An intelligent infrared gas imaging device and method based on active thermal management

Abstract

The application relates to an intelligent infrared gas imaging device and method based on active thermal management, and belongs to the technical field of gas imaging detection. The device comprises: N parallel infrared imaging channels provided with temperature control tuning infrared filters and non-refrigeration infrared detectors along an optical axis in sequence, a first thermal management component for uniformly temperature controlling the detectors to the same temperature, a second thermal management component for independently temperature controlling the filters, the central wavelengths of the N filters are tuned to the wavelengths corresponding to the absorption peaks of the measured gas and the wavelengths corresponding to the background bands without absorption of the measured gas, a collection and processing unit for synchronously acquiring the measurement images of the measured gas absorption and the background images without absorption output by the infrared imaging channels, and the unit is used for image recognition, region segmentation detection and pseudo-color superposition to realize real-time output of high-confidence gas distribution visualization results. The application improves the reliability, sensitivity and universality of gas imaging.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of gas optical imaging detection and infrared spectrum imaging technology, and in particular to an intelligent infrared gas imaging device and method based on active thermal management. BACKGROUND

[0002] Sulfur hexafluoride (SF6) gas is widely used in the electrical equipment of closed space in the power industry due to its excellent insulation and arc extinguishing performance. Real-time and accurate monitoring of the leakage of industrial gases such as SF6 is an important technical means to ensure the safe operation of equipment and meet environmental protection requirements.

[0003] Currently, gas monitoring technology has developed from traditional probe sensors to more advanced gas infrared imaging monitoring. In order to improve the specificity of monitoring and exclude background interference.

[0004] For existing solutions using non-cooled infrared detectors for gas monitoring, there are multiple inherent technical limitations, which are specifically manifested in:

[0005] Thermal drift of output signal reference. The photoelectric performance parameters of non-cooled infrared detectors (mainly microbolometer focal plane arrays) show high sensitivity to their own operating temperature. In a binocular differential system, even if two independent non-cooled cameras are of the same model and specifications, their respective temperature rise rates and response characteristic curves cannot be completely synchronized. This difference will cause the differential reference to dynamically drift over time, resulting in artifacts and background residuals in the differential image, and further causing false positives of the system.

[0006] High inherent noise level of the detector. The noise equivalent temperature difference (NETD) index of a non-cooled detector is directly related to its operating temperature. An increase in temperature will cause the inherent thermal noise level of the detector to rise, resulting in a deterioration in the signal-to-noise ratio. For weak gas leakage signals, the gray level change they produce on the image may be overwhelmed by the high background noise of the detector, resulting in a false negative, which fundamentally limits the minimum detection sensitivity of the system.

[0007] Limitations of compensation correction methods. To address the above problems, existing technologies usually use non-uniformity correction (NUC) or algorithm compensation. Non-uniformity correction requires the temporary interruption of image sequence output by a mechanical shutter, which cannot meet the requirements of continuous monitoring. A simplified linear compensation model, however, cannot accurately match the complex non-linear response drift of the detector, and the compensation effect is limited

[0008] Functional solidification and architectural defects. In addition to the above performance stability problems, the existing scheme has fundamental defects in function and architecture. First, the scheme relies on a narrow-band filter with a fixed center wavelength, resulting in a functional solidification of the entire device, becoming a "single-purpose device". If other types of gas need to be detected, the core filter element must be replaced, which severely limits the versatility and portability of the device. Second, the "single measurement channel vs single reference channel" binocular differential architecture itself also has performance bottlenecks. It identifies through single-point spectral comparison, and when there are interferents in the background, it is easy to produce false positives, and the image noise of the single reference channel limits the limit of the system detection sensitivity.

[0009] Therefore, developing a technical scheme that can not only provide a high-stability working temperature environment for the imaging system from the physical layer to suppress temperature drift and reduce noise, but also break through the technical barriers of fixed filtering to realize flexible versatility, and also surpass the limitations of the traditional binocular single-point comparison architecture to improve detection reliability and sensitivity, is an important technical subject currently faced by the field. SUMMARY

[0010] In view of the above analysis, the present application aims to disclose an intelligent infrared gas imaging device and method based on active thermal management; through the way of active thermal management, the problems of unstable imaging performance, insufficient reliability, poor versatility of being able to only fix to image one kind of gas, and limited flexibility caused by temperature fluctuations are solved.

[0011] The present application discloses an intelligent infrared gas imaging device based on active thermal management, comprising:

[0012] N parallel infrared imaging channels, N is an integer greater than or equal to 2; each channel is provided with a temperature-controlled tuning infrared filter and a non-cooled infrared detector along the optical axis; and

[0013] A first thermal management component for unified temperature control of each channel of the detector, synchronously stabilizing all the detectors at the same low temperature working point with the lowest noise;

[0014] A second thermal management component for independent temperature control of each channel of the filter, respectively tuning the center wavelengths of the N channels of the filter to the wavelengths corresponding to each absorption peak of the measured gas and the wavelengths corresponding to each background band without absorption of the measured gas;

[0015] A collection and processing unit for synchronously acquiring the measured gas absorption measurement images and non-absorption background images output by each channel of the infrared imaging channel, after image registration and background processing, performing image recognition, region segmentation detection and pseudo-color superposition, and real-time outputting high-confidence gas distribution visualization results.

[0016] Further, the first thermal management component synchronously stabilizes all the detectors at the same low-temperature working point with the lowest noise by uniformly controlling the semiconductor refrigeration elements TEC-D respectively thermally coupled to the detectors.

[0017] The second thermal management component independently controls the temperature according to the wavelength-temperature calibration model by the semiconductor refrigeration elements TEC-F respectively thermally coupled to the filters, so that the central wavelengths of M filters among the N filters are respectively located at the absorption peaks of the measured gas, and the central wavelengths of the remaining K filters are located at the background wave bands without absorption of the measured gas.

[0018] Further, the first thermal management component comprises:

[0019] N semiconductor refrigeration elements TEC-D, and a rear part of a camera core of each non-cooled infrared detector is tightly coupled to a cold end of each semiconductor refrigeration element TEC-D.

