Intelligent evaluation system and method for light absorption characteristics of cloud smoke particle groups

Abstract

The present application relates to the technical field of atmospheric environment monitoring, in particular to a cloud smoke particle group light absorption characteristic intelligent evaluation system and method, aiming at improving the precision of atmospheric environment monitoring technology, the system comprises data acquisition, data processing, multi-scale learning evaluation, database and user interaction module, the data acquisition module obtains the optical brightness image data of the cloud smoke particle group, the data processing module extracts the light feature data, the multi-scale learning evaluation module evaluates the light absorption characteristic based on the multi-scale learning model, including coarse-grained learning layer and fine-grained learning layer, combining macro and micro features to generate accurate light absorption parameter prediction value, the database module stores relevant data, the user interaction module allows users to query, modify data and control evaluation strategy, the system reduces the evaluation error by 40-60% in complex environment, significantly improves the evaluation precision, and can more accurately represent the light absorption characteristic of the cloud smoke particle group.

Description

Technical Field

[0001] This invention relates to the field of atmospheric environment monitoring technology, specifically to an intelligent evaluation system and method for the light absorption characteristics of cloud and smoke particle swarms. Background Technology

[0002] Cloud and haze particle swarms are common suspended matter in the atmosphere, and their light absorption characteristics have a significant impact on atmospheric visibility, climate change, and environmental monitoring. Traditional methods for assessing the light absorption characteristics of cloud and haze particle swarms are mainly based on the Lambert-Beer law, which assumes a simple linear relationship between light absorption and particle concentration. However, in real-world, complex atmospheric environments, the light absorption characteristics of cloud and haze particle swarms are influenced by various factors such as particle diameter, shape, density, temperature, and humidity, exhibiting a complex nonlinear relationship that traditional methods struggle to accurately assess.

[0003] In existing technologies, common evaluation methods can be broadly categorized into three types: first, theoretical calculation methods based on physical models, such as Mie scattering theory; second, empirical methods based on experimental measurements, such as direct measurement using a spectrometer; and third, data-driven methods based on simple statistical regression. However, each of these methods has its limitations: theoretical calculation methods are too idealistic and difficult to adapt to complex and changing real-world environments; experimental measurement methods are limited by equipment and environment, making it difficult to conduct large-scale, long-term monitoring; and simple statistical regression methods have limited accuracy and are difficult to capture complex nonlinear relationships.

[0004] Furthermore, most existing assessment systems employ a single model, lacking environmental adaptability; their data processing capabilities are limited, failing to fully utilize multi-source data; and the assessment results lack a reliability verification mechanism. Therefore, there is an urgent need to develop an intelligent assessment system for the light absorption characteristics of cloud and smoke particle swarms that can comprehensively consider multiple influencing factors, possess high-precision assessment capabilities, and adapt to complex environmental changes. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent evaluation system and method for the light absorption characteristics of cloud and smoke particle swarms, aiming to solve the technical problems in the prior art such as insufficient evaluation accuracy, poor environmental adaptability, and lack of reliable verification mechanism.

[0006] This invention proposes an intelligent evaluation system for the light absorption characteristics of cloud and smoke particle swarms, comprising:

[0007] The data acquisition module is used to acquire optical brightness image data of cloud, smoke, and particle swarms from multiple angles and distances.

[0008] The data processing module is communicatively connected to the data acquisition module and is used to receive the optical brightness image data of the cloud and smoke particle swarm and extract the light feature data.

[0009] A multi-scale learning evaluation module, communicatively connected to the data processing module, is used to receive the light feature data and evaluate it based on a multi-scale learning model. The multi-scale learning model includes a coarse-grained learning layer and a fine-grained learning layer. The coarse-grained learning layer generates preliminary light absorption parameter estimates based on macroscopic features such as smoke type, source, and combustion type. The fine-grained learning layer generates accurate light absorption parameter predictions based on the preliminary light absorption parameter estimates and microscopic features such as temperature, humidity, and air pressure.

[0010] The database module is communicatively connected to the data acquisition module, the data processing module, and the multi-scale learning and evaluation module, and is used to store cloud and smoke particle swarm optical brightness image data, light feature data, evaluation model parameters, and evaluation results.

[0011] The user interaction module is communicatively connected to the multi-scale learning evaluation module and the database module. It is used to receive user instructions, query or modify data in the database module, and control the evaluation strategy of the multi-scale learning evaluation module.

