Image recognition-based online assessment system, method, device and medium for visual pollution and transparency of emission plumes

Abstract

This invention discloses an online assessment system, method, equipment, and medium for visual pollution and transparency of emission plumes based on image recognition. The system includes: an image acquisition module for continuously capturing images of the plume at the chimney outlet; an image preprocessing module for receiving pollutant concentration data from a continuous emission monitoring system; an intelligent assessment module for extracting visual features of the plume from the preprocessed images and outputting plume opacity level and morphodynamic characteristics; and a correlation analysis module for receiving the plume opacity level, morphodynamic characteristics, and pollutant concentration data, performing correlation analysis by calculating correlation coefficients, and determining whether emissions are abnormal based on a multivariate model. This invention achieves real-time online monitoring of emission plumes by continuously capturing images of the plume at the chimney outlet through the image acquisition module, overcoming the limitations of manual visual inspection in terms of time and space.

Description

Technical Field

[0001] This invention belongs to the field of image monitoring technology, and in particular relates to an online assessment system, method, device and medium for visual pollution and transparency of emission plumes based on image recognition. Background Technology

[0002] In environmental protection work, monitoring and assessing the visual pollution level and transparency of gas turbine emission plumes is crucial. Currently, environmental protection departments mainly rely on the Ringelmann blackness method, which involves direct observation of plume blackness for assessment. However, this method has significant drawbacks. First, manual observation is highly subjective. Different observers may make significant deviations in their judgment of the blackness of the same plume due to differences in visual perception, experience level, and psychological state, leading to a lack of consistency and reliability in the assessment results. Second, this method cannot achieve continuous monitoring. Observation activities are limited by time and space, making it difficult to capture real-time changes during the emission process. Especially during production load adjustments or sudden emission events, it is easy to miss key anomalies, failing to meet the needs of real-time supervision. In addition, manual observation is inefficient, requiring a large investment of manpower and resources for on-site observation and data recording, which is costly and prone to errors due to fatigue. Therefore, existing technologies cannot achieve objective, accurate, and real-time assessment of the visual pollution and transparency of emission plumes. There is an urgent need for an online assessment system that can overcome subjectivity, discontinuity, and insufficient data analysis. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide an online assessment system for visual pollution and transparency of emission plumes based on image recognition, which addresses the shortcomings of the prior art. By continuously capturing images of the plumes at the chimney outlet through an image acquisition module, the system achieves real-time online monitoring of emission plumes, overcomes the limitations of manual visual inspection in time and space, and can capture dynamic changes in the emission process in a timely manner, thus meeting the requirements of continuous supervision.

[0004] The first aspect of this invention discloses an online assessment system for visual pollution and transparency of emission plumes based on image recognition, including an image acquisition module, an image preprocessing module, an intelligent assessment module, a correlation analysis module, and a data communication module; The image acquisition module is configured to continuously capture images of the plume from the chimney outlet and transmit the image data to the image preprocessing module; The image preprocessing module is connected to the data communication module and is configured to receive pollutant concentration data from the continuous emission monitoring system and dynamically adjust image processing parameters according to the pollutant concentration data. The image processing parameters include color threshold segmentation parameters and color space weight coefficients to segment the plume region and correct illumination. The intelligent evaluation module is connected to the image preprocessing module and is configured to extract visual features of the plume from the preprocessed image and output the plume blackness level and morphodynamic features, including the lift rate, diffusion angle, turbulence intensity and blackness gradient. The correlation analysis module is connected to the intelligent assessment module and the data communication module, and is configured to receive the plume opacity level, morphodynamic characteristics and pollutant concentration data, perform correlation analysis by calculating correlation coefficients, and determine whether the emissions are abnormal based on a multivariate model. The image preprocessing module, intelligent assessment module, and correlation analysis module interact with each other via a data bus, and the data communication module is configured to transmit the pollutant concentration data in real time.

[0005] In the above system, the dynamic adjustment of parameters in the image preprocessing module includes: When the water vapor concentration in the pollutant concentration data exceeds the first preset threshold, the color threshold for segmenting the plume region is adjusted from the normal range to the high humidity range, and an evaluation model specifically designed for high humidity plumes is invoked. When the concentration of a specific pollutant in the pollutant concentration data exceeds a second preset threshold, the color space weighting coefficient is adjusted to enhance the perception of the characteristic color of the specific pollutant. The normal color threshold range is defined as [R1,R2], [G1,G2], [B1,B2] in the RGB color space, and the high humidity color threshold range is defined as [R3,R4], [G3,G4], [B3,B4], where R1, R2, G1, G2, B1, B2, R3, R4, G3, G4, B3, and B4 are preset constants. The color space weighting coefficients include the red component weight wR, the green component weight wG, and the blue component weight wB, and the adjusted weights are wR', wG', and wB', which are calculated based on the pollutant concentration using a preset function.

[0006] The illumination correction in the image preprocessing module of the above system is based on Retinex theory and includes the following steps: Gaussian filtering is applied to the original image to obtain a blurred image, which is used as an estimate of the illumination components; The reflection component is calculated using a formula, and then normalized to obtain the corrected image. The original image is represented as the product of the reflection component and the illumination component, using the following formula:

[0007] The formula for calculating the reflection component is:

[0008] in, This represents the pixel value of the original image at coordinates (x, y). Represents the reflection component. Indicates the light component.