[0020] N temperature sensors, each of which is fixed at a key position of the camera core; the temperature at the key position can represent the real temperature of the detector chip and is easy to assemble / heat conduct.

[0021] A detector temperature control unit is configured to control the semiconductor refrigeration elements TEC-D connected to the camera core in a closed loop according to the temperature of the camera core collected by each temperature sensor, so as to forcibly and synchronously stabilize the camera core of all the detectors at the optimal low-temperature working point.

[0022] Further, the semiconductor refrigeration elements TEC-D in the first thermal management component are TEC1-12706, the hot end of each semiconductor refrigeration element TEC-D is connected to the heat plate of the central heat dissipation module through a graphene heat conduction pad, and the cold end is tightly coupled to the rear part of the camera core through a silver-containing heat-conducting silicone grease with a heat conduction coefficient > 7.5 W / mK.

[0023] The temperature sensors are MF52 series 10 kΩ NTC thermistors, which are respectively fixed at the key positions including the rear part of the camera core close to the coupling position of the cold end of the TEC-D.

[0024] The detector temperature control unit based on STM32 adopts a unified PID closed-loop control to forcibly and synchronously stabilize all the detectors at the optimal low-temperature working point of 15℃±0.1℃.

[0025] Further, the second thermal management component comprises N independent filter temperature control and tuning modules.

[0026] Each filter temperature control tuning module uses a temperature sensor integrated with the corresponding temperature control tuning infrared filter to collect filter temperature in real time; uses a thermally coupled semiconductor refrigeration element TEC-F to independently adjust the temperature of each filter; uses a filter tuning unit with an embedded wavelength-temperature calibration model to receive the filter temperature collected by the temperature sensor in real time, drive each TEC-F, and accurately tune the center wavelength of the corresponding filter to the set value;

[0027] The set value includes: M characteristic absorption peak wavelengths of the measured gas, and K background wavelength bands without absorption of the measured gas.

[0028] Further, each filter temperature control tuning module in the second thermal management assembly comprises:

[0029] A red copper precisely processed heat sink and a mechanical base, with the hot end connected to the central heat dissipation module;

[0030] A ring-shaped precisely tuned TEC, with the hot surface fixed to the base and the cold surface bearing the tunable infrared filter;

[0031] The tunable infrared filter, taking vanadium dioxide (VO2) as the core tuning material, is mechanically pre-tightened and fixed on the TEC cold surface through a red copper threaded compression ring;

[0032] A Pt1000 thin film platinum resistance temperature sensor, fixedly arranged on a temperature measurement base of the red copper threaded compression ring, and integrated with the filter;

[0033] A PEEK thermal insulation ring, thermally isolating the module from the aluminum alloy optical bearing plate.

[0034] Further, the acquisition and processing unit comprises:

[0035] A synchronization and acquisition module, configured to synchronously trigger N infrared detectors to collect image frames with a unified clock signal, and receive image data of N channels; the image data of N channels includes M measurement images and K background image data;

[0036] An image processing module, configured to register N images; and obtain a reference image by weighted averaging the registered background image; perform difference between the reference image and each measurement image to obtain M high signal-to-noise difference images; perform deep feature extraction and cross-validation fusion decision on the M difference images by using a neural network, and output a high-confidence gas region segmentation mask;

[0037] A result fusion display module, configured to fuse the gas region segmentation mask with an original image, and visualize and mark in real time in a pseudo-color superimposition manner.

[0038] ​Further, the image processing module comprises: a registration module, a reference image generation module, a difference module, a neural network module and a decision module; wherein,

[0039] A multi-channel registration module is configured to use the homography matrix calibrated offline to perform sub-pixel level registration on the remaining N-1 images based on one measurement image as a reference;

[0040] A background fusion and noise reduction module is configured to perform pixel-level weighted averaging on the K registered background images to generate a "super-clean" background reference image;

[0041] A difference operation module is configured to perform difference operation on the M measurement images and the background reference image respectively to obtain M high signal-to-noise ratio difference images;

[0042] A multi-dimensional feature recognition and decision module is configured to use a pre-trained U-Net CNN network to input the M difference images, recognize and segment each input difference image to output a corresponding probability map, and then perform thresholding on each probability map to obtain a binary mask corresponding to each probability map, and then perform pixel-level "logical AND" operation on the binary masks corresponding to each probability map, when the pixel area of all masks that are "true" reaches a set area, the gas is detected, and the result after the logical AND operation is output as the final segmentation mask.

[0043] Further, the measured gas is SF6, and the intelligent infrared gas imaging device is used for optical detection of SF6 gas leakage; and is configured as 4 parallel infrared imaging channels in the form of a 2x2 matrix, wherein,

[0044] Two of the four channels are configured as measurement channels by temperature control of the second thermal management assembly, and the center wavelengths of the filters are tuned to the main absorption peak 10.55 µm and the secondary absorption peak 10.8 µm of SF6 respectively; and the remaining two channels are configured as background channels, and the center wavelengths of the filters are tuned to the background reference wavelength 11.0 µm and the background reference wavelength 11.2 µm respectively.

[0045] The application also discloses a gas imaging method based on the intelligent infrared gas imaging device based on active thermal management as described above, comprising:

[0046] S1, the detector is cooled to a low temperature and stabilized; the first thermal management assembly is started to synchronously cool the infrared detectors of all N infrared imaging channels to a preset optimal low temperature working point, and waits for stabilization;

[0047] S2, the spectrum channel is set; the second thermal management assembly receives a wavelength setting instruction of each infrared imaging channel, independently tunes the center wavelengths of the N filters to the wavelengths corresponding to the absorption peaks of the measured gas and the wavelengths corresponding to the background wavelengths without absorption of the measured gas respectively, and waits for stabilization.

