheric transmission correction factor to the concentration prediction value based on the fluctuation rate of the first gray value in the continuous frame time sequence.

[0169] S641, if the fluctuation rate is greater than the preset value, the correction frequency is increased.

[0170] S642, if the fluctuation rate is less than the preset value, the correction frequency is reduced.

[0171] Because the measured transmittance on site will change dynamically with the environment on site, the real-time performance exists, and therefore, in order to ensure timely compensation for the amplitude attenuation caused by rain, fog and humidity changes, the fluctuation request of the transmittance on site is further used to dynamically adjust the correction frequency in the embodiments of the application, so as to avoid the problem that if the environment is stable, redundant calculation will be performed when a fixed frequency is used, the inversion speed is reduced, and the deviation is increased when the environment mutates and correction is not timely.

[0172] The relative standard deviation is calculated by calculating the gray value mean and gray standard deviation of each pixel point in the first gray value of multiple frames of images in the sliding window, the fluctuation rate is judged according to the difference of the relative standard deviation, the fluctuation rate can directly reflect the change of the environmental humidity (such as temperature rise-water vapor absorption enhancement-gray value reduction), is not affected by the concentration of the target gas, and can purely represent the environmental stability.

[0173] When the fluctuation rate is less than the preset value, it represents that the environment is stable, and the correction frequency can be appropriately reduced to reduce the calculation amount, such as extending the correction frequency from 0.5s to 2s; when the fluctuation rate is not less than the preset value, it represents that the environment mutates, and the correction frequency needs to be appropriately increased to ensure timely correction, such as shortening the correction frequency from 1s to 0.5s.

[0174] The application further discloses a gas leakage quantitative calibration system based on multi-spectral feature fusion, characterized by comprising:

[0175] A multi-band blackbody radiation source is configured.

[0176] A gas cloud simulation cabin is configured to build a simulated gas cloud environment and acquire configuration parameters in real time.

[0177] An environmental sensor array is configured to collect the configuration parameters to construct a four-dimensional parameter space.

[0178] An infrared thermal imager is configured to acquire infrared thermal imaging data corresponding to the simulated gas cloud environment.

[0179] A feature extraction engine is configured to associate the infrared thermal imaging data with the four-dimensional parameter space to form a structured sample set, and share the structured sample set by using a double-channel 3D-CNN to extract radiation intensity features and cloud shape features and splice them to obtain a fusion feature vector.

[0180] A concentration inversion model is configured to perform a deep learning-based nonlinear mapping on the fusion feature vector to output a concentration prediction value, and introduce an atmospheric transmission correction factor to correct the concentration prediction value to obtain an accurate concentration value.

[0181] The implementation principle is:

[0182] The fusion of physical experimental model and data-driven calibration method is realized, the nonlinear relationship of "concentration-radiation" is accurately modeled, the radiation intensity features and cloud shape features are fused into the concentration inversion process through multi-channel feature extraction, the adaptive calibration of different diffusion shapes is realized, and the influence of water vapor and dust on radiation transmission is dynamically compensated through atmospheric transmission correction factor, and the environmental robustness in the calibration process is improved.

[0183] It should be understood that although each step in the flowchart of the accompanying drawings is shown in sequence according to the direction of the arrow, these steps are not necessarily executed in sequence according to the direction of the arrow. Unless explicitly stated herein, the execution of these steps is not strictly limited in sequence, and they can be executed in other sequences.

[0184] The above are preferred embodiments of the present application, and are not intended to limit the protection scope of the present application, therefore: any equivalent changes made on the structure, shape, principle of the present application should be covered within the protection scope of the present application.

Claims

1. A gas leakage quantitative calibration method based on multispectral feature fusion, characterized in that, The method comprises the following steps: configuring a multi-band blackbody radiation source; building a simulated gas cloud environment and acquiring configuration parameters in real time, and classifying and collecting the configuration parameters to construct a four-dimensional parameter space; acquiring infrared thermal imaging data corresponding to the simulated gas cloud environment, and associating the four-dimensional parameter space to form a structured sample set; sharing the structured sample set using a dual-channel 3D-CNN to extract radiation intensity features and cloud shape features and splice them to obtain a fusion feature vector; performing a deep learning-based nonlinear mapping on the fusion feature vector to output a concentration prediction value; introducing an atmospheric transmission correction factor to correct the concentration prediction value to obtain an accurate concentration value.

2. The gas leakage quantitative calibration method based on multispectral feature fusion according to claim 1, characterized in that, The method of building a simulated gas cloud environment and acquiring configuration parameters in real time, and classifying and collecting the configuration parameters to construct a four-dimensional parameter space comprises the following steps: configuring the wind speed according to the adjustable turbulence generator, and configuring the real-time temperature according to the temperature control air curtain; acquiring the real concentration value in the simulated gas cloud environment and acquiring the current position of the thermal infrared imaging device; collecting each of the configuration parameters and classifying them into a four-dimensional parameter space containing gas concentration, observation distance, environmental temperature, and turbulence intensity, and configuring a corresponding data range for each dimension parameter based on a preset classification data amount.

