Inspection image enhancement method for safety production

Abstract

The invention specifically relates to an inspection image enhancement method for safety production, and relates to the technical field of image processing, and the method comprises the steps: collecting and preprocessing multi-source data; mie scattering characteristic analysis and transmissivity theoretical calculation; a transmissivity correction model driven by on-site concentration; carrying out image enhancement and noise reduction processing; and performing model optimization and safety production scene verification. According to the method, multi-source data acquisition and preprocessing comprehensively cover a safety production scene, and the influence of particulate matters on light propagation is accurately described by combining a Mie scattering theory and a transmissivity correction model driven by field concentration; actual interferences such as non-spherical particles, agglomeration effect and airflow unevenness are fully considered, and a pixel-level accurate transmissivity matrix is generated; differentiated noise processing is carried out for scenes with different concentrations, and key information such as equipment defects is effectively restored in combination with a detail enhancement strategy; the method provides a more reliable clear image for safety production inspection, assists inspection personnel and algorithms in accurately identifying hidden dangers, reduces the risks of missing inspection and false inspection, and guarantees the production safety.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to an inspection image enhancement method for safe production. Background Art

[0002] During production safety inspections, image acquisition is a key means of obtaining on-site information. However, particulate matter (such as dust and water mist) commonly found in industrial environments strongly scatters light. This scattering severely degrades the captured inspection images, resulting in blurry images and loss of detail. This significantly interferes with the subsequent identification of equipment defects (such as cracks and leaks), posing a significant challenge to the investigation of production safety hazards. Traditional image enhancement methods, often based on general image processing algorithms, fail to incorporate the physical properties of on-site particulate matter (such as particle size distribution and concentration). This makes it difficult to accurately restore image details in complex scattering environments and fails to meet the stringent image quality requirements for production safety.

[0003] While Mie scattering theory provides a theoretical framework for describing the interaction between particles and light, and can explain the scattering and attenuation of light from a physical perspective, it relies on idealized assumptions (such as spherical particles and uniform distribution). In actual safety production scenarios, particles have complex morphologies (including fibrous and agglomerated forms) and uneven distribution, resulting in significant deviations between theoretically calculated transmittance and the real-world scenario. Directly applying this theory to image enhancement makes it difficult to effectively restore image details, and thus fails to provide reliable visual support for inspections. Therefore, there is an urgent need for an image enhancement method that integrates Mie scattering theory with actual field data and is adapted to complex industrial environments. Summary of the Invention

[0004] The purpose of the present invention is to solve the above problems and to propose an inspection image enhancement method for safe production.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions: A patrol image enhancement method for safe production, comprising: Multi-source data acquisition and preprocessing: Collect image data, particle characteristics data, and environmental auxiliary data from inspection scenes, and perform cleaning, calibration, and standardization; Mie scattering characteristics analysis and transmittance theoretical calculation: Based on the physical properties of particles and light source characteristics, the scattering parameters and theoretical transmittance values ​​of light in a particle environment are calculated; On-site concentration-driven transmittance correction model: Establishes a mapping relationship between on-site particle concentration and transmittance to correct theoretical calculation errors; Image enhancement and noise reduction: Based on the corrected transmittance model, a degradation model is constructed. The scattered light interference is removed and optimized from the degraded image through the inversion formula. A differentiated algorithm is used to reduce noise according to the noise characteristics under different concentrations. At the same time, features related to hidden dangers are enhanced, and finally a clear image that meets objective and application indicators is output.

[0006] Preferably, multi-source data collection and preprocessing specifically include: Inspection image data: Capture images of production safety scenes, covering scenes with different lighting conditions and particle concentrations, with resolution and frame rate matching inspection requirements; Particle concentration data: Sensors deployed on-site collect particle concentration data in real time and record particle size distribution. Environmental auxiliary data: including on-site temperature, humidity, light source parameters, and shooting distance; The collected data is preprocessed, including removing motion blur, cropping invalid areas, and unifying the size and format of image data; the sensor data is calibrated for particle concentration sensors, synchronized with timestamps, and outliers are removed.