[0048] S3, synchronous parallel acquisition; all N channels are triggered synchronously to acquire the measurement image of the measured gas absorption and the background image without absorption in parallel to obtain the measurement image of the measured gas absorption and the background image without absorption;

[0049] S4, image fusion processing and identification; the acquisition processing unit performs image registration and background processing on the measured gas absorption measurement image and the background image without absorption, and then performs image recognition and region segmentation detection to obtain the outline and position of the measured gas;

[0050] S5, result visualization; the finally identified outline and position information of the measured gas is superimposed on the measurement image corresponding to the main absorption peak in real time, so as to realize intuitive and visual monitoring of the leakage position, range and diffusion trend.

[0051] The present application can realize one of the following beneficial effects:

[0052] The intelligent infrared gas imaging device and method based on active thermal management disclosed by the present application can build a reconfigurable spectral detection matrix, and provide independent and optimized working environment for the core elements in the matrix, so as to fundamentally improve the reliability, sensitivity and universality of gas imaging, and be suitable for gas leakage detection. BRIEF DESCRIPTION OF DRAWINGS

[0053] The accompanying drawings are only used for the purpose of illustrating specific embodiments and are not considered as limiting the present application, and the same reference signs represent the same parts throughout the drawings;

[0054] Figure 1 The intelligent infrared gas imaging device based on active thermal management in the embodiment of the present application is connected to the schematic block diagram;

[0055] Figure 2 The integrated three-dimensional exploded structure schematic diagram of the filter temperature control tuning module in the embodiment of the present application;

[0056] Figure 3 The gas imaging method flow chart of the intelligent infrared gas imaging device based on active thermal management in the embodiment of the present application;

[0057] Figure 4 The intelligent infrared gas imaging device for SF6 gas leakage optical detection in the embodiment of the present application is connected to the schematic block diagram;

[0058] Figure 5 The image fusion processing and identification process schematic diagram in the process of SF6 gas leakage optical detection in the embodiment of the present application;

[0059] Figure 6A comparison chart of gas detection effects of the method of the present application and a conventional threshold segmentation method at different background temperatures in an embodiment of the present application is shown in the figure;

[0060] Figure 7 An experimental result graph of improvement effects of the first thermal management component on camera temperature response and linearity in an embodiment of the present application is shown in the figure;

[0061] Figure 8 An experimental result graph of improvement effects of the first thermal management component on camera NETD, response rate and time noise in an embodiment of the present application is shown in the figure;

[0062] Figure 9 An experimental result graph of improvement effects of the first thermal management component on camera start-up thermal drift and output signal fluctuation after stabilization in an embodiment of the present application is shown in the figure.

[0063] Reference signs: 21-PEEK thermal insulation ring; 22-threaded pressure ring; 23-Pt1000 thin film platinum resistance temperature sensor; 24-tunable infrared filter; 25-ring-shaped precision tuning TEC; 26-heat sink and mechanical base;

[0064] 41-infrared imaging channel, 42-infrared detector, 43-filter temperature control tuning module, 44-detector TEC refrigeration sheet, 45-NTC thermistor; 46-detector temperature control unit; 47-multi-channel high-precision filter tuning board card; 48-host. DETAILED DESCRIPTION

[0065] The preferred embodiments of the present application will be specifically described below in conjunction with the accompanying drawings, wherein the drawings form a part of the present application and are used to explain the principles of the embodiments of the present application.

[0066] Embodiment one

[0067] One embodiment of the present application discloses an intelligent infrared gas imaging device based on active thermal management, as shown in the figure, comprising: Figure 1 N-channel parallel infrared imaging channels, N is an integer greater than or equal to 2; each channel is sequentially provided with a temperature control tuning infrared filter and a non-cooled infrared detector along an optical axis; and,

[0068] A first thermal management component is used for unified temperature control of each channel of the detector, and all the detectors are synchronously stabilized at the same low temperature working point of the lowest noise;

[0069] A second thermal management component is used for independent temperature control of each channel of the filter, and the center wavelengths of the N-channel filters are respectively tuned to the wavelengths corresponding to the absorption peaks of the measured gas and the wavelengths corresponding to the background wave bands without absorption of the measured gas;

[0070] A second thermal management component is used for independent temperature control of each channel of the filter, and the center wavelengths of the N-channel filters are respectively tuned to the wavelengths corresponding to the absorption peaks of the measured gas and the wavelengths corresponding to the background wave bands without absorption of the measured gas;

[0071] The acquisition processing unit is configured to synchronously acquire a measured image of gas absorption and a background image without absorption output by each infrared imaging channel, and to perform image recognition, region segmentation detection and pseudo-color superposition after image registration and background processing, thereby outputting a high-confidence gas distribution visualization result in real time.

[0072] The N-channel infrared imaging channels are arranged in a manner that the infrared optical axes are as concentrated as possible (for example, the 4-channel infrared imaging channels are arranged in a 2x2 matrix form), and are fixed on an integral 7075 aluminum alloy optical bearing plate that has been subjected to anode hard oxidation treatment to enhance surface hardness and heat dissipation capacity through high-precision positioning pins. This design ensures that all channel optical axes are as concentrated as possible, and can be long-term parallel and stable in a vibration and temperature change environment.

[0073] The active thermal management in the embodiment is a separate thermal management architecture, which further includes a central heat dissipation module for unified heat dissipation. The module is composed of a large-area copper fin array, multiple heat pipes and a variable-speed temperature control fan, and can efficiently handle the total heat load generated by all TEC elements.

[0074] Specifically, the uncooled infrared detector in each infrared imaging channel is mainly a microbolometer focal plane array (FPA) detector. In order to make the key performance parameters of each detector, including resolution, pixel size, response rate and noise level, highly consistent, and thus to maximize the elimination of fixed pattern noise and drift introduced by the differences between the detectors themselves, the same model and batch of detectors are selected for each detector.

[0075] The noise equivalent temperature difference (NETD) index of the uncooled detector is directly related to its operating temperature. An increase in temperature will cause the intrinsic thermal noise level of the detector to rise, resulting in a deterioration of the signal-to-noise ratio. To address the impact of temperature changes on the performance of the uncooled infrared detector, the uncooled infrared detector in the embodiment is actively thermally managed.