[0009] In the above system, the visual feature extraction of smoke plumes by the intelligent evaluation module includes both traditional image processing methods and deep learning methods. The conventional image processing method includes calculating the contrast between the average gray level of the smoke plume region and the average gray level of the pure sky region, and mapping the contrast to the Ringelmann black level. The formulas for calculating the average gray level of the smoke plume region and the average gray level of the clean sky region are as follows:

[0010] Where G represents the average gray level, and n represents the total number of pixels in the region. This represents the grayscale value of the i-th pixel; The formula for calculating the contrast ratio is:

[0011] Where C represents contrast. This represents the average gray level of the plume region. This represents the average gray level of the pure sky area; The deep learning method includes using a convolutional neural network model to classify images to directly output the smoke plume opacity level.

[0012] The convolutional neural network model in the above system includes convolutional layers, pooling layers, and fully connected layers; The convolutional layer uses a 3×3 convolutional kernel, with a stride of 1 and padding of 1. The pooling layer uses a 2×2 pooling kernel with a step size of 2 to perform max pooling operation. The fully connected layer is configured to integrate features and output classification results; The convolutional neural network model is obtained by training labeled plume images, and the training process uses the cross-entropy loss function and the backpropagation algorithm.

[0013] The morphological dynamics feature analysis of the intelligent evaluation module in the above system includes: The rise rate is calculated using the optical flow method, which determines the rise distance of the plume per unit time based on the pixel motion vectors of consecutive image frames. The diffusion angle is calculated by extracting the plume contour using an edge detection algorithm and using the chimney outlet as a reference point to calculate the angle change between the contour and the vertical direction. Turbulence intensity is quantified by statistically analyzing the standard deviation of pixel distribution within the reference region of the plume profile; The blackness gradient is obtained by meshing the plume region and calculating the average grayness gradient of adjacent meshes. The formula for calculating the grayness gradient is as follows: ,

[0014] in, Represents a grid The average gray value within, This represents the grayscale gradient in the horizontal direction. This represents the grayscale gradient in the vertical direction.

[0015] In the above system, the multivariate model of the correlation analysis module determines the emission anomaly as follows: The correlation coefficient between visual blackness and pollutant concentration exceeds the preset threshold, and the pollutant concentration exceeds the standard, and the visual blackness exceeds the standard or the plume rise rate decreases abnormally and the diffusion angle increases abnormally. The correlation coefficient is the Pearson correlation coefficient. The pollutant concentration exceeding the standard and the visual blackness exceeding the standard are determined by comparing real-time data with historical normal ranges. The abnormal decrease in plume rise speed is defined as the speed being lower than 80% of the historical average speed. The abnormal increase in diffusion angle is defined as the angle being greater than 120% of the historical average diffusion angle.

[0016] The formula for calculating the Pearson correlation coefficient in the above system is as follows:

[0017] in, Indicates the first The visual blackness value at a given moment. Represents the average value of visual blackness. Indicates the first The pollutant concentration value at a given time. This represents the average concentration of pollutants, and n represents the number of data points.

[0018] The system also includes an early warning module, configured to generate alarm information when the correlation analysis module determines that emissions are abnormal, and transmit it to an external monitoring platform via a network interface.

[0019] The second aspect of this invention discloses an online assessment method for visual pollution and transparency of emission plumes based on image recognition, comprising the following steps: Step S1: Continuously capture images of the plume at the chimney outlet using an image acquisition device, and transmit the image data to the preprocessing step; Step S2: Receive pollutant concentration data from the continuous emission monitoring system, and dynamically adjust the image processing parameters according to the pollutant concentration data. The image processing parameters include color threshold segmentation parameters and color space weight coefficients to perform image segmentation and illumination correction on the plume region. Step S3: Extract visual features of the plume from the preprocessed image and output the plume opacity level and morphodynamic features, including lift rate, diffusion angle, turbulence intensity and opacity gradient. Step S4: Receive the plume opacity level, morphodynamic characteristics and pollutant concentration data, perform correlation analysis by calculating correlation coefficients, and determine whether the emissions are abnormal based on a multivariate model; In steps S2, S3 and S4, data interaction is performed through a data bus, and the pollutant concentration data is transmitted in real time through data communication.

[0020] A third aspect of the present invention discloses an electronic device, comprising: a memory and a processor, wherein the processor and the memory are connected; The memory is used to store programs; The processor invokes a program stored in the memory to execute the method provided in the second aspect embodiment described above.

[0021] The fourth aspect of the present invention discloses a computer-readable storage medium having a computer program stored thereon, the computer program being executed by a computer to perform the method provided in the second aspect embodiment.