[0007] Preferably, Mie scattering characteristic analysis and transmittance theoretical calculation specifically include: Based on the Mie scattering theory formula, the scattering cross section, extinction cross section, and scattering phase function of a single particle are calculated; Combined with the particle concentration, the volume scattering coefficient and volume extinction coefficient are calculated, and the theoretical value of transmittance is derived to output the theoretical transmittance distribution at different positions and concentrations.

[0008] Preferably, the transmittance correction model driven by on-site concentration specifically includes: A regression method is used, with measured concentration and environmental parameters as input, theoretical transmittance as initial value, and the corrected transmittance is output; Considering the spatial heterogeneity of particle concentration in the scene, the discrete sensor concentration data is expanded into a continuous concentration field through interpolation, and then combined with the correction model to obtain the spatial transmittance distribution; Outputs a transmittance matrix that matches the inspection image pixels and is verified by field data.

[0009] Preferably, the image enhancement and noise reduction processing specifically includes: Degradation model formula: For any pixel in the image, the degradation model is:

[0010] in, is the pixel value of the degraded image; Provide a clear image of the target; is the corrected transmittance; is the atmospheric light component; is the noise component.

[0011] Preferably, inversion and optimization from degradation to clarity are also included; Remove scattered light interference from degraded images and restore the object's original reflective properties; According to the degradation model deformation, we get:

[0012] in, This is a clear image of the preliminary inversion; Based on the noise characteristics at different concentrations, a differentiation algorithm is used to suppress noise without destroying key details; Enhance the features related to hidden dangers in the image and improve the recognition efficiency of manual or algorithmic methods.

[0013] Preferably, special optimizations to adapt to safe production scenarios are also included: For dynamic scenes, the inversion and noise reduction algorithms are lightweight; In scenes with no light or high-density whiteout, infrared image-assisted enhancement is combined; the contour information of the infrared image is superimposed on the visible light inversion result to retain the location information of the device.

[0014] Preferably, the final output is an enhanced clear image, which must meet the following requirements: Objective indicators: PSNR>30dB and SSIM>0.9 compared with the reference clear image; PSNR is peak signal-to-noise ratio, and SSIM is structural similarity.

[0015] Preferably, it also includes: model optimization and safety production scenario verification, specifically including: Calculate the PSNR and SSIM of the enhanced image and the reference clear image to evaluate the fidelity; Calculate the noise variance reduction ratio to evaluate the noise reduction effect; The image clarity is evaluated by production safety inspectors; the recognition accuracy before and after enhancement is tested using a defect recognition algorithm.

[0016] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: 1. This invention comprehensively covers production safety scenarios through multi-source data acquisition and preprocessing, and combines Mie scattering theory with a transmittance correction model driven by on-site concentration to accurately characterize the impact of particulate matter on light propagation. It fully considers actual interference such as non-spherical particles, agglomeration effects, and uneven airflow to generate a pixel-level accurate transmittance matrix. It differentially processes noise for different concentration scenarios and combines detail enhancement strategies to effectively restore key information such as equipment defects. It provides more reliable and clear images for production safety inspections, assists inspection personnel and algorithms in accurately identifying hidden dangers, reduces the risk of missed detection and false detection, and ensures production safety.

[0017] 2. Through multi-cooperative operations, the present invention lays a solid foundation from basic data, adapts theoretical models to reality, then goes on to the practical application of image enhancement, and finally continuously optimizes through scenario verification; for different production safety scenarios, it can retroactively adjust Mie scattering parameters, correct model algorithms, and optimize filtering strategies to continuously adapt to scenario requirements; improve the accuracy of manual identification of equipment defects, and meet objective indicators. It not only meets the current inspection image enhancement needs, but also reserves optimization space for future scene changes, providing long-term stable and efficient technical support for production safety inspections and promoting the development of intelligent inspections. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Further details, features and advantages of the present application are disclosed in the following description of exemplary embodiments in conjunction with the accompanying drawings, in which: Figure 1 Flow chart of the method of the present invention. DETAILED DESCRIPTION

[0019] Several embodiments of the present application will be described in more detail below with reference to the accompanying drawings so that those skilled in the art can implement the present application. The present application can be embodied in many different forms and for many different purposes and should not be limited to the embodiments described herein. These embodiments are provided to make the present application comprehensive and complete and to fully convey the scope of the present application to those skilled in the art. The embodiments do not limit the present application.