[0076] Preferably, in the embodiment, the first thermal management assembly synchronously stabilizes all detectors at the same low-temperature operating point with the lowest noise by uniformly controlling the temperature of the semiconductor refrigeration elements TEC-D that are respectively thermally coupled to each detector. The semiconductor refrigeration element TEC is a solid-state heat pump without mechanical movement, small in size, fast in response, and capable of precise temperature control ±0.01°C. It does not require refrigerant and has no vibration noise, making it very suitable for temperature control applications in the scenario of the embodiment.

[0077] Further, the first thermal management assembly includes:

[0078] N semiconductor refrigeration elements TEC-D, the cold end of each semiconductor refrigeration element TEC-D is tightly coupled with the rear part of the camera core of one non-refrigeration infrared detector;

[0079] N temperature sensors, each temperature sensor is respectively fixed on the key position of the camera core; the temperature of the key position can represent the real temperature of the detector chip and is easy to assemble / heat conduct;

[0080] A detector temperature control unit is used to control the semiconductor refrigeration element TEC-D connected with the camera core according to the temperature of the camera core collected by each temperature sensor, so as to control the camera core of all detectors to be forced and synchronized to be stabilized at the optimal low-temperature working point.

[0081] In a specific scheme, the model of the semiconductor refrigeration element TEC-D in the first thermal management assembly is TEC1-12706, the hot end of each semiconductor refrigeration element TEC-D is connected with the heat plate of the central heat dissipation module through the graphene heat conduction pad, and the cold end is tightly coupled with the rear part of the camera core through the silver-containing heat-conducting silicone grease with a heat conduction coefficient of >7.5 W / mK;

[0082] The temperature sensor is a MF52 series 10 kΩ NTC thermistor, which is respectively fixed on the key position including the rear part of the camera core close to the coupling position of the cold end of the TEC-D;

[0083] The detector temperature control unit based on STM32 adopts a unified PID closed-loop control to force and synchronize all detectors to be stabilized at the optimal low-temperature working point of 15℃±0.1℃.

[0084] Specifically, the temperature control and tuning infrared filter in each infrared imaging channel is a temperature control and tuning infrared filter taking vanadium dioxide (VO2) ) and other temperature-sensitive materials as the core, and the center wavelength of the filter can be located at the set wavelength by accurately controlling the temperature of the filter. In order to realize the selection and control of the center wavelength of each filter, the filter is actively thermally managed in the embodiment.

[0085] Preferably, in the embodiment, the second thermal management assembly independently controls the temperature of each filter through the semiconductor refrigeration element TEC-F which is respectively thermally coupled with each filter, according to the wavelength-temperature calibration model, tunes the center wavelengths of M filters in N filters to be located at each absorption peak of the measured gas, and tunes the center wavelengths of the remaining K filters to be located at each background waveband without absorption of the measured gas.

[0086] Further, the second thermal management assembly for actively thermally managing the filter includes N independent filter temperature control and tuning modules.

[0087] Each filter temperature control tuning module adopts a temperature sensor integrated with the corresponding temperature control tuning infrared filter to collect the filter temperature in real time; adopts a semiconductor refrigeration element TEC-F thermally coupled with each temperature control tuning infrared filter to independently adjust the temperature of each filter; adopts a filter tuning unit with a built-in wavelength-temperature calibration model to receive the filter temperature collected by the temperature sensor in real time, and drive each TEC-F to accurately tune the center wavelength of the corresponding filter to the set value;

[0088] The set value includes M characteristic absorption peak wavelengths of the measured gas and K background wavelength bands of the measured gas without absorption; wherein the M characteristic absorption peaks include the main absorption peak and the secondary absorption peak of the measured gas, which can be selected according to specific test requirements; the K background wavelength bands can also be selected according to specific test requirements.

[0089] In one specific scheme, each filter temperature control tuning module in the second thermal management assembly, as shown in Figure 2 , includes:

[0090] A red copper precisely processed heat sink and a mechanical base 26, the hot end of which is connected to a central heat dissipation module;

[0091] A ring-shaped precisely tuned TEC 25, the hot face of which is fixed to the base, and the cold face of which carries a tunable infrared filter 24;

[0092] The tunable infrared filter 24 takes vanadium dioxide (VO2) as the core tuning material, and is mechanically pre-tightened and fixed to the TEC cold face through a red copper threaded compression ring 22;

[0093] A Pt1000 thin film platinum resistance temperature sensor 23, which is fixed to a temperature measurement base of the red copper threaded compression ring, and is integrated with the filter;

[0094] A PEEK thermal insulation ring 21, which thermally insulates the module from an aluminum alloy optical carrying plate;

[0095] The filter tuning unit adopts a FPGA-based multi-channel high-precision filter tuning board card to perform high-precision acquisition and digital filtering on the Pt1000 feedback signal corresponding to each filter, independently generates a high-frequency PWM drive signal corresponding to each temperature control target to control the temperature of each ring-shaped precisely tuned TEC, and realizes rapid and accurate tuning of the center wavelength of the filter.

[0096] Specifically, the acquisition and processing unit includes:

[0097] ​a synchronization and acquisition module, configured to synchronously trigger N-channel infrared detectors to acquire image frames by using a unified clock signal, and configured to receive image data of the N channels, wherein the image data of the N channels includes M-channel measurement images and K-channel background image data;

[0098] an image processing module, configured to register the N-channel images, and configured to obtain a reference image by weighted averaging the registered background images, and configured to obtain M high signal-to-noise difference images by differentiating the reference image and each measurement image;

[0099] a result fusion and display module, configured to fuse the gas region segmentation mask and the original image, and configured to real-time visualize the leakage location, range and diffusion trend in a pseudo-color superimposed manner.