[0022] Compared with existing technologies, this invention has the following advantages: By continuously capturing images of the plume at the chimney outlet through the image acquisition module, real-time online monitoring of emitted plumes is achieved, overcoming the limitations of manual visual inspection in terms of time and space. It can promptly capture dynamic changes during the emission process, meeting the needs of continuous monitoring. The image preprocessing module dynamically adjusts image processing parameters, such as color threshold segmentation parameters and color space weighting coefficients, based on pollutant concentration data from the Continuous Emission Monitoring System (CEMS). This enables the system to adapt to complex and changing emission environments. For example, it optimizes the segmentation threshold at high water vapor concentrations or enhances color perception when the concentration of specific pollutants (such as NO2) increases, effectively reducing identification errors caused by environmental interference and significantly improving the objectivity and accuracy of the assessment. This solves the problems of strong subjectivity and inconsistent data in manual judgment. The intelligent assessment module provides multi-dimensional plume state information by extracting plume opacity levels and morphodynamic characteristics (such as rise rate, diffusion angle, turbulence intensity, and opacity gradient), achieving a more comprehensive visual pollution assessment and compensating for the shortcomings of manual visual inspection that relies solely on opacity levels. The correlation analysis module integrates data on plume opacity level, morphodynamic characteristics, and pollutant concentrations based on a multivariate model. By calculating correlation coefficients, it performs correlation analysis to comprehensively determine emission anomalies. For example, it can provide timely warnings even when plume opacity is within acceptable limits but plume morphology is abnormal, enhancing the system's reliability and practicality. Overall, the system significantly reduces reliance on manual labor and improves monitoring efficiency through automated image recognition and data fusion, providing stable and reliable data support for environmental decision-making.

[0023] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0024] Figure 1 This is a system architecture diagram of Embodiment 1 of the present invention.

[0025] Figure 2 This is a flowchart of the method in Embodiment 2 of the present invention. Detailed Implementation

[0026] Example 1 like Figure 1 As shown, an online assessment system for visual pollution and transparency of emission plumes based on image recognition includes an image acquisition module, an image preprocessing module, an intelligent assessment module, a correlation analysis module, and a data communication module. The image acquisition module is configured to continuously capture images of the plume from the chimney outlet and transmit the image data to the image preprocessing module; The image preprocessing module is connected to the data communication module and is configured to receive pollutant concentration data from the continuous emission monitoring system and dynamically adjust image processing parameters according to the pollutant concentration data. The image processing parameters include color threshold segmentation parameters and color space weight coefficients to segment the plume region and correct illumination. The intelligent evaluation module is connected to the image preprocessing module and is configured to extract visual features of the plume from the preprocessed image and output the plume blackness level and morphodynamic features, including the lift rate, diffusion angle, turbulence intensity and blackness gradient. The correlation analysis module is connected to the intelligent assessment module and the data communication module, and is configured to receive the plume opacity level, morphodynamic characteristics and pollutant concentration data, perform correlation analysis by calculating correlation coefficients, and determine whether the emissions are abnormal based on a multivariate model. The image preprocessing module, intelligent assessment module, and correlation analysis module interact with each other via a data bus, and the data communication module is configured to transmit the pollutant concentration data in real time.

[0027] During implementation, the image acquisition module continuously captures images of the plume from the chimney outlet using a high-definition camera. The camera is installed at a distance of 1.5 times the height of the chimney; for example, if the chimney is 100 meters high, the camera is set at 150 meters, with the lens vertically aimed at the chimney outlet to ensure coverage of the plume and the sky background. Camera parameters are automatically adjusted according to ambient light: under strong midday sunlight, the ISO is set to 100, the shutter speed to 1 / 1000 second, and the aperture to f / 8; at night, the ISO is increased to 1600, the shutter speed is decreased to 1 / 60 second, and the aperture is widened to f / 2.8. Through automatic exposure and white balance functions, images are transmitted to the image preprocessing module at a rate of 25 frames per second. This automatic adjustment of camera parameters ensures stable and high-quality image acquisition, providing a consistent input source for subsequent processing.

[0028] The image preprocessing module is connected to the data communication module to receive pollutant concentration data from the continuous emission monitoring system in real time. When the water vapor concentration exceeds the first preset threshold of 50 grams per cubic meter, the module dynamically adjusts the color threshold for plume region segmentation. The normal color threshold range is defined in the RGB color space as red component R from 100 to 150, green component G from 100 to 150, and blue component B from 150 to 200; after adjustment, the high humidity color threshold range is set to red component R from 120 to 170, green component G from 120 to 170, and blue component B from 170 to 220, making the system more sensitive to white or gray plumes. Simultaneously, an evaluation model specifically designed for high humidity plumes is invoked, which is trained based on a high humidity image dataset. When the NO2 concentration exceeds the second preset threshold of 10 mg / m³, the module adjusts the color space weight coefficients. The normal weights are: red component weight wR 0.3, green component weight wG 0.5, and blue component weight wB 0.2. New weights are calculated using a linear function based on the NO2 concentration. For example, at a concentration of 15 mg / m³, the new weights are wR' 0.5, wG' 0.3, and wB' 0.2, enhancing the perception of brownish-red hues. After parameter adjustment, plume region segmentation is performed: first, a color threshold is used to initially screen image regions; then, a gray-level co-occurrence matrix is ​​applied to calculate texture features, such as regions with a contrast value greater than 50, which are identified as plume boundaries. Next, illumination correction is performed. Based on Retinex theory, the original image is Gaussian filtered using a 5x5 filter kernel and a standard deviation of 1.5, resulting in a blurred image as an estimate of the illumination components. Then, the reflection component is calculated using a formula: the original image pixel value divided by the illumination component estimate. The reflection component is then normalized, scaling the pixel values ​​to the range of 0 to 255. By dynamically adjusting image processing parameters and illumination correction, adaptive processing for different emission conditions is achieved, improving the accuracy and robustness of plume segmentation.