3. The gas leakage quantitative calibration method based on multispectral feature fusion according to claim 2, characterized in that, The method of acquiring infrared thermal imaging data corresponding to the simulated gas cloud environment, and associating the four-dimensional parameter space to form a structured sample set comprises the following steps: collecting a plurality of infrared thermal imaging images corresponding to the configuration parameters; calculating the gray matrix of each infrared thermal imaging image and forming a three-dimensional tensor with the corresponding frame time sequence; associating a plurality of the three-dimensional tensors with the four-dimensional parameter space to obtain the structured sample set.

4. The gas leakage quantitative calibration method based on multispectral feature fusion according to claim 3, characterized in that, The method of sharing the structured sample set using a dual-channel 3D-CNN to extract radiation intensity features comprises the following steps: performing distance attenuation correction on the gray matrix based on the observation distance; performing radiation gray calibration on the gray matrix based on the environmental temperature; extracting features from the corrected gray matrix based on a 3D convolution layer, specifically, the first layer extracts local radiation intensity differences corresponding to the three-dimensional tensor, the second layer extracts time variation trends corresponding to the radiation intensity, and the third layer extracts spatial distribution of the radiation intensity.

5. The gas leakage quantitative calibration method based on multispectral feature fusion according to claim 3, characterized in that, The method of sharing the structured sample set using a dual-channel 3D-CNN to extract cloud shape features comprises the following steps: performing binary processing on the gray matrix to obtain a cloud space mask and calculating a corresponding morphological parameter sequence based on the cloud space mask; extracting features from the morphological parameter sequence based on a 3D convolution layer, specifically, the first layer extracts local morphological changes, the second layer extracts spatial evolution rules of the morphology, and the third layer extracts the correlation between the morphological parameter sequence and the turbulence intensity.

6. The multispectral feature fusion based gas leak quantification calibration method of claim 1, wherein, The method of introducing an atmospheric transmission correction factor to correct the concentration prediction value to obtain an accurate concentration value comprises the following steps: acquiring a preset calibration transmittance; calculating a real-time on-site measured transmittance; The atmospheric transmission correction factor is calculated based on the calibrated transmittance and the measured transmittance, and the concentration accurate value is calculated by the following formula: ， wherein, the concentration prediction value characterized as an output of a deep learning nonlinear mapping, is the atmospheric transmission correction factor, the calibration transmittance characterized as a transmittance at a wavelength of , and a temperature of , the in-situ measured transmittance characterized as a transmittance at a wavelength of , and a temperature of .

7. The gas leakage quantitative calibration method based on multispectral feature fusion according to claim 6, characterized in that, The real-time measured transmittance is calculated, including the following steps: The actual water vapor absorption band is collected by the thermal infrared imaging device, and a first gray value is calculated; A second gray value of the multi-band blackbody radiation source at the same temperature and the same band is obtained; The measured transmittance is calculated based on the first gray value and the second gray value combined with the following formula, ， wherein characterized by the first grey value, characterized by the second grey value, characterized by a water vapor transmissivity calibration factor of a laboratory calibration.

8. The gas leakage quantitative calibration method based on multispectral feature fusion according to claim 2, characterized in that, Further comprising the following steps: The wind speed is monitored in real time, and it is judged whether it is greater than the wind speed threshold; If not, load the standard model; If greater, activate the turbulent enhancement mode; Under the standard model, load the standard weight set whose radiation intensity weight is greater than or equal to the cloud cluster shape weight; When the turbulent enhancement mode is activated, load the high turbulent weight set whose radiation intensity is less than the cloud cluster shape weight.

9. The multispectral feature fusion based gas leak quantification calibration method of claim 1, wherein, The concentration prediction value is corrected by introducing the atmospheric transmission correction factor to obtain the concentration accurate value, further comprising the following steps: The correction frequency of the concentration prediction value by the atmospheric transmission correction factor is dynamically adjusted based on the fluctuation rate of the first gray value in the continuous frame number time sequence; If the fluctuation rate is greater than the preset value, the correction frequency is increased; If the fluctuation rate is less than the preset value, the correction frequency is reduced.

10. A gas leakage quantitative calibration system based on multispectral feature fusion, characterized in that, It includes: A configured multi-band blackbody radiation source; A gas cloud simulation cabin for building a simulated gas cloud environment and obtaining configuration parameters in real time; An environmental sensor array for classifying and collecting the configuration parameters to construct a four-dimensional parameter space; An infrared thermal imager for obtaining infrared thermal imaging data corresponding to the simulated gas cloud environment; A feature extraction engine for associating the infrared thermal imaging data with the four-dimensional parameter space to form a structured sample set, and using a double-channel 3D-CNN to share the structured sample set to extract radiation intensity features and cloud cluster shape features and splice them to obtain a fusion feature vector; A concentration inversion model for performing deep learning-based nonlinear mapping on the fusion feature vector to output a concentration prediction value, and introducing an atmospheric transmission correction factor to correct the concentration prediction value to obtain a concentration accurate value.

Images