[0100] More specifically, the image processing module includes a registration module, a reference image generation module, a difference module, a neural network module and a decision module, wherein,

[0101] a multi-channel registration module, configured to use a single-shot matrix obtained by offline calibration to perform sub-pixel level registration on the remaining N-1 images based on one measurement image as a reference;

[0102] a background fusion and denoising module, configured to perform pixel-level weighted averaging on the K-channel registered background images to generate a "super-clean" background reference image;

[0103] a difference operation module, configured to differentiate the M-channel measurement images and the background reference image respectively to obtain M high signal-to-noise ratio difference images;

[0104] a multi-dimensional feature recognition and decision module, configured to use a pre-trained U-Net CNN network to recognize and segment each difference image to output a corresponding probability map, and configured to threshold each probability map to obtain a binary mask corresponding to each probability map, and then perform pixel-level "logical AND" operation on the binary masks corresponding to each probability map, and when the pixel region that is "true" in all masks reaches a set area, the gas is detected, and the result after the logical AND operation is output as the final segmentation mask.

[0105] More specifically, the result fusion and display module is configured to real-time superimpose the extracted measured gas region on the measurement image corresponding to the main absorption peak in a high-contrast pseudo-color form, to realize intuitive and visual monitoring of the leakage location, range and diffusion trend.

[0106] Specifically, the processing process includes:

[0107] 1) After obtaining the mask image of the measured gas, the gray value of the corresponding gas region in the image is mapped to color information with visual discrimination through a preset color look-up table (CLUT);

[0108] For example, a "Jet" color spectrum from blue to red can be used, in which blue represents low concentration and red represents high concentration.

[0109] 2) Using the mask image as an overlay mask, the color regions are overlaid on the real-time measurement image corresponding to the main absorption peak to form the final monitoring result image.

[0110] 3) The image is updated in real time on the display, and the operator can visually observe the position, shape, diffusion trend and relative concentration distribution of the measured gas.

[0111] Example Two

[0112] A gas imaging method based on the intelligent infrared gas imaging device based on active thermal management as described in the above example is given in this embodiment, as shown in Figure 3 , comprising:

[0113] S1, the detector is cooled to a low temperature; the first thermal management component is started to synchronously cool the infrared detectors of all N infrared imaging channels, and waits for them to be stably maintained at a preset optimal low temperature working point;

[0114] Specifically, by using the feedback of the temperature sensor, the semiconductor cooler TEC-D in each channel is driven to work, so that the detector is stably maintained at a preset optimal low temperature working point;

[0115] S2, spectrum channel setting; the second thermal management component receives the wavelength setting instruction of each infrared imaging channel, independently tunes the center wavelengths of the N filters to the wavelengths corresponding to the absorption peaks of the measured gas and the wavelengths corresponding to the background wavebands without absorption of the measured gas, and waits for stabilization;

[0116] Specifically, according to the built-in wavelength-temperature calibration model, the high-precision temperature sensor is used to feedback, and the semiconductor cooler TEC-F in each channel is independently driven to accurately tune the N filters to their target wavelengths; the target wavelengths of each channel are M measurement wavelengths and K background wavelengths, respectively;

[0117] S3, synchronous and parallel acquisition; all N channels are synchronously triggered to perform parallel image sequence acquisition to obtain the measurement image of the measured gas absorption and the background image without absorption;

[0118] S4, image fusion processing and recognition; the image acquisition and processing unit processes the measured image of the measured gas absorption and the background image without absorption, performs image registration and background processing, and then performs image recognition, region segmentation and detection to obtain the outline and position of the measured gas;

[0119] S5, result visualization; the final recognized outline and position information of the measured gas is superimposed on the measured image corresponding to the main absorption peak in real time, and intuitive and visual monitoring of the leakage position, range and diffusion trend is realized.

[0120] Specifically, in S4, the following steps are included:

[0121] a) pixel-level registration of N images;

[0122] b) average the images of K background channels to generate a "super-clean" background reference image;

[0123] c) difference between the images of M measurement channels and the "super-clean" background reference image;

[0124] d) input the M difference images into the CNN model for recognition, and output the outline and position of the measured gas through cross-validation and fusion decision.

[0125] Embodiment three

[0126] In this embodiment, a scheme for SF6 gas leakage optical detection using an intelligent infrared gas imaging device based on active thermal management is also given, as shown in Figure 4 A four-channel parallel infrared imaging array is used, which is fixed on a whole 7075 aluminum alloy optical bearing plate through high-precision positioning pins in the form of 2x2 by four structural identical infrared imaging channels 41. The surface hardness and heat dissipation capacity of the plate are enhanced by anode hard anodizing treatment. This design ensures the long-term parallelism and stability of all channel optical axes in the vibration and temperature change environment.

[0127] In this embodiment, a separate thermal management architecture is used, and the heat is finally uniformly dissipated by a central heat dissipation module composed of a large-area copper fin array, multiple heat pipes and a variable speed temperature control fan, which can efficiently handle the total heat load generated by all TEC elements.

[0128] The first thermal management component comprises four main cooling TECs (model TEC1-12706). Their hot ends are connected to the heat spreader of the central heat dissipation module via graphene thermal pads, while their cold ends are tightly coupled to the rear of each camera module via silver-containing thermal grease (thermal conductivity >7.5W / mK). Four MF52 series 10K NTC thermistors 45 serve as feedback sensors, controlled by a unified PID closed-loop control unit 46 based on an STM32 microprocessor. This unit incorporates over-temperature protection and PID parameter self-tuning functions, ensuring that all four detectors are forced and synchronously stabilized at their optimal low-temperature operating point of 15℃±0.1℃ under different ambient temperatures.

[0129] The structure of the second thermal management component can be referenced. Figure 2 As shown, each channel is equipped with an independent integrated filter temperature control tuning module 43. The heat sink and mechanical base 26 are precision-machined from copper, and their hot ends are also connected to the central heat dissipation module. A customized ring-shaped precision-tuned TEC has its hot side fixed on the base. A Pt1000 thin-film platinum resistance temperature sensor 23 is pre-fixed on a specially machined temperature sensing base on a copper threaded pressure ring 22. Vanadium dioxide (V₂O₃) is used as the base. A tunable infrared filter 24, with Pt1000 as the core tuning material, is placed on the cold surface of the annular TEC 25 and mechanically pre-tightened by a copper threaded clamping ring 22. The entire module is connected to the aluminum alloy optical carrier plate via a PEEK thermal insulation ring 21. An FPGA-based multi-channel high-precision filter tuning board is responsible for independent and parallel temperature control tuning. It can generate high-frequency (e.g., 20kHz) PWM drive signals and perform high-precision acquisition and digital filtering of the weak resistance signals fed back by the Pt1000 sensor.