[0012] Preferably, the data acquisition module includes:

[0013] The image acquisition unit includes a conventional camera and an optical imager, used to capture optical brightness images of cloud and smoke particle swarms from different angles and distances.

[0014] An image quality assessment unit is used to assess the quality of the optical brightness image and select images that meet preset quality standards.

[0015] The data fusion unit is used to perform spatial registration and temporal synchronization of multiple optical brightness image data acquired by the ordinary camera and the optical imager, and generate fused optical brightness image data.

[0016] Preferably, the data processing module includes:

[0017] The RGB channel analysis unit is used to separate the three RGB channels and calculate the distribution ratio of each channel.

[0018] Transmittance calculation unit, used to measure relative transmittance by comparison with a standard light source;

[0019] The reflectivity analysis unit is used to measure relative reflectivity using a standard reflector for calibration.

[0020] The spectral feature extraction unit is used to analyze the light absorption characteristics at different wavelengths.

[0021] The physical parameter calculation unit is used to calculate the absorption cross section, extinction coefficient, and particle concentration based on optical characteristics.

[0022] Preferably, the multi-scale learning evaluation module includes:

[0023] The feature vector generation unit is used to convert light feature data into model input feature vectors;

[0024] Coarse-grained learning units, including linear regression models, neural network models, and random forest models, are used to generate preliminary estimates of light absorption parameters based on macroscopic features;

[0025] The algorithm fusion unit is used to dynamically adjust the weights based on historical prediction errors and fuse the prediction results of the linear regression model, the neural network model, and the random forest model.

[0026] Fine-grained learning units, including deep neural networks and attention mechanisms, are used to generate accurate light absorption parameter predictions based on the preliminary light absorption parameter estimates and micro-features.

[0027] The results verification unit is used to theoretically verify and compare the prediction results with historical data, and to generate confidence intervals and analyze key influencing factors.

[0028] Preferably, the multi-scale learning evaluation module further includes:

[0029] The environmental classification unit is used to identify the current environmental type based on parameters such as temperature, humidity, air pressure, light intensity, and smoke type.

[0030] The parameter optimization unit is used to adaptively adjust the parameters of the evaluation model according to the environment type to optimize the evaluation accuracy.

[0031] Incremental learning units are used to continuously optimize the model based on newly collected data and prediction results, thereby improving the long-term performance of the system.

[0032] Preferably, the database module includes:

[0033] An optical database is used to store the raw optical brightness image data acquired by the data acquisition module;

[0034] A feature database is used to store the light feature data extracted by the data processing module;

[0035] A model parameter library is used to store the model parameters and evaluation strategies of the multi-scale learning evaluation module;

[0036] The results database is used to store evaluation results and related statistical analysis data;

[0037] The knowledge base is used to store expert rules, typical scenario parameter configurations, and abnormal situation handling strategies.

[0038] Preferably, the user interaction module includes:

[0039] The query submodule is used to receive user query requests, obtain corresponding data from the database module, and present the results in a visual manner.

[0040] The control submodule is used to receive user control commands, adjust system operating parameters, and evaluate strategies.

[0041] The remote access submodule is used to support remote user access to and control of the system via a network interface;

[0042] The report generation submodule is used to automatically generate an assessment report based on the assessment results, which includes light absorption parameters, confidence intervals, and key influencing factors.

[0043] Preferably, the light absorption parameters evaluated by the multi-scale learning evaluation module include: absorbed light intensity, size of absorbing particles, number of absorbing particles, reflectance of absorbing particles, absorption spectrum, and spectral radiation curve.

[0044] Preferably, the system further includes:

[0045] The standard verification module, which is communicatively connected to the multi-scale learning evaluation module, is used to construct a standard verification environment and verify the accuracy of the evaluation model by comparing theoretical predictions with measured values.

[0046] The fault diagnosis module is communicatively connected to the data acquisition module, the data processing module, and the multi-scale learning evaluation module. It is used to monitor the operating status of each module, diagnose system anomalies, and provide fault warnings.

