VOCs gas leakage real-time monitoring system based on dual-light image fusion

Abstract

The invention discloses a VOCs gas leakage real-time monitoring system based on dual-light image fusion, which comprises a dual-light acquisition module, a preprocessing module and a fusion and identification module, and is characterized in that the dual-light acquisition module comprises a visible light lens and an infrared lens; the preprocessing module comprises a gas path dust removal unit and a signal conditioning unit; the fusion and identification module carries out dynamic registration, multi-scale fusion and intelligent noise reduction processing on the preprocessed image. According to the VOCs gas leakage real-time monitoring system based on dual-light image fusion disclosed by the invention, through linkage of the dual-light acquisition module, the preprocessing module and the fusion and identification module, high-sensitivity detection and real-time early warning of VOCs leakage are realized in a complex industrial scene, and meanwhile, the false alarm rate is reduced.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of VOCs gas leakage monitoring, and particularly relates to a VOCs (volatile organic compounds) gas leakage real-time monitoring system based on dual-light image fusion. BACKGROUND

[0002] Traditional VOCs gas leakage monitoring has the following difficulties: 1. Limitations of single modal detection: Traditional infrared gas detection relies on single waveband (such as mid-wave infrared 3.0-3.5 μm) absorption characteristics, which is easily disturbed by environmental background (such as high-temperature pipelines, steam, etc.), resulting in high false alarm rate.

[0003] Visible light images cannot capture gas leakage characteristics and can only assist in locating the leakage point, which needs to rely on manual inspection, resulting in low efficiency and safety hazards.

[0004] 2. Insufficient adaptability to complex scenes: High-risk scenes such as oil refineries and chemical plants have a large number of pipelines, valves and other high-density equipment, and traditional sensors (such as PID, FID) need to be in close contact, which is difficult to cover hidden leakage points.

[0005] Existing optical gas imaging (OGI) equipment has insufficient sensitivity to small leaks (<0.4 g / min), and relies on manual image interpretation, which cannot achieve real-time early warning.

[0006] 3. Defects in algorithm and hardware cooperation: Traditional image fusion technology is not optimized for gas leakage dynamic diffusion characteristics, resulting in poor separation effect of gas clouds and background.

[0007] Existing systems lack all-weather monitoring capability, and high temperature, high humidity and dusty environments can easily cause sensor performance degradation.

[0008] Therefore, the above problems are further improved. SUMMARY

[0009] The main purpose of the present application is to provide a VOCs gas leakage real-time monitoring system based on dual-light image fusion, which realizes high-sensitivity detection and real-time early warning of VOCs leakage in complex industrial scenes through the linkage of dual-light acquisition module, preprocessing module, fusion and recognition module, while reducing the false alarm rate.

[0010] To achieve the above purpose, the present application provides a VOCs gas leakage real-time monitoring system based on dual-light image fusion, which comprises a dual-light acquisition module, a preprocessing module, a fusion and recognition module, wherein: The dual light acquisition module includes a visible light lens and an infrared lens, which respectively obtain a visible light image and an infrared image, and the visible light lens and the infrared lens are cooperatively processed; The preprocessing module includes an air path dust removal unit and a signal conditioning unit, the air path dust removal unit filters impurities including dust and oil mist in the air through a multi-stage filter element to prevent attachment to the lens surface or interference with infrared radiation transmission, and the signal conditioning unit performs non-uniformity correction on the infrared image; The fusion and recognition module performs dynamic registration, multi-scale fusion and intelligent noise reduction processing on the preprocessed image.

[0011] As a further preferred technical solution of the above technical solution, the cooperative processing is specifically implemented as: Optical axis coaxial calibration: the center axes of the visible light lens and the infrared lens are aligned through a mechanical structure to ensure that the fields of view of the two lenses completely coincide and eliminate spatial misplacement caused by optical axis deviation; Synchronous zooming and focusing: step motors are used to drive the two lenses to zoom synchronously, and laser ranging feedback is used to adjust the focus in real time to ensure consistent image clarity at different distances; Timing synchronization: based on a hardware trigger signal, the visible light and infrared image acquisition frame rates are synchronized to avoid time difference problems in dynamic scenes.

[0012] As a further preferred technical solution of the above technical solution, the dynamic registration is specifically implemented as: Feature extraction: using the SIFT algorithm, key points and feature descriptors are extracted from the visible light image and the infrared image; Feature matching: KD-Tree is used to quickly match the feature points of the two images, and the RANSAC algorithm is used to eliminate false matching points; Affine transformation: an affine transformation matrix is calculated according to the matching points to map the infrared image to the visible light image coordinate system, eliminating the difference in viewing angle; Real-time update: the registration parameters are dynamically updated for each frame of image to adapt to the gimbal rotation or target movement scene.

