Intelligent calibration method of refrigeration type infrared thermal imaging gas leakage detection system
By dynamically compensating for environmental parameters and fusing multimodal features, the calibration lag and sensitivity issues of infrared gas detection systems in complex environments have been resolved, enabling real-time and accurate detection of gas leaks. This technology is suitable for high-risk fields such as petrochemicals and natural gas transportation.
Patent Information
- Application Number
- CN202510880085.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-27
- Publication Date
- 2025-10-17
AI Technical Summary
Traditional infrared gas detection systems are greatly affected by changes in ambient temperature and humidity, suffer from calibration delays, cannot adapt to the field environment in real time, have low sensitivity to minor leaks, are easily affected by light and smoke, require a large amount of labeled data for deep learning models, while infrared gas leak data is scarce, and performance evaluation indicators are not optimized for dynamic gas diffusion scenarios.
By dynamically compensating for environmental parameters and fusing multimodal features, including temperature drift compensation, motion artifact elimination, baseline dynamic correction, image spatiotemporal fusion, matched filtering enhancement, and leakage quantification analysis, combined with temperature and humidity sensors, an improved YOLOv8 model, and a Kalman filter, real-time calibration and accurate leakage assessment are achieved.
It significantly improves the detection accuracy and adaptability in complex industrial scenarios, enhances the real-time detection capability of gas leaks, and is suitable for high-risk fields such as petrochemicals and natural gas transportation.
Smart Images
Figure CN120800675A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application belongs to the technical field of gas detection, and particularly relates to an intelligent calibration method of a refrigeration type infrared thermal imaging gas leakage detection system. BACKGROUND
[0002] The infrared gas detection equipment is an instrument for detecting gas concentration, leakage and the like by using infrared principle. The traditional infrared thermal imaging detection relies on the difference between the gas and the background temperature, but the change of the environmental temperature and humidity will significantly affect the sensitivity of the detector, leading to false positives / missed reports (for example, the increase of the humidity will interfere with the infrared radiation of the gas absorption band); and the existing calibration relies on the laboratory calibration, cannot adapt to the fluctuation of the on-site environment in real time, and needs frequent manual intervention; and the traditional frame difference method and the optical flow method have low sensitivity to small leakage and are easily disturbed by light, smoke and the like; the deep learning model needs a large amount of labeled data, and the infrared gas leakage data is scarce; and the performance evaluation indexes (such as NETD and MRTD) of the existing refrigeration type infrared imaging system are only for static targets and are not optimized for the dynamic scene of gas diffusion.
[0003] Therefore, the above problems are further improved. SUMMARY
[0004] The main purpose of the present application is to provide an intelligent calibration method of a refrigeration type infrared thermal imaging gas leakage detection system, which solves the problems of environmental interference sensitivity and calibration lag in the traditional infrared gas detection by means of dynamic compensation of environmental parameters and multi-modal feature fusion, significantly improves the detection accuracy and adaptability in complex industrial scenes, and is suitable for leakage monitoring in high-risk fields such as petroleum and chemical industry and natural gas transportation.
[0005] To achieve the above purpose, the present application discloses an intelligent calibration method of a refrigeration type infrared thermal imaging gas leakage detection system, comprising the following steps: Step S1: dynamic compensation of environmental parameters, temperature drift compensation, motion artifact elimination and baseline dynamic correction processing are respectively performed; Step S2: multi-modal data fusion, image space-time fusion, matching filter enhancement and leakage quantization analysis processing are respectively performed.
[0006] As a further preferred technical solution of the above technical solution, step S1 is specifically implemented as: Step S1.1: for temperature drift compensation, the environmental parameters are collected in real time by a temperature and humidity sensor, the working temperature of the detector is combined, a non-linear regression model of temperature-radiation attenuation is established, and the infrared radiation value is dynamically corrected based on the following formula: , is the original radiation intensity, and are the temperature and humidity change amounts, And To calibrate the temperature and humidity compensation coefficient through experiments, The calibrated leakage area radiation intensity, specifically: Real-time data acquisition, the sensor updates the environmental parameters synchronously every frame; Model iteration, optimize the compensation coefficient according to historical data to adapt to long-term temperature drift; Detector adjustment, dynamically adjust the integration time and gain based on environmental parameters to compensate for radiation attenuation and ensure that the detector works in the best linear region; Step S1.2: For motion artifact elimination, combine the optical flow method with the improved YOLOv8 model to distinguish between real leakage diffusion and background motion noise, wherein: For the optical flow method, calculate the pixel motion vector of adjacent frames, filter out the continuous diffusion features, and perform K-means clustering on the optical flow results to retain areas with consistent diffusion direction; For the improved YOLOv8 model, add a time attention module in the target detection layer to strengthen the temporal correlation of the leakage area and suppress transient interference; Step S1.3: For baseline dynamic correction, dynamically adjust the baseline radiation value based on environmental parameters to eliminate long-term drift of background infrared radiation, the formula is: , is the initial baseline value, And is the time decay coefficient, is the air pressure compensation factor, is the air pressure change, specifically: Sliding window mean, update the baseline every 30 seconds, use a 5-minute window to smooth environmental fluctuations; Adaptive filtering, predict the baseline trend through a Kalman filter for real-time correction.