[0020] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. It will be further understood that terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the relevant art and / or the context of this specification, and will not be interpreted in an idealized or overly formal sense unless expressly defined as such herein.

[0021] Example 1 The specific implementation method is combined with the attached Figure 1 Provide detailed explanation.

[0022] Attachment Figure 1 A flowchart of an inspection image enhancement method for safe production provided by an embodiment of the present invention shows the complete steps from multi-source data acquisition and preprocessing to model optimization and safe production scenario verification.

[0023] See also Figure 1 , a patrol image enhancement method for safe production, including the following parts: Multi-source data acquisition and preprocessing: Collect image data, particle characteristics data, and environmental auxiliary data from inspection scenes, and perform cleaning, calibration, and standardization; Specifically include: Inspection image data: Images of production safety sites (such as workshops, coal mines, and chemical plants) are captured using fixed cameras, drones, or handheld inspection equipment. These images cover scenes with varying lighting conditions (daytime / nighttime) and particulate matter concentrations (low / medium / high pollution). The resolution and frame rate must match inspection requirements (for example, high-definition images are required to identify millimeter-level defects). Particle concentration data: The particle concentration (unit: ), and record the particle size distribution of the particles (such as 0.1-10 The proportion of particles in the range needs to be measured by a particle size analyzer), because the Mie scattering characteristics are closely related to the particle size; Environmental auxiliary data: including on-site temperature, humidity (affecting the aggregation state of particulate matter), light source parameters (such as the wavelength and intensity of the inspection light), shooting distance (the length of the light propagation path, which affects scattering attenuation), etc., collected through temperature and humidity sensors, light source controllers and other equipment; The collected data is preprocessed, including removing motion blur (caused by device jitter), cropping invalid areas (such as black edges), and unifying the size and format of image data; calibrating the particle concentration sensor (eliminating zero drift), synchronizing timestamps (ensuring that the image matches the concentration data at the corresponding moment), and eliminating outliers (such as jump data caused by sensor failure).

[0024] Mie scattering characteristics analysis and theoretical transmittance calculation: Based on the physical properties of particles (particle size, refractive index) and light source characteristics, the scattering parameters and theoretical transmittance values ​​of light in a particulate matter environment are calculated; Specifically include: Based on the Mie scattering theory formula, the scattering cross section (the ratio of scattered light energy to incident light energy), extinction cross section (the total attenuation of scattering and absorption), and scattering phase function (the distribution of scattered light in space) of a single particle are calculated; Combined with the particle concentration (number of particles per unit volume), the volume scattering coefficient and volume extinction coefficient are calculated, and the theoretical value of transmittance is derived to output the theoretical transmittance distribution at different positions and concentrations (because the particle concentration in the scene may be uneven, the transmittance needs to reflect spatial differences).

[0025] First determine the parameters: Particle intrinsic parameters (the basis for determining scattering characteristics): particle size distribution and complex refractive index ; Light source and propagation parameters (determine the propagation conditions of light): wavelength of incident light and light propagation distance ; Calculation of Mie scattering characteristics of a single particle: Dimensionless particle size parameter: (used to uniformly describe the relative relationship between particle size and wavelength)

[0026] is the particle size (radius); is the wavelength of the incident light; when When ≤1, Mie scattering degenerates into Rayleigh scattering; when When ≥1, it is close to geometric optical scattering.

[0027] Scattering efficiency factor:

[0028] Absorption efficiency factor: (indicates the efficiency of absorption)

[0029] Extinction efficiency factor: (indicates the total efficiency of scattering and absorption)

[0030] in represents taking the real part of a complex number, 、 is the recursive coefficient of Mie scattering, which needs to be calculated through complex Bessel functions and Hankel functions. In practical applications, Mie scattering calculation libraries such as MiePlot can be called to implement it.

[0031] Convert the efficiency factor into actual physical cross-section (unit: ), which represents the range of the particle's effect on light: Scattering cross section:

[0032] Extinction cross section:

[0033] The scattering characteristics of a single particle need to be combined with the particle concentration to expand the macroscopic effect of the particle group, and the key parameters describing the propagation of light in granular media are obtained: Particle number concentration, the total number of particles per unit volume, is obtained by integrating the particle size distribution:

[0034] 、 are the minimum and maximum particle sizes, respectively.