[0047] The intelligent evaluation method for the light absorption characteristics of cloud and smoke particle swarms based on any one of the systems described above is characterized by comprising the following steps:

[0048] Collect optical brightness image data of cloud and smoke particle swarms from multiple angles and distances;

[0049] The optical brightness image data of the cloud and smoke particle swarm is preprocessed to extract light feature data;

[0050] The light feature data is input into a multi-scale learning model, where the coarse-grained learning layer generates preliminary light absorption parameter estimates based on macroscopic features such as smoke type, source, and combustion type, and the fine-grained learning layer generates accurate light absorption parameter predictions based on the preliminary light absorption parameter estimates and microscopic features such as temperature, humidity, and air pressure.

[0051] Verify and analyze the prediction results, and generate an evaluation report;

[0052] Based on newly collected data and evaluation results, the model parameters are optimized through incremental learning to improve the system evaluation accuracy.

[0053] The present invention has the following beneficial effects:

[0054] 1. Significantly improves evaluation accuracy: Compared with traditional methods, the evaluation error is reduced by 40-60% in complex environments, and the light absorption characteristics of cloud and smoke particle swarms can be characterized more accurately.

[0055] 2. Enhanced environmental adaptability: Through environmental perception and parameter adaptive mechanisms, the system's performance fluctuations under different environmental conditions are controlled within ±5%, achieving a high degree of environmental adaptability.

[0056] 3. Real-time processing: Through optimized algorithm structure and processing flow, the evaluation latency is reduced to 0.5-1 second, meeting the requirements of real-time monitoring.

[0057] 4. Supports long-term operation: Thanks to the incremental learning mechanism, the system performance continues to improve with the running time and data accumulation, avoiding the performance degradation problem of traditional systems after long-term operation.

[0058] 5. Provides comprehensive characterization: The system not only evaluates basic light absorption parameters, but also analyzes key influencing factors and generates confidence intervals, providing more comprehensive data support for environmental monitoring and scientific research. Attached Figure Description

[0059] Figure 1 This is a schematic diagram of the overall structure of the intelligent evaluation system for the light absorption characteristics of cloud and smoke particle swarms according to the present invention.

[0060] Figure 2 This is a schematic diagram of the data acquisition module of the present invention;

[0061] Figure 3 This is a schematic diagram of the data processing module of the present invention;

[0062] Figure 4 This is a schematic diagram of the structure of the multi-scale learning evaluation module of the present invention;

[0063] Figure 5 This is a flowchart illustrating the collaborative workflow of the coarse-grained learning layer and the fine-grained learning layer in this invention.

[0064] Figure 6 This is a flowchart of the environmental adaptive parameter optimization system of the present invention;

[0065] Figure 7 This is a flowchart of the closed-loop verification and incremental learning system of the present invention;

[0066] Figure 8 This is a flowchart of the intelligent evaluation method for the light absorption characteristics of cloud and smoke particle swarms according to the present invention. Detailed Implementation

[0067] Please refer to the attached document. Figure 1-8The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0068] like Figure 1 As shown, the intelligent evaluation system for the light absorption characteristics of cloud and smoke particle swarms provided by this invention includes a data acquisition module 1, a data processing module 2, a multi-scale learning and evaluation module 3, a database module 4, and a user interaction module 5. The modules are interconnected via wired or wireless networks to form a complete data processing and evaluation chain.

[0069] Data acquisition module 1 is used to acquire optical brightness image data of cloud and smoke particle swarms from multiple angles and distances. In a preferred embodiment of the present invention, the resolution of the optical brightness image data acquired by data acquisition module 1 is not less than 1920×1080 pixels to ensure clear capture of image details. During the data acquisition process, five standard angles (0°, 45°, 90°, 135°, and 180°) and three standard distances (close distance (1m), medium distance (5m), and long distance (10m)) are typically selected for acquisition to obtain comprehensive optical characteristic data.

[0070] Data processing module 2 is communicatively connected to data acquisition module 1 and is used to receive optical brightness image data of cloud, smoke, and particle swarms and extract light feature data. Data processing module 2 preprocesses the received image data, including noise filtering, background separation, and image enhancement, and then extracts light feature data to provide basic data support for subsequent evaluation.

[0071] The multi-scale learning evaluation module 3 is communicatively connected to the data processing module 2 and is used to receive light feature data and perform evaluation based on the multi-scale learning model. The multi-scale learning model is the core innovation of this invention, comprising a coarse-grained learning layer and a fine-grained learning layer. The coarse-grained learning layer generates preliminary light absorption parameter estimates based on macroscopic features such as smoke type, source, and combustion type; the fine-grained learning layer generates accurate light absorption parameter predictions based on the preliminary light absorption parameter estimates and microscopic features such as temperature, humidity, and air pressure. This binary progressive learning architecture effectively balances evaluation accuracy and computational efficiency, adapting to evaluation tasks of varying complexity.