[0007] As a further preferred technical solution of the above technical solution, step S2 is specifically implemented as: Step S2.1: For image spatio-temporal fusion, for the compensated and calibrated data, to improve the spatio-temporal consistency of the leakage area, specifically: align the calibrated infrared images of the last 5 frames, and weighted average to suppress random noise; and through SIFT feature point matching, superimpose the infrared leakage area onto the visible light image to assist in positioning; Step S2.2: For matched filter enhancement, for the compensated and calibrated data, enhance the signal-to-noise ratio of the gas feature absorption band, specifically: design a Gaussian type matched filter to suppress background radiation in other bands, thereby matching the band; and dynamically adjust the filter bandwidth according to environmental parameters to adapt to absorption peak shift under different temperature and humidity; Step S2.3: For the leakage quantification analysis, the compensated calibrated data is combined with the calibration results to achieve accurate leakage evaluation, specifically: Column concentration calculation: Based on the corrected radiation value, the gas concentration is inverted through the Beer-Lambert law: , is the gas absorption coefficient, is the optical path length, is expressed as the reference radiation intensity; Step S3.3: For the leakage quantification analysis, the motion vector of the leakage area is extracted by the optical flow method, and the diffusion speed and direction are calculated.
[0008] As a further preferred technical solution of the above technical solution, for step S3.3, the alarm threshold is dynamically adjusted according to the leakage area, diffusion speed and column concentration.
[0009] As a further preferred technical solution of the above technical solution, the detector is a medium-wave refrigeration-type infrared detector, which is used to obtain an infrared characteristic absorption image of gas leakage. BRIEF DESCRIPTION OF DRAWINGS
[0010] Figure 1 is a method flowchart of the present application. DETAILED DESCRIPTION
[0011] The following description is used to disclose the present application so that those skilled in the art can implement the present application. The preferred embodiments in the following description are only used as examples, and other obvious modifications can be thought of by those skilled in the art. The basic principles of the present application defined in the following description can be applied to other embodiments, modifications, improvements, equivalents and other technical solutions without departing from the spirit and scope of the present application.
[0012] In the preferred embodiments of the present application, those skilled in the art should note that the detector and the like involved in the present application can be regarded as prior art.
[0013] Preferred embodiments.
[0014] As Figure 1 shown, the present application discloses an intelligent calibration method of a refrigeration-type infrared thermal imaging gas leakage detection system, comprising the following steps: Step S1: Environmental parameter dynamic compensation, temperature drift compensation, motion artifact elimination and baseline dynamic correction processing are performed respectively; Step S2: Multimodal data fusion, image space fusion, matching filter enhancement and leakage quantification analysis processing are performed respectively.