[0035] The volume scattering coefficient and volume extinction coefficient describe the total scattering / extinction ability of all particles in a unit volume to light: Volume scattering coefficient:

[0036] Volume extinction coefficient:

[0037] If the particles are monodisperse (uniform particle size), it can be simplified to , polydisperse systems require integral calculations.

[0038] Scattering phase function (describes the angular distribution of scattered light in space, used to analyze the source of haze in the image (forward scattering will cause image blur, and backward scattering will increase background noise)):

[0039] is the scattering angle, Scattering amplitude function, forward scattering The higher the proportion of ≈0°, the more severe the image blur.

[0040] Derivation of the theoretical transmittance value: Transmittance is the ratio of the remaining energy after light passes through a particulate matter environment to the incident energy. It directly determines the degree of image attenuation and is derived based on the Lambert-Beer law: If the concentration of particulate matter in the scene is uniform (such as a closed workshop), the light intensity decays exponentially with the propagation distance, and the transmittance formula is:

[0041] Spatial distribution of transmittance in non-uniform concentration scenarios: In actual inspection scenarios, particulate matter concentrations may vary spatially (for example, concentrations are higher in areas near dust sources). Therefore, point-by-point transmittance calculations are required: Divide the scene into grid units (such as 10cm×10cm), and obtain the value of each unit by interpolation. ; For each pixel , calculate the cumulative extinction along the light propagation path:

[0042] ) is a pixel The corresponding propagation distance, the integral result is the total attenuation on the path.

[0043] On-site concentration-driven transmittance correction model: Establishes a mapping relationship between on-site particle concentration and transmittance, corrects theoretical calculation errors, and improves the practical applicability of transmittance; Specifically include: Analyze the deviation between the theoretical transmittance (derived from Mie scattering characteristics analysis and transmittance theoretical calculation steps) and the actual image attenuation: In high-concentration scenarios, the scattering of non-spherical particles (such as fibrous dust) may deviate from the Mie theory (Mie assumes spherical particles), resulting in higher theoretical transmittance (actual attenuation is more severe), which needs to be corrected using measured data. On-site measured particulate matter concentration data include: Concentration values: Laser dust monitors and PM2.5 monitors collect real-time concentration data at different locations and times. This data provides a realistic basis for corrections and directly reflects the extent of particulate pollution on-site. For example, concentrations at coal mine excavation faces can fluctuate significantly due to operational changes, and this data can capture these real-world changes.

[0044] Particle size distribution supplement: In addition to concentration, the particle size distribution measured by the particle size analyzer (such as 0.1-10 Because Mie scattering is sensitive to particle size, the equivalent particle size conversion of non-spherical particles (such as fiber dust) will affect the fit of the theoretical model. This correction is more accurate.

[0045] Synchronous image degradation features: Average Grayscale: This is the average grayscale value of image pixels. Particle scattering can cause the image to appear gray and reduce brightness. The average grayscale can directly reflect attenuation. For example, in high-density scenes, the image will appear gray, and the average grayscale will be significantly lower than in clean scenes. This can quantify the degree of attenuation and help determine the theoretical transmittance error.

[0046] Contrast refers to the grayscale difference between the brightest and darkest pixels in an image. Particle scattering blurs details and reduces contrast. For example, when a weld seam is clean, it has high contrast and is clear. However, at high concentrations, scattering reduces the grayscale difference between the weld seam and the background, resulting in low contrast. Combining contrast with theoretical transmittance can determine the accuracy of the model's prediction of detail attenuation.

[0047] Using regression methods (such as linear regression and random forest), with measured concentration and environmental parameters as input and theoretical transmittance as initial value, the corrected transmittance is output; For example, for coal mine scenes, a concentration-shape factor correction term can be introduced to adjust the extinction coefficient ; Considering the spatial heterogeneity of particulate matter concentration in the scene (e.g. high concentration near the dust source), the discrete sensor concentration data is extended to a continuous concentration field through interpolation (e.g. Kriging interpolation), and then combined with the correction model to obtain the spatialized transmittance distribution (each pixel corresponds to a transmittance value); Output the transmittance matrix verified by field data that matches the pixels of the inspection image.