[0029] The intelligent evaluation module extracts visual features of the plume from the preprocessed image, outputting the plume's blackness level and morphological dynamics. The blackness level is evaluated using a combination of traditional image processing methods and deep learning methods. Traditional image processing methods calculate the contrast between the average grayscale of the plume region and the average grayscale of the clear sky region; for example, the average grayscale of the plume region... The average gray level of the pure sky area is 120. The contrast ratio C is 80, which is equal to the absolute value 120 minus 80 divided by 80, resulting in 0.5. According to the preset mapping relationship, a contrast ratio of 0.5 corresponds to a Ringelmann blackness level of 2. The deep learning method uses a pre-trained convolutional neural network model. The model structure includes convolutional layers, pooling layers, and fully connected layers. The convolutional layers use 3x3 kernels with a stride of 1 and padding of 1. The pooling layers use 2x2 kernels with a stride of 2, performing max pooling. The fully connected layers integrate features and output the classification results. During model training, a large number of labeled plume images are used, and the weights are optimized using the cross-entropy loss function and backpropagation algorithm. After inputting the pre-processed image, the model directly outputs the blackness level, such as the highest probability level being level 2. Morphodynamic features include lift rate, diffusion angle, turbulence intensity, and blackness gradient. The lift rate is calculated using optical flow. For two consecutive frames, the Lucas-Kanade method is applied to analyze pixel motion vectors. For example, if a specific point at the top of the plume rises 5 meters per unit time, the lift rate is 5 meters per second. The diffusion angle is calculated by extracting the plume contour using an edge detection algorithm, and then calculating the angle between the contour and the vertical direction using the chimney outlet as a reference point; for example, the average angle is 25 degrees. Turbulence intensity is quantified by statistically analyzing the standard deviation of the pixel distribution within the reference region of the plume contour; for example, a standard deviation of 8 indicates a high degree of contour irregularity. The blackness gradient is calculated by dividing the plume region into a 10x10 grid, calculating the average gray value of each grid, and then using the central difference method to calculate the grayness gradient between adjacent grids, including the horizontal gradient. Equals average gray level divided by 2, vertical gradient Similar calculations were performed. By integrating multiple assessment methods, a comprehensive quantitative analysis of plume visual pollution and transparency was achieved, improving the accuracy of the assessment results.

[0030] The correlation analysis module receives plume opacity level and morphodynamic characteristics from the intelligent assessment module, as well as pollutant concentration data transmitted from the data communication module. It performs correlation analysis by calculating correlation coefficients. Using the Pearson correlation coefficient formula, it calculates the correlation between visual opacity values ​​and dust concentration values. For example, taking data from five time points with visual opacity values ​​of 2, 3, 2, 4, and 3, and dust concentration values ​​of 10, 15, 12, 20, and 18 mg / m³, the calculated correlation coefficient r is 0.9, exceeding the preset threshold of 0.8. The correlation analysis module determines whether emissions are abnormal based on a multivariate model. The conditions are: the correlation coefficient between visual opacity and pollutant concentration exceeds the threshold; pollutant concentration exceeds the standard; and visual opacity exceeds the standard or the plume rise velocity decreases abnormally while the diffusion angle increases abnormally. Pollutant concentration exceeding the standard is determined by comparing real-time data with historical normal ranges. For example, the historical normal range for dust concentration is 5 to 15 mg / m³; a current value of 20 mg / m³ is considered exceeding the standard. Visual opacity exceeding the standard is defined as exceeding level 3. An abnormal decrease in plume rise velocity is defined as a velocity falling below 80% of the historical average velocity. For example, if the historical average velocity is 6 m / s, a current velocity of 4.8 m / s would be considered an abnormal decrease. An abnormal increase in diffusion angle is defined as an angle exceeding 120% of the historical average diffusion angle. For example, if the historical average angle is 20 degrees, a current angle of 24 degrees would be considered an abnormal increase. When all conditions are met, emissions are considered abnormal. Through multivariate correlation analysis, reliable detection of abnormal events is achieved, reducing false alarms caused by single factors.

[0031] The data communication module transmits pollutant concentration data from the external Continuous Emission Monitoring System (CEMS) in real time. The image preprocessing module, intelligent assessment module, and correlation analysis module interact via a data bus to ensure the coordinated operation of all system components. By integrating image acquisition, preprocessing, intelligent assessment, and correlation analysis, the entire system achieves objective, real-time online assessment of the visual pollution and transparency of emission plumes, providing continuous data support for environmental regulation.