[0130] Two of the four channels are used as measurement channels by temperature control of the second thermal management component, with the center wavelength of the filter tuned to the main absorption peak of SF6 at 10.55µm and the secondary absorption peak at 10.8µm, respectively; the remaining two channels are used as background channels, with the center wavelength of the filter tuned to the background reference bands of 11.0µm and 11.2µm, respectively.

[0131] More information of gas absorption can be obtained by selecting the primary and secondary absorption peaks of SF6 gas, and the selected two background reference wavebands are located in the near-infrared waveband adjacent to the SF6 gas without absorption effect, which can truly reflect the background thermal radiation information of the scene. According to the Planck blackbody radiation law, at a typical environmental temperature, the radiation intensity and dynamic response of the background thermal radiation with temperature change in the selected measurement waveband and reference waveband are highly consistent. This high correlation makes the background radiation form a 'common mode signal' in the dual-channel which is easy to eliminate by difference operation. Compared with selecting reference wavebands with a long spectral distance, the scheme of the embodiment of the application can more thoroughly eliminate the interference of complex thermal background, significantly reduce the dependence on subsequent image processing algorithms, and fundamentally improve the detection signal-to-noise ratio and reliability of the system for trace SF6 gas leakage in a real industrial environment.

[0132] Moreover, the FPGA-based multi-path high-precision filter tuning board card 47 is adopted in the second thermal management component, and according to the built-in and precisely calibrated wavelength-temperature calibration lookup table (LUT), the target wavelength is converted into a high-precision target temperature setting value. Then, the board card reads the resistance values of the four Pt1000 sensors at a higher sampling rate of 100 Hz, drives the four annular TEC tuning films through an independent PID control loop, and accurately stabilizes the four filters at the respective target temperatures within a few seconds, with a stabilization accuracy better than ±0.05℃.

[0133] The detectors in the 4-path parallel infrared imaging channel are forced and synchronized to be stabilized at an optimal low-temperature working point of 15℃±0.1℃ by the temperature control of the first thermal management component.

[0134] The acquisition and processing unit adopts a central control and image processing host (configured as an industrial PC with a four-channel CameraLink image acquisition card and an NVIDIA RTX series GPU) for unified scheduling.

[0135] In the process of optical detection of SF6 gas leakage, the following steps are included:

[0136] Step 1, detector low-temperature stabilization: the first thermal management component is started, and the STM32 controller thereof monitors the four NTC resistance values at a sampling rate of 10 Hz, drives the four detector TEC refrigeration films 44 through PWM signals, and stabilizes all four infrared detectors 42 at 15℃ within about 3 minutes after the system is started.

[0137] Step 2, spectral channel setting: the host 48 sends wavelength setting instructions of the four channels to the multi-path high-precision filter tuning board card 47 of the second thermal management component through a serial port:

[0138] Channel one, as measurement channel A, tunes the filter to the primary absorption peak 10.55µm of SF6;

[0139] Channel two, as measurement channel B, the filter is tuned to the secondary absorption peak of SF6 10.8µm;

[0140] Channel three, as background channel X: the filter is tuned to the background reference waveband 11.0µm;

[0141] Channel four, as background channel Y: the filter is tuned to the background reference waveband 11.2µm.

[0142] Step 3, synchronous parallel acquisition: after the temperature of all detectors and filters is stable, the host computer 48 triggers all four channels to perform 14-bit RAW format image sequence acquisition at a frame rate of 30Hz, and obtains measurement image A, measurement image B, background image X and background image Y, respectively.

[0143] Step 4, image fusion processing and recognition: the software application (developed based on C++ and CUDA) on the host computer 48 processes each frame of image array in real time; as shown in the figure, the processing process includes: Figure 5

[0144] Multi-channel image registration: taking the image of channel one as the reference main image, the remaining three images are subjected to perspective transformation through the 3x3 homography transformation matrix calibrated offline, to achieve sub-pixel level accurate alignment.

[0145] Background fusion noise reduction: the registered background image X and background image Y are subjected to pixel-level weighted averaging (the weight can be adaptively adjusted according to the signal-to-noise ratio of the corresponding waveband, and the default is 0.5), to generate a “super-clean” background reference image (Ref_clean) with significant noise suppression.

[0146] Difference enhancement: subtracting the “super-clean” background reference image Ref_clean from the measurement image A and the measurement image B, respectively, generates two difference images (Diff_A, Diff_B) with high signal-to-noise ratio.

[0147] Multi-dimensional feature recognition and decision: taking the difference images Diff_A and Diff_B as a double-channel input, and sending them into a pre-trained convolutional neural network (CNN) model based on U-Net architecture for recognition and segmentation. The model outputs two corresponding probability maps (Prob_A, Prob_B). The decision unit first thresholdizes (such as 0.5) the two probability maps to obtain two binary masks, and then performs pixel-level “logical and” operation on the two masks. Only when the pixel area where both masks are “true” reaches a certain area, the system finally determines that there is SF6 gas leakage, and outputs the result of the logical and operation as the final segmentation mask. ​

[0148] Step 5, Result Visualization: The machine segments the gas leak area into a mask, renders it as a semi-transparent red pseudo-color layer, overlays it onto the original video stream of measurement channel A in real time, and outputs it to the display device through the HDMI interface to form the final visualized monitoring result.