[0048] Mie theory assumes that particles are spherical, but in real-world scenarios such as textile workshop fiber dust and irregular coal dust in coal mines, the scattering characteristics differ greatly from spheres. For example, fiber dust scatters more in a specific direction, and the theoretical calculation of the extinction coefficient will be smaller, resulting in overestimation of the theoretical transmittance (feeling less light attenuation, but the actual attenuation is more severe); When the humidity and temperature are high in the field, particles tend to agglomerate, the equivalent particle size becomes larger, and the scattering cross-section changes. For example, water mist particles agglomerate into large droplets, and the theoretical model does not consider agglomeration, which will make the calculated transmittance inconsistent with the actual value (due to stronger attenuation of large particles), and real-time data correction is needed; In addition to particle morphology, uneven airflow and light in the field can also amplify the deviation. For example, local airflow vortex in the workshop causes uneven particle distribution, and the theoretical model calculates based on uniform distribution, which will result in higher theoretical transmittance in the vortex area (high concentration), and the actual image will be darker and the attenuation will be more severe.

[0049] Linear regression (basic but practical) is suitable for preliminary correction, assuming that the theoretical transmittance error is linearly related to the measured concentration deviation and environmental parameter difference.

[0050] For example, a linear model is constructed with the theoretical transmittance-actual attenuation corresponding transmittance as the dependent variable, and the measured concentration and theoretical assumed concentration difference, temperature difference as independent variables. The advantage is that it is simple to understand and fast to calculate, suitable for quick preliminary correction, such as routine workshop inspection, where environmental parameters fluctuate slightly and linear models can quickly adapt.

[0051] Random forest (dealing with complex nonlinearity) In real-world scenarios, the interaction between particle morphology, concentration, and environment is complex and nonlinear. Random forest can handle multi-dimensional input (such as incorporating concentration, particle size distribution, temperature, humidity, and particle morphology proportion) and uncover hidden patterns. For example, in a high-humidity and high-concentration coal mine scenario, particle agglomeration and morphology change are complex. With random forest, these factors can be integrated to accurately correct the theoretical transmittance. The disadvantage is that the calculation is slightly complex and requires certain computing power support.

[0052] Field sensors are discrete points (for example, one every 10 meters in a workshop). To obtain a continuous concentration field, kriging interpolation, based on spatial autocorrelation, can estimate concentrations at any location based on discrete points, providing a defined error range. This method is suitable for scenarios where particle concentrations vary spatially (for example, near a dust source, where concentrations gradually change from high to low). It is more accurate than simple linear interpolation and can capture spatial trends in concentrations.

[0053] Operation process: first organize the discrete concentration data of the sensor and record the position coordinates and concentration values Analyze the spatial variation function of concentration (such as Gaussian model, spherical model) to determine the change pattern of concentration with distance; based on the variation function, grid the scene (such as 1m×1m grid), interpolate the concentration point by point, and construct a continuous concentration field.

[0054] With the continuous concentration field, combined with a correction model (such as a random forest or a linear model with a shape factor), the corrected transmittance is calculated for each grid cell (corresponding to an image pixel). The final output is a transmittance matrix that matches the inspection image pixels. The value corresponding to each pixel accurately reflects the actual transmittance at that location after particle attenuation, laying the foundation for subsequent image enhancement (pixel-level correction).

[0055] The corrected transmittance matrix provides pixel-level precision. Each value is calibrated based on field measurements, accurately reflecting the attenuation of light passing through a particulate environment before reaching the camera in inspection scenarios. For example, in a workshop image, the matrix can accurately depict differences in different areas (high concentration and low transmittance near equipment exhaust ports, while low concentration and high transmittance farther away), enabling subsequent image enhancement to target each pixel.

[0056] Image enhancement and noise reduction are based on a transmittance matrix and an atmospheric scattering model to invert a clear image. This modified matrix is ​​more accurate, allowing the inverted image to best reproduce the actual scene. For example, it precisely removes the haze caused by particle scattering while preserving details of equipment defects (cracks and leaks) for enhanced enhancement.