[0015] Specifically, step S1 is specifically implemented as: Step S1.1: For temperature drift compensation, real-time acquisition of environmental parameters by temperature and humidity sensor, combined with the working temperature of the detector, a nonlinear regression model of temperature-radiation attenuation is established, and the model dynamically corrects the infrared radiation value based on the following formula: , is the original radiation intensity, and are the temperature and humidity change amounts, and are the temperature and humidity compensation coefficients calibrated by experiment, is the calibrated radiation intensity of the leakage area (unit: W / m²·sr·μm), which is specifically: Real-time data acquisition, the sensor updates the environmental parameters synchronously every frame; Model iteration, optimize the compensation coefficients according to historical data to adapt to long-term temperature drift; Detector adjustment, dynamically adjust the integration time and gain based on environmental parameters to compensate for radiation attenuation and ensure that the detector works in the best linear region; Step S1.2: For motion artifact elimination, combined with the optical flow method and the improved YOLOv8 model, distinguish between real leakage diffusion and background motion noise, where: For the optical flow method, calculate the pixel motion vector of adjacent frames, filter out the continuous diffusion features (leakage gas motion direction consistent), and perform K-means clustering on the optical flow results to retain the areas with consistent diffusion direction, which is specifically implemented as: First, optical flow motion vector preprocessing: Input: Previous frame optical flow vector field , where N is the number of feature points; Vector amplitude: ; Vector direction: ; Filtering conditions: Amplitude threshold filtering (remove noise and stationary points): ; Direction consistency preliminary screening (retain those falling in the diffusion area): ; w is the local neighborhood window (recommended 5×5); Second, K-means clustering based on direction-position joint features: Feature vector construction: For each candidate vector vi, construct a 4-dimensional feature: ; (xi y i ) are feature point coordinates (normalized to [0, 1]); α are amplitude weight coefficients (recommended α =0.1); Clustering objective function: ; K Determined by the elbow rule (usually K =2~4); Force the direction difference to be greater than 45 degrees when initializing the clustering center (avoid local optimum).
[0016] Direction consistency verification: For each cluster C k , calculate the direction dispersion: .
[0017] Third, leakage area extraction and visualization: A. Spatial connectivity analysis: 1. For each retained cluster C k , extract its pixel coordinate set P k ; 2. Perform morphological closing operation (3x3 kernel) to fill the holes; 3. Remove areas with an area less than Amin (recommended Amin=0.1%×image area Amin=0.1%×image area); For diffusion direction field visualization: ; N(x,y) is the optical flow vector in the neighborhood of (x,y); Use HSV color coding for direction (hue) and amplitude (brightness).
[0018] For the improved YOLOv8 model, add a temporal attention module in the target detection layer to strengthen the temporal correlation of the leakage area and suppress transient interference (such as flying birds and vehicles); Step S1.3: For baseline dynamic correction, dynamically adjust the baseline radiation value based on environmental parameters to eliminate the long-term drift of background infrared radiation, the formula is: , is the initial baseline value, and are time decay coefficients, is the air pressure compensation factor, is the air pressure change amount, specifically: Sliding window mean, update baseline every 30 seconds, 5 minutes window to smooth environment fluctuation; Adaptive filter, predict baseline trend by Kalman filter, real-time correction.
[0019] More specifically, step S2 is implemented as: Step S2.1: For image spatio-temporal fusion, for the compensated calibrated data, to improve the spatio-temporal consistency of the leakage region, specifically: align the consecutive 5 frames of calibrated infrared images (obtained by a cooled infrared imaging module), weighted average to suppress random noise; and through SIFT feature point matching, superimpose the infrared leakage region to the visible light image to assist positioning; For weighted average to suppress random noise, specifically: First, input data preprocessing: Input: consecutive 5 frames of calibrated infrared images {It}t=15, each frame size M×N; Prior conditions: Sub-pixel level alignment has been completed by optical flow method (displacement error <0.3 pixels); Radiation calibration compensates for environmental temperature drift (accuracy ±0.1℃); Second, frame alignment based on motion compensation: 1. Reference frame selection, take the middle frame I3 as the reference (highest signal-to-noise ratio), calculate the dense optical flow field of the remaining frames to I3; 2. Sub-pixel level remapping: Perform bidirectional deformation on each frame It (avoiding holes); Use bicubic interpolation to maintain temperature gradient continuity.
[0020] Third, adaptive weighted fusion: Weight matrix design (considering signal-to-noise ratio and motion consistency): ; Signal-to-noise ratio weight (based on local standard deviation): ; Optical flow confidence weight (inverse projection error): ; Temporal decay weight (suppress motion blur): ; The fusion formula is: .
[0021] Fourth, noise suppression post-processing: Step 1: Median filter in time domain, take median of 5-frame sequence for each pixel position (x, y): Step 2: Spatial guided filter, take Ifused as guide image, preserve edges while smoothing homogenous regions.