[0149] This invention achieves significantly superior overall performance compared to existing technologies by organically combining separate active thermal management of the detector and filter, multi-channel parallel detection, and multi-dimensional information fusion decision-making. (Refer to...) Figure 6 to Figure 9 The experimental comparison results fully demonstrate the beneficial effects of the present invention, which are specifically reflected in multiple aspects:

[0150] First, the physical basis of this invention stems from the fundamental performance improvement of the first thermal management component on all imaging channel detectors, providing a crucial physical prerequisite for all high-precision detection. (Refer to...) Figure 7 After implementing active temperature control, the camera response consistency was significantly improved, and the linear fitting determination coefficient R² increased from 0.9829 to 0.9998. (Reference) Figure 8 Experimental results of the first thermal management component's improvement on camera NETD, response rate, and temporal noise; Figure 8 The comparison chart of noise equivalent temperature difference (NETD) shows that the noise equivalent temperature difference (NETD) of the camera has been optimized from 14.4mK without temperature control to 11.6mK, with a performance improvement of about 19.4% and a significant optimization of the core signal-to-noise ratio; Figure 8 The temperature comparison graph shows that the response was optimized from 62.2℃ (uncontrolled) to 69.3℃. Figure 8 The comparison chart of time noise shows that the time noise was optimized from 0.89±0.55 for uncontrolled temperature to 0.81±0.28. Figure 8 The comparison graph of the temperature response curves of the average gray value shows that the average gray value is high and unstable when the temperature is not controlled; the average gray value is low and stable after temperature control, showing a positive proportional relationship with the increase of temperature.

[0151] More importantly, refer to Figure 9 The experimental results of the first thermal management component on improving camera power-on thermal drift and output signal fluctuation after stabilization are shown in the complete temperature response diagram and the stabilization stage analysis diagram. It can be seen that after active temperature control, the curve quickly stabilizes at the target operating point and converges rapidly and remains near a constant value, basically eliminating thermal drift. From the stability comparison diagram and the relative stability comparison diagram, it can be seen that after active temperature control, the histogram distribution is significantly narrowed, the signal is more concentrated, and the long-term operating reference height of the system is highly stable. The standard deviation of the signal output after stabilization is reduced from 54.4DN before temperature control to 30.2DN, and the stability is improved by 44.5%. This fundamentally solves the reference drift problem of long-term operation of multi-channel differential systems.

[0152] On the basis of this high stability and low noise, the neural network algorithm of the application can maximize its powerful feature recognition capability. Figure 6 As shown in the figure, Figure 6 a and c in the figure are real images of SF6 gas leakage under different temperature conditions, Figure 6 b in the figure is a segmented image of the SF6 gas region under each temperature condition extracted by the traditional threshold method; Figure 6 d in the figure is a segmented image of the SF6 gas region under each temperature condition extracted by the scheme of the embodiment of the application;

[0153] Furthermore, the biggest innovation of the application is to completely break through the architecture bottleneck of the traditional binocular differential scheme, and through multi-channel parallel detection and information fusion, a qualitative leap is realized in the two dimensions of reliability and sensitivity. The four-channel (double measurement + double background) architecture in the embodiment of the application brings double advantages: first, by cross-verification of the signals of two different characteristic absorption peaks (10.55µm and 10.8µm) of SF6 gas, the matching recognition of the "spectrum fingerprint" of the gas is realized, which can fundamentally eliminate the false positives caused by interference substances with single-band absorption characteristics, greatly improving the reliability and accuracy of detection. Second, by averaging the pixels of two independent background channels (11µm and 11.2µm), a "super-clean" background reference image with deep suppression of random noise is generated, which greatly improves the signal-to-noise ratio of the final difference image compared with a single background channel, thereby improving the minimum detection sensitivity of the system to a new height.

[0154] In summary, through the synergistic innovation of "detector low-temperature stability", "multi-channel spectrum cross-verification and background noise reduction", and "neural network intelligent recognition", the application finally provides a next-generation gas optical imaging solution that far surpasses existing technologies in reliability, sensitivity, and versatility.

[0155] The above is only the preferred specific embodiment of the application, but the protection scope of the application is not limited thereto, and any changes or replacements within the technical scope disclosed by the application can be easily thought of by those skilled in the art, which should be covered within the protection scope of the application.

Claims

1. An active thermal management based intelligent infrared gas imaging device, characterized in that, The application relates to an active thermal management intelligent infrared gas imaging device. The device comprises: N parallel infrared imaging channels, N being an integer greater than or equal to 2; Each channel is provided with a temperature-controlled tuning infrared filter and a non-cooled infrared detector along an optical axis; A first thermal management component is used for unified temperature control of the detectors in each channel, and all the detectors are synchronously stabilized at the same low-temperature working point of the lowest noise; A second thermal management component is used for independent temperature control of the filters in each channel, and the center wavelengths of the N filters are tuned to the wavelengths corresponding to the absorption peaks of the measured gas and the wavelengths corresponding to the background wave bands without absorption of the measured gas, respectively; A collection and processing unit is used for synchronously collecting the measurement images and the background images of the measured gas output by the infrared imaging channels, performing image registration, background processing, image recognition, region segmentation detection and pseudo-color superposition, and outputting a high-confidence gas distribution visualization result in real time. The collection and processing unit comprises: A synchronization and collection module is used for synchronously triggering the N infrared detectors to collect image frames by using a unified clock signal, and receiving the image data of the N channels; the image data of the N channels comprises M measurement images and K background image data; An image processing module is used for registering the N images, performing weighted average on the registered background images to obtain a reference image, performing difference between the reference image and each measurement image to obtain M high-signal-to-noise difference images, and outputting a high-confidence gas region segmentation mask by using a neural network to perform deep feature extraction and cross-validation fusion decision on the M difference images; A result fusion display module is used for fusing the gas region segmentation mask with the original image, and visualizing and marking in a pseudo-color superposition mode in real time.

2. The active thermal management intelligent infrared gas imaging device according to claim 1, wherein The first thermal management component is used for synchronously stabilizing all the detectors at the same low-temperature working point of the lowest noise by performing unified temperature control on the semiconductor refrigeration elements TEC-D which are respectively thermally coupled with the detectors in each channel; The second thermal management component is used for tuning the center wavelengths of M channels of the N filters to be located at the absorption peaks of the measured gas and tuning the center wavelengths of the remaining K channels to be located at the background wave bands without absorption of the measured gas by performing independent temperature control on the semiconductor refrigeration elements TEC-F which are respectively thermally coupled with the filters in each channel according to a wavelength-temperature calibration model.