[0057] Image enhancement and noise reduction: Based on the modified transmittance model, a degradation model is constructed. Scattered light interference is removed from the degraded image through an inversion formula and optimized. A differentiated algorithm is used to reduce noise based on noise characteristics at different concentrations. Features related to hidden dangers are also enhanced, ultimately outputting a clear image that meets both objective and application indicators. Specifically include: Based on the modified transmittance model and incorporating the physical mechanisms of image degradation, this approach uses precise inversion and targeted noise reduction algorithms to restore inspection images contaminated by particle scattering to a clear state while preserving key details required for safe production, such as equipment cracks, instrument readings, and leak traces. Its core logic is to first eliminate scattering blur and then suppress noise interference.

[0058] Degradation model formula: For any pixel in the image, the degradation model is:

[0059] in, is the pixel value of the degraded image (image to be processed); A clear image of the target (to be restored); is the corrected transmittance (spatial matrix, one value per pixel); is the atmospheric light component (the average intensity of scattered light in the environment, which affects the overall brightness offset of the image); is the noise component (including sensor thermal noise, random spot noise of scattering particles, etc.); atmospheric light Accurate acquisition of dark channel: Using the improved dark channel prior method, the pixels are filtered in the area with low transmittance after correction (such as the darker equipment shadow in the image), and the average is taken after eliminating the interference of high light, so as to avoid the traditional method in the scene with complex light source (such as multiple lights in the workshop). Overestimation.

[0060] noise Characteristic analysis: The probability distribution of noise is statistically analyzed by calculating the residual between the degraded image and the synchronously collected low-concentration clear image. At low particle concentrations, Gaussian white noise is dominant (with small variance), while at high concentrations, Poisson distribution is present (caused by the random characteristics of particle scattering), providing a basis for the subsequent selection of noise reduction algorithms.

[0061] Inversion and optimization from degradation to clarity; Remove scattered light interference from degraded images and restore the object's original reflective properties; According to the degradation model deformation, we get:

[0062] in, This is a clear image of the preliminary inversion; when When approaching 0 (high concentration area, light almost completely attenuated), directly fill in the weighted average of the neighborhood pixels to avoid numerical overflow caused by division operations (such as image black hole repair in high dust area of coal mine underground); Due to the local error of the transmittance model, guided filtering is performed on the inverted image (with the corrected transmittance as the guide image) to smooth the factors caused by fluctuations, while preserving the edges of the equipment (such as the outline of the pipeline); For different concentration levels, different algorithms are used to suppress noise while preserving key details; Low concentration scene (noise mainly in the form of Gaussian noise): Applicable scenarios: such as daily inspection in chemical plant, image with slight blur, noise showing as uniform graininess.

[0063] Algorithm selection: wavelet domain hard threshold denoising, after wavelet decomposition of the image, set dynamic threshold in high frequency subband (containing noise and details); reduce threshold for equipment texture area (such as valve thread) to preserve details, and increase threshold for background area (such as wall surface) to enhance denoising effect.

[0064] Medium to high concentration scene (noise mainly in the form of Poisson combined with impulse noise): Applicable scenarios: such as coal mine tunneling face, water mist covered cooling tower, image with a large number of random speckle and blocky blur. Algorithm selection: First, remove impulse noise (such as bright spots caused by large particle scattering) through adaptive median filtering; Then, use non-local means filtering (NL-means), utilize the correlation of repeated structures in the image (such as repeated bolts on the surface of the equipment), set a search window related to transmittance (window size reduced in high transmittance area to preserve details, window size enlarged in low transmittance area to enhance smoothing).

[0065] Noise suppression effect verification: calculate the structural similarity (SSIM) and edge preservation index (EPI) of the image before and after denoising to ensure that the EPI value of the key edges of the equipment (such as the outline of the crack) does not decrease by more than 5% (to avoid excessive denoising leading to loss of defects).

[0066] Enhance features related to hidden dangers in the image (such as color abnormalities, shape distortions), improve the recognition efficiency of artificial or algorithm.

[0067] Contrast adaptive adjustment: use limited contrast adaptive histogram equalization (CLAHE) to enhance contrast in local areas (such as equipment surface), avoid amplification of background noise caused by global equalization (such as highlighting the slight rust on the surface of the pipeline from the dark background).

[0068] Color correction: To address color shifts caused by particle scattering (e.g., stronger blue light scattering, resulting in a bluish image), the gain of each of the RGB channels is adjusted based on the relationship between the light source wavelength and transmittance (e.g., increasing the weight of the red light channel to restore the device's true color).