[0013] As a further preferred technical solution of the above technical solution, the multi-scale fusion is specifically implemented as: Wavelet decomposition: the infrared image is decomposed into low-frequency approximation components and high-frequency detail components through 3-layer discrete wavelet transform; Feature extraction: the edge information of the gas cloud in the high-frequency detail components of the infrared image is retained, and the background noise is suppressed; Weighted fusion: the RGB channels of the visible light image and the infrared high-frequency detail components are superimposed according to the weight to enhance the contrast of the gas edge; Wavelet reconstruction, inverse transform generates fusion image, ensures that the gas cloud is highlighted in the visible light background.

[0014] As a further preferred technical solution of the above technical solution, the intelligent noise reduction is specifically implemented as: Network improvement, based on YOLOv8 architecture, increase channel attention module, strengthen gas feature extraction; introduce lightweight Backbone, improve inference speed; Training strategy, using synthetic data enhancement, combined with real industrial scene dataset, improve model robustness; Post-processing, through confidence threshold and non-maximum suppression, filter false alarm targets. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 It is a structural schematic diagram of the present application.

[0016] Figure 2 It is a flow chart of dynamic registration of the present application.

[0017] Figure 3 It is a flow chart of multi-scale fusion of the present application. DETAILED DESCRIPTION

[0018] 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 as examples, and other obvious modifications can be thought 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.

[0019] In the preferred embodiments of the present application, those skilled in the art should note that the visible light and the like involved in the present application can be regarded as prior art.

[0020] Preferred embodiments.

[0021] As Figures 1-3 shown, the present application discloses a VOCs gas leakage real-time monitoring system based on dual-light image fusion, which comprises a dual-light acquisition module, a preprocessing module (i.e. Figure 1 the preprocessing unit), a fusion and recognition module, wherein: The dual-light acquisition module includes a visible light lens and an infrared lens (installed on the detector), which respectively obtain visible light (environmental perception) images and infrared (gas signature) images, and perform collaborative processing of the visible light lens and the infrared lens (visible light lens: captures optical images of the scene, providing high-resolution visual information (such as the appearance of the equipment and the spatial positioning of the leak location) for detailed identification and spatial coordinate calibration of the target object; infrared lens: utilizes the absorption characteristics of VOCs gas for infrared radiation in specific bands (such as the mid-infrared band of 3-5μm or 8-14μm, with different absorption peaks for different gases) to detect changes in the infrared radiation intensity of the gas cloud. When VOCs leak, the gas molecules absorb infrared radiation in the environment, resulting in "dark spots" or abnormal radiation intensity in the infrared image of the area); The pre-processing module includes an air path dust removal unit and a signal conditioning unit. The air path dust removal unit filters impurities such as dust and oil mist in the air through a multi-stage filter element (PTFE membrane + activated carbon) to prevent them from adhering to the lens surface or interfering with infrared radiation transmission (preventing false signals or image blur caused by impurities). The signal conditioning unit (using an FPGA chip) performs non-uniformity correction (NUC) on the infrared image. Due to the differences in the response characteristics of each pixel of the infrared detector, the uncorrected image will have fixed pattern noise (FPN). NUC eliminates background noise by calibrating the offset and gain coefficient of each pixel in real time, improving the signal-to-noise ratio of the infrared image, making the radiation anomaly of the VOCs gas cloud more clearly discernible. For non-uniformity correction, the response curve of each pixel of the infrared detector is obtained through a blackbody radiation source (normal temperature and high temperature), and the gain and offset coefficient of each pixel are calculated in real time through FPGA. The formula is: ; After the non-uniformity correction, the coordinates are The output grayscale value (or quantitative value such as radiation brightness) of the pixel is the target result of the correction process and is used for accurate analysis of subsequent images (for example, in VOCs gas leakage monitoring, the pixel value of the infrared image that clearly reflects the gas distribution after correction) so that different pixels can output more consistent and accurate values ​​under the same real radiation input. The coordinates collected by the infrared detector The original grayscale value of the pixel point (the uncorrected original response output. Due to the detector manufacturing process and other reasons, different pixels respond differently to the same radiation, so the original value will have non-uniformity problems and is the "raw material" for correction processing.

[0022] For coordinates The pixel gain coefficient is used to compensate for the gain difference of the detector pixel response; Offset coefficients of the pixel points for coordinates Offset coefficients of the pixel points for coordinates

[0023] Dynamic update: automatically trigger NUC every 30 minutes to adapt to temperature drift; Temporal noise reduction: use 3-frame weighted average method (weights 0.5, 0.3, 0.2) to suppress random noise.