3. The active thermal management intelligent infrared gas imaging device according to claim 2, wherein The first thermal management component comprises: N semiconductor refrigeration elements TEC-D, and the cold end of each semiconductor refrigeration element TEC-D is tightly coupled with the rear part of the camera core of a non-cooled infrared detector; N temperature sensors, each of which is fixed at a key position of the camera core; the temperature of the key position can represent the real temperature of the detector chip and is easy to assemble and conduct heat; A detector temperature control unit is used for performing closed-loop temperature control on the semiconductor refrigeration elements TEC-D which are connected with the camera core according to the temperature of the camera core collected by each temperature sensor, and forcibly and synchronously stabilizing the camera core of all the detectors at the optimal low-temperature working point.

4. The actively thermal managed smart infrared gas imaging device according to claim 3, wherein, the semiconductor refrigeration element TEC-D in the first thermal management component is of the model TEC1-12706, the hot end of each semiconductor refrigeration element TEC-D is connected with the heat spreading plate of the central heat dissipation module through a graphene thermal conductive pad, and the cold end is tightly coupled with the rear part of the camera core through a silver-containing thermal conductive silicone grease with a thermal conductivity > 7.5 W / mK; the temperature sensor is a MF52 series 10kΩ NTC thermistor, which is fixed at key positions including the rear part of the camera core near the coupling position of the TEC-D cold end; the STM32-based detector temperature control unit uses unified PID closed-loop control to forcibly and synchronously stabilize all detectors at an optimal low-temperature working point of 15℃±0.1℃.

5. The actively thermal managed smart infrared gas imaging device according to claim 2, wherein, in the second thermal management component, N independent filter temperature control and tuning modules are included; each filter temperature control and tuning module uses a temperature sensor integrated with the corresponding temperature-controlled and tuned infrared filter to collect filter temperature in real time; a semiconductor refrigeration element TEC-F is used to independently adjust the temperature of each filter, which is thermally coupled with each temperature-controlled and tuned infrared filter; a filter tuning unit with a built-in wavelength-temperature calibration model receives the filter temperature collected by the temperature sensor in real time, drives each TEC-F, and accurately tunes the center wavelength of the corresponding filter to a set value; the set value includes M characteristic absorption peak wavelengths of the measured gas and K background wavelength bands without absorption of the measured gas.

6. The actively thermal managed smart infrared gas imaging device according to claim 5, wherein, each filter temperature control and tuning module in the second thermal management component includes: a red copper precisely processed heat sink and mechanical base, the hot end of which is connected to the central heat dissipation module; a ring-shaped precision tuning TEC, the hot face of which is fixed to the base, and the cold face of which carries a tunable infrared filter; a tunable infrared filter, which uses vanadium dioxide as the core tuning material and is mechanically pre-tightened and fixed to the TEC cold face through a red copper threaded compression ring; a Pt1000 thin-film platinum resistance temperature sensor, which is fixed to the temperature measurement base of the red copper threaded compression ring and is integrated with the filter; a PEEK thermal insulation ring, which thermally isolates the module from the aluminum alloy optical carrier plate.

7. The actively thermal managed smart infrared gas imaging device according to claim 1, wherein, the image processing module includes a multi-channel registration module, a background fusion and noise reduction module, a difference operation module, and a multi-dimensional feature recognition and decision module; wherein, the multi-channel registration module uses one measurement image as a reference to perform sub-pixel level registration on the remaining N-1 images using an offline calibrated homography matrix; the background fusion and noise reduction module performs pixel-level weighted averaging on K registered background images to generate a "super-clean" background reference image; the difference operation module separately differentiates M measurement images from the background reference image to obtain M high signal-to-noise ratio difference images. The multi-dimensional feature recognition and decision module adopts a pre-trained U-Net CNN network, takes M differential images as input, recognizes and segments each input differential image to output a corresponding probability map, and then thresholdizes each probability map to obtain a binary mask corresponding to each probability map, and then performs a pixel-level "logical AND" operation on the binary masks corresponding to each probability map. When the pixel region that is "true" in all masks reaches a set area, the gas is detected, and the result after the logical AND operation is output as the final segmentation mask.

8. The actively thermal managed smart infrared gas imaging device according to any one of claims 1-7, wherein, the measured gas is SF6, and the smart infrared gas imaging device is used for optical detection of SF6 gas leakage; and is configured as 4 parallel infrared imaging channels in a 2x2 matrix form, wherein, 2 of the 4 channels are configured as measurement channels by temperature control of the second thermal management component, and the center wavelengths of the filters are tuned to the main absorption peak 10.55 pm and the secondary absorption peak 10.8 pm of SF6, respectively; and the remaining 2 channels are configured as background channels, and the center wavelengths of the filters are tuned to the background reference wavelength band 11.0 pm and the background reference wavelength band 11.2 pm, respectively.

9. A gas imaging method based on the active thermal management based intelligent infrared gas imaging apparatus according to any one of claims 1-8, characterized in that, comprising: S1, detector low-temperature stabilization; the first thermal management component is started to synchronously cool the infrared detectors of all N infrared imaging channels, and waits for them to be stabilized at a preset optimal low-temperature working point; S2, spectral channel setting; the second thermal management component receives wavelength setting instructions for each infrared imaging channel, independently tunes the center wavelengths of the N filters to the wavelengths corresponding to the absorption peaks of the measured gas and the wavelengths corresponding to the background wavelength bands without absorption of the measured gas, respectively, and waits for stabilization; S3, synchronous parallel acquisition; all N channels perform parallel image sequence acquisition under synchronous triggering to obtain measurement images of the measured gas absorption and background images without absorption; S4, image fusion processing and recognition; the acquisition and processing unit performs image registration and background processing on the synchronous acquired measurement images of the measured gas absorption and background images without absorption, and then performs image recognition, region segmentation and detection to obtain the outline and position of the measured gas; S5, result visualization; the finally recognized outline and position information of the measured gas are superimposed on the measurement image corresponding to the main absorption peak in real time to realize intuitive and visual monitoring of the leakage position, range and diffusion trend.

Images