[0069] Key area sharpening: Use the Laplace operator to sharpen preset areas of interest (such as instrument dials and flange interfaces), highlighting the details of scale lines and sealing surfaces to ensure that inspectors can read values ​​or determine whether there is a leak.

[0070] Special optimizations to adapt to safe production scenarios: For dynamic scenarios such as drone inspections, the inversion and noise reduction algorithms are lightweight. For example, the transmittance matrix is ​​calculated in blocks, and simplified filtering is applied to non-critical areas (such as the sky and open ground), ensuring that the single-frame image processing time is less than 100ms (meeting the requirement for 25fps real-time display). In completely dark or high-concentration whiteout scenes (such as the early stages of a dust explosion), infrared image enhancement is combined with auxiliary technology. By taking advantage of the weak scattering characteristics of infrared light, the contour information of the infrared image is superimposed on the visible light inversion result to retain the location information of the equipment (for example, to determine the distribution of equipment near trapped personnel). By pre-marking critical production safety areas (such as pressure vessel welds and cable joints), protection weights are set during the noise reduction and enhancement process to ensure that the pixel values ​​in these areas are not excessively modified (for example, the grayscale difference of tiny cracks in welds is increased by more than 20% after enhancement, while the noise variance is reduced by more than 50%).

[0071] The final output is an enhanced clear image, which must meet the following requirements: Objective indicators: PSNR>30dB and SSIM>0.9 compared to the reference clear image (in medium-density scenes); PSNR is peak signal-to-noise ratio, SSIM is structural similarity; Application indicators: The manual identification accuracy of equipment defects (such as cracks with a diameter greater than 2mm) is improved by >30%, providing high-quality input for subsequent model optimization and safe production scenario verification. It can also be directly integrated into the display terminal or defect recognition algorithm of the inspection system.

[0072] Model optimization and safety production scenario verification: Verify the image enhancement effect, optimize the model based on the special needs of safety production scenarios (such as defect recognition accuracy), and improve practicality; Specifically include: Calculate the PSNR (peak signal-to-noise ratio) and SSIM (structural similarity) of the enhanced image and the reference clear image (such as the image at the same location without particles) to evaluate the fidelity; calculate the noise variance reduction ratio to evaluate the noise reduction effect; Production safety inspectors assess image clarity (e.g., whether equipment nameplates and welds are clearly visible). Defect recognition algorithms (e.g., YOLO and Faster R-CNN) are used to test recognition accuracy (e.g., missed detection rate, false detection rate) before and after enhancement to ensure that the enhanced images improve the detection rate of hidden dangers (e.g., cable damage and pipeline leaks). And optimize the corresponding steps based on the safe production scenario.

[0073] If the enhancement effect is poor in certain scenes (such as high-humidity dusty environments), retrace the Mie scattering characteristic analysis and transmittance theoretical calculation steps and correct the particle refractive index in the Mie scattering calculation (humidity affects the optical properties of the particle surface); If the transmittance model error is large, adjust the regression algorithm of the transmittance correction model step driven by the field concentration (such as introducing a deep learning model to fit the nonlinear relationship); If excessive noise reduction results in loss of detail, optimize the filtering parameters of the image enhancement and noise reduction steps (e.g., reduce the smoothing strength of the non-local mean). The optimized models and enhanced algorithms adapted to specific production safety scenarios (such as chemical workshops and underground coal mines) can be integrated into the inspection system for real-time operation.

[0074] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0075] The above description is merely illustrative of certain exemplary embodiments of the present invention. It goes without saying that those skilled in the art will be able to modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and description are illustrative in nature and should not be construed as limiting the scope of protection of the claims.

[0076] It should be noted that, in this document, if there are relational terms such as first and second, etc., they are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprises", "comprising" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprising a ..." does not exclude the presence of other identical elements in the process, method, article or device that includes the element.

[0077] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0078] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0079] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0080] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0081] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0082] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

[0083] The above description is merely illustrative of certain exemplary embodiments of the present invention. It goes without saying that those skilled in the art will be able to modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and description are illustrative in nature and should not be construed as limiting the scope of protection of the claims.