[0032] In one embodiment, the dynamic adjustment of parameters in the image preprocessing module includes: When the water vapor concentration in the pollutant concentration data exceeds the first preset threshold, the color threshold for segmenting the plume region is adjusted from the normal range to the high humidity range, and an evaluation model specifically designed for high humidity plumes is invoked. When the concentration of a specific pollutant in the pollutant concentration data exceeds a second preset threshold, the color space weighting coefficient is adjusted to enhance the perception of the characteristic color of the specific pollutant. The normal color threshold range is defined as [R1,R2], [G1,G2], [B1,B2] in the RGB color space, and the high humidity color threshold range is defined as [R3,R4], [G3,G4], [B3,B4], where R1, R2, G1, G2, B1, B2, R3, R4, G3, G4, B3, and B4 are preset constants. The color space weighting coefficients include the red component weight wR, the green component weight wG, and the blue component weight wB, and the adjusted weights are wR', wG', and wB', which are calculated based on the pollutant concentration using a preset function.

[0033] During implementation, when the water vapor concentration exceeds the first preset threshold of 50 grams per cubic meter, the module automatically adjusts the color threshold for plume region segmentation. For example, the normal color threshold range in the RGB color space is 100 to 150 for the red component R, 100 to 150 for the green component G, and 150 to 200 for the blue component B; after adjustment, the high humidity color threshold range is 120 to 170 for the red component R, 120 to 170 for the green component G, and 170 to 220 for the blue component B. The system directly applies the new range by querying the preset threshold mapping table, making the segmentation algorithm more sensitive to white or gray plumes. Simultaneously, it calls an evaluation model specifically designed for high humidity plumes, trained on a high humidity image dataset, capable of identifying plume features under water vapor interference. When the NO2 concentration exceeds the second preset threshold of 10 mg / m³, the module adjusts the color space weighting coefficients. The normal weights are: red component weight wR 0.3, green component weight wG 0.5, and blue component weight wB 0.2. After adjustment, the new weights are wR' 0.5, wG' 0.3, and wB' 0.2. The system directly outputs the results through a linear interpolation function, enhancing the perception of brownish-red features in the image. Through dynamic parameter adjustment, adaptive optimization of image processing is achieved, improving the accuracy and environmental adaptability of plume segmentation.

[0034] In one embodiment, the illumination correction of the image preprocessing module is based on Retinex theory and includes the following steps: Gaussian filtering is applied to the original image to obtain a blurred image, which is used as an estimate of the illumination components; The reflection component is calculated using a formula, and then normalized to obtain the corrected image. The original image is represented as the product of the reflection component and the illumination component, using the following formula:

[0035] The formula for calculating the reflection component is:

[0036] in, This represents the pixel value of the original image at coordinates (x, y). Represents the reflection component. Indicates the light component.

[0037] In implementation, the original image is Gaussian filtered using a 5x5 filter kernel and a standard deviation of 1.5, directly yielding a blurred image as an estimate of the illumination components. For example, the input image pixel value matrix is ​​filtered to generate a smoothed version, and each pixel value in the illumination component estimation matrix is ​​a weighted average of the original pixels. Then, the reflection component is calculated by dividing the original image pixel value by the illumination component estimate; for example, the original pixel value of 150 at coordinates (x, y) is divided by the illumination estimate of 120, resulting in a reflection component of 1.25. Finally, the reflection component is normalized, scaling the value to the range of 0 to 255; for example, multiplying by 255 and rounding, yielding a corrected image pixel value of 191. Illumination correction using Retinex theory eliminates the influence of variable illumination, providing a stable and consistent image foundation for subsequent analysis.

[0038] In one embodiment, the plume visual feature extraction of the intelligent evaluation module includes traditional image processing methods and deep learning methods; The conventional image processing method includes calculating the contrast between the average gray level of the smoke plume region and the average gray level of the pure sky region, and mapping the contrast to the Ringelmann black level. The formulas for calculating the average gray level of the smoke plume region and the average gray level of the clean sky region are as follows:

[0039] Where G represents the average gray level, and n represents the total number of pixels in the region. This represents the grayscale value of the i-th pixel; The formula for calculating the contrast ratio is:

[0040] Where C represents contrast. This represents the average gray level of the plume region. This represents the average gray level of the pure sky area; The deep learning method includes using a convolutional neural network model to classify images to directly output the smoke plume opacity level.

[0041] In practice, traditional image processing methods map Ringelmann black levels by calculating the contrast between the smoke plume region and the clear sky region, such as the average gray level of the smoke plume region. The average gray level of the pure sky area is 120. The value is 80, and the calculated contrast ratio (C) is 0.5, directly corresponding to a blackness level of 2. The deep learning method uses a convolutional neural network model. The model takes a pre-processed image as input and outputs the blackness level probability directly through forward propagation. For example, after calculating the input image, the model outputs a probability of 0.85 for level 2 and a probability of 0.15 for level 3, ultimately determining the blackness level as 2. By combining traditional and deep learning methods, redundant verification of blackness evaluation is achieved, improving the accuracy and robustness of the results.

[0042] In one embodiment, the convolutional neural network model includes convolutional layers, pooling layers, and fully connected layers; The convolutional layer uses a 3×3 convolutional kernel, with a stride of 1 and padding of 1. The pooling layer uses a 2×2 pooling kernel with a step size of 2 to perform max pooling operation. The fully connected layer is configured to integrate features and output classification results; The convolutional neural network model is obtained by training labeled plume images, and the training process uses the cross-entropy loss function and the backpropagation algorithm.