[0024] The fusion and recognition module respectively performs dynamic registration, multi-scale fusion and intelligent noise reduction processing on the preprocessed images (including visible light images and infrared images).

[0025] Specifically, for cooperative processing, the specific implementation is: Optical axis coaxial calibration: align the center axes of the visible light lens and the infrared lens through mechanical structure, ensure that the fields of view of the two lenses completely overlap, and eliminate spatial misplacement caused by optical axis offset; Synchronous zooming and focusing: use a stepper motor to drive the two lenses to synchronously zoom (in the range of 30-150mm), and adjust the focus in real time through laser ranging feedback to ensure consistent image clarity at different distances; Timing synchronization: based on hardware trigger signals (such as GPIO pulses), ensure that the frame rates of visible light and infrared image acquisition are synchronized (such as 30fps), and avoid time difference problems in dynamic scenes.

[0026] More specifically, for dynamic registration (eliminate the offset of dual-optical images caused by viewing angle difference through feature point matching (SIFT algorithm) and affine transformation), the specific implementation is: Feature extraction: use SIFT (Scale-Invariant Feature Transform) algorithm to extract key points (such as corner points, edge intersection points) and feature descriptors (SIFT scale invariance: extract multi-scale feature points through Gaussian difference (DoG) pyramid to ensure matching robustness at different resolutions) from visible light images and infrared images; Feature matching: quickly match the feature point pairs (i.e. the key points and feature descriptors obtained above) of the two images through KD-Tree, and eliminate false matching points combined with RANSAC algorithm; Affine transformation: calculate the affine transformation matrix (including translation, rotation, scaling parameters) according to the matching point pairs, map the infrared image to the visible light image coordinate system, and eliminate the viewing angle difference; affine transformation matrix: Real-time update: dynamically update the registration parameters for each frame of image to adapt to the pan-tilt rotation or target moving scene.

[0027] Further, for multi-scale fusion (using wavelet transform to extract gas features in the infrared image, superimposed with the visible light background, and enhance the edge contrast of gas cloud), the specific implementation is: Wavelet decomposition, 3-layer discrete wavelet transform (DWT) is performed on the infrared image, decomposed into low-frequency approximation component (LL) and high-frequency detail component (LH, HL, HH); Daubechies4 (db4) wavelet basis is selected, and the decomposition formula is: ; This is the wavelet coefficient, which represents the signal After discrete wavelet transform with db4 as the wavelet basis, the transform result at scale j , displacement k reflects the "similarity" of the original signal and db4 wavelet basis function at this scale and displacement, which is the key data for subsequent signal analysis (such as feature extraction in VOCs gas leakage monitoring) based on wavelet coefficients.

[0028] Feature extraction, retaining gas cloud edge information (such as gradient change) in the high-frequency detail component of the infrared image, suppressing background noise; Weighted fusion, superimposing the RGB channels of the visible light image and the infrared high-frequency detail component according to the weight (such as infrared weight 0.7, visible light 0.3), enhancing the edge contrast of the gas; weighted fusion model: Infrared high-frequency component weight = 0.7 (highlighting the gas edge); Visible light weight = 0.3 (retaining background details); The fusion formula is: , is the weight of the infrared high-frequency detail component, is the visible light weight, is the infrared high-frequency detail component, is the visible light image, represents the final fused image, which is the result image obtained by superimposing the infrared high-frequency component image and the visible light image according to a certain weight, and it fuses the gas-related features (such as VOCs gas cloud edge) in the infrared image and the background detail information in the visible light image, which is used for subsequent more intuitive and accurate analysis (such as in VOCs gas leakage monitoring, clearly presenting the fusion view of gas distribution and scene background).

[0029] Wavelet reconstruction, inverse transform to generate a fusion image, ensuring that the gas cloud is highlighted in the visible light background.

[0030] Further, for intelligent noise reduction (based on YOLOv8 improved deep learning model, distinguishing between real leakage and interference (such as steam, dust), reducing false alarm rate) is implemented as: Network improvement, based on YOLOv8 architecture, add channel attention module (ECA-Net), strengthen gas feature extraction; Introduce lightweight Backbone (such as MobileNetV3), improve inference speed; ECA-Net module: embed high-efficiency channel attention mechanism in Backbone, calculate channel weight, formula: , is the feature map, is the Sigmoid function; Training strategy, use synthetic data enhancement (add steam, dust interference), combined with real industrial scene dataset, improve model robustness; Training data synthesis: use CycleGAN to generate interference scene images containing steam and dust, enhance model anti-interference ability; Post-processing, through confidence threshold (>0.9) and non-maximum suppression (NMS), filter false alarm targets; Push video stream through RTSP protocol, synchronize alarm information to the cloud.