Claims

1. A patrol image enhancement method for safe production, characterized in that: include: Multi-source data acquisition and preprocessing: Collect image data, particle characteristics data, and environmental auxiliary data from inspection scenes, and perform cleaning, calibration, and standardization; Mie scattering characteristics analysis and transmittance theoretical calculation: Based on the physical properties of particles and light source characteristics, the scattering parameters and theoretical transmittance values ​​of light in a particle environment are calculated; On-site concentration-driven transmittance correction model: Establishes a mapping relationship between on-site particle concentration and transmittance to correct theoretical calculation errors; Image enhancement and noise reduction: Based on the corrected transmittance model, a degradation model is constructed. The scattered light interference is removed and optimized from the degraded image through the inversion formula. A differentiated algorithm is used to reduce noise according to the noise characteristics under different concentrations. At the same time, features related to hidden dangers are enhanced, and finally a clear image that meets objective and application indicators is output.

2. The inspection image enhancement method for production safety according to claim 1, characterized in that: Multi-source data collection and preprocessing, including: Inspection image data: Capture images of production safety scenes, covering scenes with different lighting conditions and particle concentrations, with resolution and frame rate matching inspection requirements; Particle concentration data: Sensors deployed on-site collect particle concentration data in real time and record particle size distribution. Environmental auxiliary data: including on-site temperature, humidity, light source parameters, and shooting distance; The collected data is preprocessed, including removing motion blur, cropping invalid areas, and unifying the size and format of image data; the sensor data is calibrated for particle concentration sensors, synchronized with timestamps, and outliers are removed.

3. The inspection image enhancement method for production safety according to claim 1, characterized in that: Mie scattering characteristics analysis and transmittance theoretical calculation, including: Based on the Mie scattering theory formula, the scattering cross section, extinction cross section, and scattering phase function of a single particle are calculated; Combined with the particle concentration, the volume scattering coefficient and volume extinction coefficient are calculated, and the theoretical value of transmittance is derived to output the theoretical transmittance distribution at different positions and concentrations.

4. The inspection image enhancement method for production safety according to claim 3, characterized in that: The transmittance correction model driven by field concentration includes: A regression method is used, with measured concentration and environmental parameters as input, theoretical transmittance as initial value, and the corrected transmittance is output; Considering the spatial heterogeneity of particle concentration in the scene, the discrete sensor concentration data is expanded into a continuous concentration field through interpolation, and then combined with the correction model to obtain the spatial transmittance distribution; Outputs a transmittance matrix that matches the inspection image pixels and is verified by field data.

5. The inspection image enhancement method for production safety according to claim 4, characterized in that: Image enhancement and noise reduction processing, including: Degradation model formula: For any pixel in the image, the degradation model is: in, is the pixel value of the degraded image; Provide a clear image of the target; is the corrected transmittance; is the atmospheric light component; is the noise component.

6. The inspection image enhancement method for production safety according to claim 5, characterized in that: Also included is inversion and optimization from degradation to clarity; Remove scattered light interference from degraded images and restore the object's original reflective properties; According to the degradation model deformation, we get: in, This is a clear image of the preliminary inversion; Based on the noise characteristics at different concentrations, a differentiation algorithm is used to suppress noise without destroying key details; Enhance the features related to hidden dangers in the image and improve the recognition efficiency of manual or algorithmic methods.

7. The inspection image enhancement method for production safety according to claim 6, characterized in that: It also includes special optimizations to adapt to safe production scenarios: For dynamic scenes, the inversion and noise reduction algorithms are lightweight; In scenes with no light or high-density whiteout, infrared image-assisted enhancement is combined; the contour information of the infrared image is superimposed on the visible light inversion result to retain the location information of the device.

8. The inspection image enhancement method for production safety according to claim 7, characterized in that: The final output is an enhanced clear image, which must meet the following requirements: Objective indicators: PSNR>30dB and SSIM>0.9 compared with the reference clear image; PSNR is peak signal-to-noise ratio, and SSIM is structural similarity.

9. The inspection image enhancement method for production safety according to claim 1, characterized in that: Also includes: Model optimization and safety production scenario verification, including: Calculate the PSNR and SSIM of the enhanced image and the reference clear image to evaluate the fidelity; Calculate the noise variance reduction ratio to evaluate the noise reduction effect; The image clarity is evaluated by production safety inspectors; the recognition accuracy before and after enhancement is tested using a defect recognition algorithm.

Images