[0043] In implementation, the model includes convolutional layers, pooling layers, and fully connected layers. The convolutional layers use 3x3 kernels with a stride of 1 and padding of 1 to convolve the input image. For example, if the input image size is 224×224, the output feature map size remains unchanged after convolution, and each feature map is enhanced with an activation function to improve non-linearity. The pooling layers use 2x2 kernels with a stride of 2 to perform max pooling, effectively halving the feature map size to 112×112 while preserving the main features. The fully connected layers integrate all features and output the classification result, mapping the feature vectors to six categories and outputting the probability of each blackness level. The model is trained using labeled images, with weights optimized through backpropagation. After training, the model is directly applied to real-time images. By concretizing the model structure, efficient feature extraction is achieved, ensuring fast and accurate blackness classification.

[0044] In one embodiment, the morphodynamic feature analysis of the intelligent evaluation module includes: The rise rate is calculated using the optical flow method, which determines the rise distance of the plume per unit time based on the pixel motion vectors of consecutive image frames. The diffusion angle is calculated by extracting the plume contour using an edge detection algorithm and using the chimney outlet as a reference point to calculate the angle change between the contour and the vertical direction. Turbulence intensity is quantified by statistically analyzing the standard deviation of pixel distribution within the reference region of the plume profile; The blackness gradient is obtained by meshing the plume region and calculating the average grayness gradient of adjacent meshes. The formula for calculating the grayness gradient is as follows: ,

[0045] in, Represents a grid The average gray value within, This represents the grayscale gradient in the horizontal direction. This represents the grayscale gradient in the vertical direction.

[0046] In practice, the rise rate is calculated using optical flow, based on pixel motion vectors from consecutive image frames. For example, applying the Lucas-Kanade method to two adjacent frames analyzes the motion of pixels at the top of the plume, resulting in a rise distance of 5 meters per unit time and a rise rate of 5 meters per second. The spread angle is calculated by extracting the plume contour using an edge detection algorithm, using the chimney outlet as a reference point to calculate the angle between the contour and the vertical direction. For example, after using the Canny algorithm to detect edges, the angle is calculated to be 25 degrees. Turbulence intensity is quantified by statistically analyzing the standard deviation of pixel distribution within the reference region. For example, in a 100×100 pixel area, a standard deviation of 8 indicates a high degree of contour irregularity. The blackness gradient is calculated by meshing the plume region, calculating the average gray value of each mesh, and then using the central difference method to calculate the gradient, such as the horizontal gradient. The calculated result is 2, vertical gradient. The calculation result is 1.5. Through detailed quantification of morphological characteristics, a comprehensive analysis of plume dynamics was achieved, enhancing the depth of visual pollution assessment.

[0047] In one embodiment, the multivariate model of the correlation analysis module determines the emission anomaly based on the following conditions: The correlation coefficient between visual blackness and pollutant concentration exceeds the preset threshold, and the pollutant concentration exceeds the standard, and the visual blackness exceeds the standard or the plume rise rate decreases abnormally and the diffusion angle increases abnormally. The correlation coefficient is the Pearson correlation coefficient. The pollutant concentration exceeding the standard and the visual blackness exceeding the standard are determined by comparing real-time data with historical normal ranges. The abnormal decrease in plume rise speed is defined as the speed being lower than 80% of the historical average speed. The abnormal increase in diffusion angle is defined as the angle being greater than 120% of the historical average diffusion angle.

[0048] During implementation, the model checks whether the correlation coefficient between visual opacity and pollutant concentration exceeds a preset threshold of 0.8. For example, the Pearson correlation coefficient is calculated to be 0.9, exceeding the threshold. Pollutant concentration exceeding the standard is determined by comparing real-time data with historical normal ranges. For example, the historical normal range for dust concentration is 5 to 15 mg / m³, while the current value of 20 mg / m³ is considered exceeding the standard. Visual opacity exceeding the standard is defined as exceeding level 3; the current opacity level is 2, not exceeding the standard. An abnormal decrease in plume rising speed is defined as a speed lower than 80% of the historical average speed (the historical average speed is 6 m / s, and the current speed of 4.8 m / s is considered an abnormal decrease). An abnormal increase in diffusion angle is defined as an angle greater than 120% of the historical average diffusion angle (the historical average angle is 20 degrees, and the current angle of 24 degrees is considered an abnormal increase). When these conditions are met, emissions are deemed abnormal. Through multi-variable comprehensive judgment, accurate identification of abnormal events is achieved, improving the reliability of the system.

[0049] In one embodiment, the formula for calculating the Pearson correlation coefficient is:

[0050] in, Indicates the first The visual blackness value at a given moment. Represents the average value of visual blackness. Indicates the first The pollutant concentration value at a given time. This represents the average concentration of pollutants, and n represents the number of data points.