[0022] For auxiliary positioning, the specific implementation is: First, SIFT feature extraction and matching; 1. Dual-modal feature extraction: Infrared image IIR: extract the outline of the leakage area as a mask Mleak, and only extract SIFT features in the outer region of the mask (to avoid interference from the leakage area); Visible light image IVIS: extract SIFT features throughout the image; 2. Cross-modal feature matching: Use the FLANN matcher (KD-Tree acceleration) and ratio test to eliminate false matches; Second, perspective transformation matrix estimation: 1. Robust optimization: fit the homography matrix through RANSAC; 2. Reprojection error verification, retain matching points that meet the requirements; Third, leakage area superposition and visualization: 1. Infrared mask transformation: transform the leakage mask Mleak to the visible light coordinate system through H; use bilinear interpolation to maintain edge smoothness; 2. Pseudo-color fusion, heat map rendering: map the infrared temperature data to the Jet color scale; transparency blending: superimpose only in the leakage area.
[0023] Step S2.2: For matched filter enhancement, enhance the signal-to-noise ratio of the gas feature absorption band for the compensated calibrated data, specifically: design a Gaussian-type matched filter to suppress the background radiation of other bands, thereby matching the band; and dynamically adjust the filter bandwidth according to the environmental parameters to adapt to the absorption peak shift under different temperature and humidity, specifically: Absorption peak shift model (temperature and humidity influence): ; T: temperature (℃), T0 reference temperature; H: humidity (%RH), H0 reference humidity; α, β, γ: material characteristic coefficients; Bandwidth adaptive formula: ; B0: nominal bandwidth under standard conditions; κT, κH: temperature and humidity sensitivity coefficients (determined through calibration experiments).
[0024] Step S2.3: For the leakage quantification analysis, the precise leakage evaluation is realized by combining the calibration results for the compensated calibrated data, specifically: Column concentration calculation: Based on the corrected radiation value, the gas concentration is inverted by Beer-Lambert law: , is the gas absorption coefficient, is the optical path length, is expressed as the reference radiation intensity (background radiation value when there is no leakage, obtained by baseline dynamic correction, unit: W / m²·sr·μm); Step S3.3: For the leakage quantification analysis, the motion vector of the leakage area is extracted by the optical flow method, and the diffusion speed and direction are calculated, specifically: First, the optical flow field motion vector extraction (based on the improved Horn-Schunck algorithm); Energy function minimization: ; I(x,y,t): pixel intensity of the time sequence infrared image sequence; u,v: x / y direction motion vector to be solved (unit: pixel / frame); Ix,Iy,It: image gradient in x / y / t direction (calculated by Sobel operator); λ: smoothing weight coefficient (typical value 0.1~1.0, suppress noise); Iterative solution formula (Gauss-Seidel iteration): ; is the mean value of the current pixel field; Termination condition: .
[0025] Second, leakage area segmentation and vector clustering: Dynamic threshold segmentation: ; V=(u,v) is the motion vector field; is the empirical coefficient; Then DBSCAN clustering (noise robustness).
[0026] Third, diffusion speed and direction calculation: Physical speed conversion: ; d: observation distance (m); f: camera focal length (pixels); Δt: frame interval time (s); Main diffusion direction (PCA extraction): ; covariance matrix: ; diffusion rate scalar: .
[0027] Fourth, the environmental coupling correction model: wind speed impact compensation: ; Vwind: ambient wind speed vector (need to be measured synchronously); d0: attenuation distance constant, related to gas density; turbulent diffusion coefficient: .
[0028] Further, for step S3.3, the alarm threshold is dynamically adjusted according to the leakage area, diffusion speed and column concentration (PPM·M).
[0029] Further, the detector is a medium wave refrigeration type infrared detector, which is used to obtain the infrared characteristic absorption image of gas leakage.
[0030] The method of the application is realized by the following hardware modules: 1. Refrigeration type infrared imaging module Core components: medium wave refrigeration type infrared detector (response wave band 3.2-3.5 μm), optical lens.
[0031] Function: capture infrared characteristic absorption image of gas leakage.
[0032] 2. Multi-environment sensor module Components: temperature and humidity sensor, barometer, GPS positioning module.
[0033] Function: real-time acquisition of environmental parameters and transmission to calibration module.
[0034] 3. Intelligent calibration module Algorithm layer: radiation compensation model, motion feature extraction model, adaptive learning unit.
[0035] Output: calibrated infrared image and leakage quantization data (column concentration, diffusion speed).
[0036] 4. Data processing and user interface Function: visualize the leakage area (superimpose visible light and infrared image), generate detection report, trigger hierarchical alarm.
[0037] It is worth mentioning that the technical features of the detector and other technical features involved in the present patent application should be regarded as prior art, and the specific structure, working principle and possible control mode and spatial arrangement mode involved in these technical features can be selected by using conventional selection in the art, and should not be regarded as the invention point of the present patent, and the present patent will not be further expanded and described in detail.