[0031] Preferably, the present application also includes all-weather monitoring architecture: Active temperature control system: built-in semiconductor cooling module, ensures stable operation of infrared detector in -20℃~60℃ environment.

[0032] Explosion-proof and protective design: IP67 protection level, meets ExdbIICT6 explosion-proof certification, suitable for flammable and explosive areas, including: Explosion-proof shell: double-layer 304 stainless steel structure, explosion-proof joint gap ≤0.1mm, meets ExdbIICT6 standard.

[0033] Heat dissipation design: filled with heat-conducting silica gel inside, external heat dissipation fins, ensures safe heat dissipation through explosion-proof gap.

[0034] Sealing protection: IP67 level sealing ring (red mark), all interfaces use M20 waterproof connector (blue mark).

[0035] Preferably, the present application also includes a cloud collaborative early warning module: Edge computing: built-in AI chip (such as Huawei Ascend 310), realizes local real-time gas recognition and leakage positioning.

[0036] Multi-level alarm: upload leakage coordinates and concentration curve to smart park platform through 4G / 5G, trigger sound and light alarm and work order push.

[0037] It is worth mentioning that the visible light 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 of these technical features can be selected conventionally 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 can be replaced equivalently, and any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A real-time VOCs gas leakage monitoring system based on dual-light image fusion, characterized in that: It includes dual-light acquisition module, pre-processing module, fusion and recognition module, among which: The dual-light acquisition module includes a visible light lens and an infrared lens, which respectively obtain visible light images and infrared images, and perform collaborative processing on the visible light lens and the infrared lens; The pre-processing module includes an air path dust removal unit and a signal conditioning unit. The air path dust removal unit filters impurities such as dust and oil mist in the air through a multi-stage filter element to prevent them from adhering to the lens surface or interfering with infrared radiation transmission. The signal conditioning unit performs non-uniformity correction on the infrared image. The fusion and recognition module performs dynamic registration, multi-scale fusion and intelligent noise reduction on the pre-processed images.

2. A real-time VOCs gas leakage monitoring system based on dual-light image fusion according to claim 1, characterized in that: The specific implementation of collaborative processing is: Optical axis coaxial calibration: The central axes of the visible light lens and the infrared lens are aligned through a mechanical structure to ensure that the fields of view of the two lenses completely overlap and eliminate spatial misalignment caused by optical axis offset; Synchronous zoom and focus: A stepper motor drives the two lenses for synchronous zoom, and laser ranging feedback is used to adjust the focus in real time to ensure consistent image clarity at different distances. Timing synchronization: Based on hardware trigger signals, it ensures the synchronization of visible light and infrared image acquisition frame rates to avoid time difference problems in dynamic scenes.

3. The real-time VOCs gas leakage monitoring system based on dual-light image fusion according to claim 2 is characterized in that: The specific implementation for dynamic registration is: Feature extraction: Use the SIFT algorithm to extract key points and feature descriptors from visible light images and infrared images; Feature matching: KD-Tree is used to quickly match feature point pairs between two images, and the RANSAC algorithm is used to eliminate mismatched points. Affine transformation: Calculate the affine transformation matrix based on the matching point pairs, map the infrared image to the visible light image coordinate system, and eliminate the perspective difference; Real-time update: Dynamically update the registration parameters for each frame to adapt to the scene of gimbal rotation or target movement.

4. The real-time VOCs gas leakage monitoring system based on dual-light image fusion according to claim 3 is characterized in that: The specific implementation of multi-scale fusion is as follows: Wavelet decomposition: perform three-layer discrete wavelet transform on the infrared image to decompose it into low-frequency approximate components and high-frequency detail components; Feature extraction: retaining the edge information of gas clouds in the high-frequency detail components of infrared images and suppressing background noise; Weighted fusion: superimposes the RGB channels of the visible light image and the infrared high-frequency detail components according to weights to enhance the gas edge contrast; Wavelet reconstruction and inverse transform are used to generate fused images, ensuring that the gas clouds are highlighted against the visible light background.

5. The real-time VOCs gas leakage monitoring system based on dual-light image fusion according to claim 4 is characterized in that: The specific implementation of intelligent noise reduction is as follows: Network improvements: Based on the YOLOv8 architecture, a channel attention module is added to enhance gas feature extraction; a lightweight backbone is introduced to improve inference speed; Training strategies, using synthetic data augmentation combined with real industrial scenario datasets to improve model robustness; Post-processing, filtering out false alarm targets through confidence threshold and non-maximum suppression.

Images