[0051] In practice, the formula is used to calculate the correlation between visual blackness values ​​and pollutant concentration values. For example, taking data from five time points with visual blackness values ​​of 2, 3, 2, 4, and 3, and dust concentration values ​​of 10, 15, 12, 20, and 18 mg / m³, the Pearson correlation coefficient is calculated to be 0.9. The system directly applies the formula and outputs the result without displaying intermediate calculation steps. Through the Pearson correlation coefficient, a quantitative correlation between visual and concentration data is achieved, providing an objective basis for anomaly detection.

[0052] In one embodiment, an early warning module is further included, configured to generate alarm information when the correlation analysis module determines that the emissions are abnormal, and transmit the alarm information to an external regulatory platform via a network interface.

[0053] During implementation, when the correlation analysis module determines abnormal emissions, the early warning module generates an alarm message, such as the text "Emission Abnormality: Opacity Level 2, Dust Concentration Exceeds Standard, Abnormal Morphological Characteristics." The alarm is transmitted to an external monitoring platform via a network interface, for example, using the TCP / IP protocol to send it to a designated port on the server with IP address 192.168.1.100. This real-time early warning function enables timely environmental regulatory responses, improving the system's practicality and timeliness.

[0054] Example 2 like Figure 2 As shown, an online assessment method for visual pollution and transparency of emission plumes based on image recognition includes the following steps: Step S1: Continuously capture images of the plume at the chimney outlet using an image acquisition device, and transmit the image data to the preprocessing step; Step S2: Receive pollutant concentration data from the continuous emission monitoring system, and dynamically adjust the image processing parameters according to the pollutant concentration data. The image processing parameters include color threshold segmentation parameters and color space weight coefficients to perform image segmentation and illumination correction on the plume region. Step S3: Extract visual features of the plume from the preprocessed image and output the plume opacity level and morphodynamic features, including lift rate, diffusion angle, turbulence intensity and opacity gradient. Step S4: Receive the plume opacity level, morphodynamic characteristics and pollutant concentration data, perform correlation analysis by calculating correlation coefficients, and determine whether the emissions are abnormal based on a multivariate model; In steps S2, S3 and S4, data interaction is performed through a data bus, and the pollutant concentration data is transmitted in real time through data communication.

[0055] The online assessment method for visual pollution and transparency of emission plumes based on image recognition provided in this embodiment has the same implementation principle and technical effect as the system embodiment in Embodiment 1. For the sake of brevity, any parts not mentioned in the method embodiment can be referred to the corresponding content in Embodiment 1.

[0056] Example 3 A computer-readable storage medium having a computer program stored thereon, the computer program being executed by a computer to perform the online assessment method for visual pollution and transparency of emission plumes based on image recognition as described in Embodiment 2 above.

[0057] Example 4 An electronic device includes: a memory and a processor, wherein the processor and the memory are connected; The memory is used to store programs; The processor calls a program stored in the memory to execute an online assessment method for visual pollution and transparency of emission plumes based on image recognition, as described in Embodiment 2.

[0058] It should be noted that the electronic device mentioned may be, but is not limited to, personal computers (PCs), tablet computers, mobile internet devices (MIDs), etc.

[0059] It should be noted that processors, memory, and other components that may be present in electronic devices are electrically connected to each other, directly or indirectly, to enable data transmission or interaction. For example, processors, memory, and other components may be electrically connected to each other via one or more communication buses or signal lines.

[0060] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0061] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can also be implemented in other ways. The system embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0062] In addition, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0063] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, laptop, server, mobile phone, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0064] The above description is merely a preferred embodiment of the present invention and does not constitute any limitation on the present invention. Any simple modifications, alterations, or equivalent structural changes made to the above embodiments based on the technical essence of the present invention shall still fall within the protection scope of the present invention.

Claims

1. An online assessment system for visual pollution and transparency of emission plumes based on image recognition, characterized in that, It includes an image acquisition module, an image preprocessing module, an intelligent evaluation module, a correlation analysis module, and a data communication module; The image acquisition module is configured to continuously capture images of the plume from the chimney outlet and transmit the image data to the image preprocessing module; The image preprocessing module is connected to the data communication module and is configured to receive pollutant concentration data from the continuous emission monitoring system and dynamically adjust image processing parameters according to the pollutant concentration data. The image processing parameters include color threshold segmentation parameters and color space weight coefficients to segment the plume region and correct illumination. The intelligent evaluation module is connected to the image preprocessing module and is configured to extract visual features of the plume from the preprocessed image and output the plume blackness level and morphodynamic features, including the lift rate, diffusion angle, turbulence intensity and blackness gradient. The correlation analysis module is connected to the intelligent assessment module and the data communication module, and is configured to receive the plume opacity level, morphodynamic characteristics and pollutant concentration data, perform correlation analysis by calculating correlation coefficients, and determine whether the emissions are abnormal based on a multivariate model. The image preprocessing module, intelligent assessment module, and correlation analysis module interact with each other via a data bus, and the data communication module is configured to transmit the pollutant concentration data in real time.