[0038] For those skilled in the art, the technical solutions recorded in the foregoing embodiments can still be modified, or some technical features therein can be replaced by equivalents, and any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.
Claims
1. An intelligent calibration method for a refrigerated infrared thermal imaging gas leak detection system, characterized in that: The following steps are involved: Step S1: Dynamic compensation of environmental parameters, including temperature drift compensation, motion artifact elimination and baseline dynamic correction processing; Step S2: Multimodal data fusion, performing image spatiotemporal fusion, matched filter enhancement, and leakage quantification analysis processing respectively.
2. The intelligent calibration method of a refrigerated infrared thermal imaging gas leak detection system according to claim 1, characterized in that: Step S1 is specifically implemented as follows: Step S1.1: For temperature drift compensation, the temperature and humidity sensors are used to collect environmental parameters in real time. Combined with the detector's operating temperature, a nonlinear regression model of temperature-radiation attenuation is established. The model dynamically corrects the infrared radiation value based on the following formula: , is the original radiation intensity, and are the temperature and humidity changes, and is the temperature and humidity compensation coefficient calibrated by experiment, is the radiation intensity of the leakage area after calibration, specifically: Real-time data acquisition, the sensor synchronously updates the environmental parameters every frame; Model iteration, optimizing compensation coefficients based on historical data to adapt to long-term temperature drift; Detector adjustment: dynamically adjust the integration time and gain based on environmental parameters to compensate for radiation attenuation and ensure that the detector operates in the optimal linear region; Step S1.2: For motion artifact removal, the optical flow method is combined with the improved YOLOv8 model to distinguish between real leakage diffusion and background motion noise, where: For the optical flow method, the motion vectors of pixels in adjacent frames are calculated, the continuous diffusion features are screened out, and K-means clustering is performed on the optical flow results to retain the areas with consistent diffusion directions; For the improved YOLOv8 model, a temporal attention module is added to the object detection layer to strengthen the temporal correlation of the leakage area and suppress transient interference; Step S1.3: For baseline dynamic correction, the baseline radiation value is dynamically adjusted based on environmental parameters to eliminate the long-term drift of background infrared radiation. The formula is: , is the initial baseline value, and is the time attenuation coefficient, is the pressure compensation factor, is the air pressure change, specifically: Sliding window mean, updating the baseline every 30 seconds, using a 5-minute window to smooth out environmental fluctuations; Adaptive filtering uses the Kalman filter to predict baseline change trends and make real-time corrections.
3. The intelligent calibration method of a refrigerated infrared thermal imaging gas leak detection system according to claim 2, characterized in that: Step S2 is specifically implemented as follows: Step S2.1: For image spatiotemporal fusion, the compensated and calibrated data is used to improve the spatiotemporal consistency of the leakage area. Specifically, five consecutive calibrated infrared images are aligned and weighted averaged to suppress random noise. Furthermore, SIFT feature point matching is used to overlay the infrared leakage area onto the visible light image to assist in positioning. Step S2.2: For matched filter enhancement, based on the compensated and calibrated data, the signal-to-noise ratio of the characteristic gas absorption band is enhanced. Specifically, a Gaussian matched filter is designed to suppress background radiation from other bands, thereby matching the bands. The filter bandwidth is dynamically adjusted according to environmental parameters to adapt to absorption peak shifts under different temperature and humidity conditions. Step S2.3: For leakage quantification analysis, the compensated and calibrated data is combined with the calibration results to achieve accurate leakage assessment, specifically: Column concentration calculation: Based on the corrected radiation value, the gas concentration is inverted using the Beer-Lambert law: , is the gas absorption coefficient, is the optical path length, Expressed as reference radiation intensity; Step S3.3: For leakage quantification analysis, the motion vector of the leakage area is extracted by the optical flow method, and the diffusion speed and direction are calculated.
4. The intelligent calibration method of a refrigerated infrared thermal imaging gas leak detection system according to claim 3, characterized in that: For step S3.3, the alarm threshold is dynamically adjusted according to the leakage area, diffusion rate and column concentration.
5. The intelligent calibration method of a refrigerated infrared thermal imaging gas leak detection system according to claim 4, characterized in that: The detector is a medium-wave cooled infrared detector used to obtain infrared characteristic absorption images of gas leaks.