2. The system according to claim 1, characterized in that, The dynamic adjustment of parameters in the image preprocessing module includes: When the water vapor concentration in the pollutant concentration data exceeds the first preset threshold, the color threshold for segmenting the plume region is adjusted from the normal range to the high humidity range, and an evaluation model specifically designed for high humidity plumes is invoked. When the concentration of a specific pollutant in the pollutant concentration data exceeds a second preset threshold, the color space weighting coefficient is adjusted to enhance the perception of the characteristic color of the specific pollutant. The normal color threshold range is defined as [R1,R2], [G1,G2], [B1,B2] in the RGB color space, and the high humidity color threshold range is defined as [R3,R4], [G3,G4], [B3,B4], where R1, R2, G1, G2, B1, B2, R3, R4, G3, G4, B3, and B4 are preset constants. The color space weighting coefficients include the red component weight wR, the green component weight wG, and the blue component weight wB, and the adjusted weights are wR', wG', and wB', which are calculated based on the pollutant concentration using a preset function.

3. The system according to claim 1, characterized in that, The illumination correction in the image preprocessing module is based on Retinex theory and includes the following steps: Gaussian filtering is applied to the original image to obtain a blurred image, which is used as an estimate of the illumination components; The reflection component is calculated using a formula, and then normalized to obtain the corrected image. The original image is represented as the product of the reflection component and the illumination component, using the following formula: The formula for calculating the reflection component is: in, This represents the pixel value of the original image at coordinates (x, y). Represents the reflection component. Indicates the light component.

4. The system according to claim 1, characterized in that, The visual feature extraction of plumes in the intelligent evaluation module includes traditional image processing methods and deep learning methods. The conventional image processing method includes calculating the contrast between the average gray level of the smoke plume region and the average gray level of the pure sky region, and mapping the contrast to the Ringelmann black level. The formulas for calculating the average gray level of the smoke plume region and the average gray level of the clean sky region are as follows: Where G represents the average gray level, and n represents the total number of pixels in the region. This represents the grayscale value of the i-th pixel; The formula for calculating the contrast ratio is: Where C represents contrast. This represents the average gray level of the plume region. This represents the average gray level of the pure sky area; The deep learning method includes using a convolutional neural network model to classify images to directly output the smoke plume opacity level.

5. The system according to claim 4, characterized in that, The convolutional neural network model includes convolutional layers, pooling layers, and fully connected layers; The convolutional layer uses a 3×3 convolutional kernel, with a stride of 1 and padding of 1. The pooling layer uses a 2×2 pooling kernel with a step size of 2 to perform max pooling operation. The fully connected layer is configured to integrate features and output classification results; The convolutional neural network model is obtained by training labeled plume images, and the training process uses the cross-entropy loss function and the backpropagation algorithm.

6. The system according to claim 1, characterized in that, The morphological dynamics feature analysis of the intelligent evaluation module includes: The rise rate is calculated using the optical flow method, which determines the rise distance of the plume per unit time based on the pixel motion vectors of consecutive image frames. The diffusion angle is calculated by extracting the plume contour using an edge detection algorithm and using the chimney outlet as a reference point to calculate the angle change between the contour and the vertical direction. Turbulence intensity is quantified by statistically analyzing the standard deviation of pixel distribution within the reference region of the plume profile; The blackness gradient is obtained by meshing the plume region and calculating the average grayness gradient of adjacent meshes. The formula for calculating the grayness gradient is as follows: ， in, Represents a grid The average gray value within, This represents the grayscale gradient in the horizontal direction. This represents the grayscale gradient in the vertical direction.

7. The system according to claim 1, characterized in that, The multivariate model in the correlation analysis module determines the emission anomalies based on the following conditions: The correlation coefficient between visual blackness and pollutant concentration exceeds the preset threshold, and the pollutant concentration exceeds the standard, and the visual blackness exceeds the standard or the plume rise rate decreases abnormally and the diffusion angle increases abnormally. The correlation coefficient is the Pearson correlation coefficient. The pollutant concentration exceeding the standard and the visual blackness exceeding the standard are determined by comparing real-time data with historical normal ranges. The abnormal decrease in plume rise speed is defined as the speed being lower than 80% of the historical average speed. The abnormal increase in diffusion angle is defined as the angle being greater than 120% of the historical average diffusion angle.

8. An online assessment method for visual pollution and transparency of emission plumes based on image recognition, characterized in that, Includes the following steps: Step S1: Continuously capture images of the plume at the chimney outlet using an image acquisition device, and transmit the image data to the preprocessing step; Step S2: Receive pollutant concentration data from the continuous emission monitoring system, and dynamically adjust the image processing parameters according to the pollutant concentration data. The image processing parameters include color threshold segmentation parameters and color space weight coefficients to perform image segmentation and illumination correction on the plume region. Step S3: Extract visual features of the plume from the preprocessed image and output the plume opacity level and morphodynamic features, including lift rate, diffusion angle, turbulence intensity and opacity gradient. Step S4: Receive the plume opacity level, morphodynamic characteristics and pollutant concentration data, perform correlation analysis by calculating correlation coefficients, and determine whether the emissions are abnormal based on a multivariate model.

9. An electronic device, characterized in that, include: A memory and a processor, wherein the processor and the memory are connected; The memory is used to store programs; The processor invokes a program stored in the memory to execute the method of claim 8.

10. A computer-readable storage medium, characterized in that, It stores a computer program, which is executed by a computer to perform the method of claim 8.

Images