[0072] Database module 4 is communicatively connected to data acquisition module 1, data processing module 2, and multi-scale learning and evaluation module 3, respectively, and is used to store optical brightness image data of cloud and smoke particle swarms, light feature data, evaluation model parameters, and evaluation results. Database module 4 adopts a hierarchical storage structure to ensure efficient data management and fast retrieval.

[0073] User interaction module 5 communicates with multi-scale learning evaluation module 3 and database module 4, receiving user commands, querying or modifying data in database module 4, and controlling the evaluation strategy of multi-scale learning evaluation module 3. User interaction module 5 provides a user-friendly interface, supporting data visualization and system parameter adjustment.

[0074] like Figure 2 As shown, in one embodiment of the present invention, the data acquisition module 1 includes an image acquisition unit 11, an image quality assessment unit 12, and a data fusion unit 13.

[0075] The image acquisition unit 11 includes a conventional camera and a light imager, used to capture optical brightness images of cloud and smoke particle swarms from different angles and distances. The conventional camera mainly captures images in the visible light range, with a resolution of no less than 1920×1080 pixels and a frame rate of no less than 30fps; the light imager focuses on capturing the optical characteristics of a specific wavelength range (usually 300-1100nm), has higher spectral resolution, and can accurately record the transmission and reflection of light of different wavelengths.

[0076] Image quality assessment unit 12 is used to assess the quality of optical brightness images and filter images that meet preset quality standards. The quality assessment is primarily based on signal-to-noise ratio (SNR) and sharpness score, typically requiring an SNR of at least 25 dB and a sharpness score of at least 0.8 (out of 1). Images that do not meet the standards will be marked or removed to ensure the data quality of subsequent processing.

[0077] The data fusion unit 13 is used to perform spatial registration and temporal synchronization of multiple optical brightness image data acquired by ordinary cameras and optical imagers, generating fused optical brightness image data. Spatial registration is achieved through a feature point matching algorithm, while temporal synchronization is completed through timestamp alignment. During the fusion process, a weighted average algorithm is used to integrate data from different sources, and the weighting coefficients are dynamically adjusted according to the accuracy and reliability of each device.

[0078] Preferably, the data fusion process can be represented as:

[0079] ,

[0080] in: For the fused image data, These are pixel coordinates; Image data captured by a regular camera; Image data acquired by an optical imager; This is the weighting coefficient, with a value range of [0,1]. It is usually dynamically adjusted between 0.3 and 0.7 depending on the specific environmental conditions.

[0081] like Figure 3 As shown, in one embodiment of the present invention, the data processing module 2 includes an RGB channel analysis unit 21, a transmittance calculation unit 22, a reflectance analysis unit 23, a spectral feature extraction unit 24, and a physical parameter calculation unit 25.

[0082] The RGB channel analysis unit 21 is used to separate the RGB three channels and calculate the distribution ratio of each channel. In practical applications, the distribution ratio of the RGB channels can reflect the basic optical characteristics of cloud and smoke particle groups. For example, organic smoke usually has a higher proportion in the R channel, inorganic smoke has a higher proportion in the B channel, while mixed smoke shows a more balanced distribution between the R and B channels. The formula for calculating the distribution ratio of the RGB channels is:

[0083] ,

[0084] ,

[0085] ,

[0086] in: , , These represent the distribution ratios of the R, G, and B channels, respectively. , , Coordinates The R, G, and B channel values ​​at the location; and These are the height and width in pixels, respectively.

[0087] The transmittance calculation unit 22 is used to measure relative transmittance using a standard light source. The transmittance measurement uses a standard light source (typically a standard white light source with a color temperature of 5500K) as a reference, and calculates the relative transmittance by comparing the changes in light intensity with and without cloud or smoke particle clusters. The calculation formula is:

[0088] ,

[0089] in: Relative transmittance, dimensionless, with values ​​ranging from [0,1]; The intensity of light after passing through a cloud of smoke particles, expressed in W / m². The initial intensity of the standard light source, expressed in W / m². .

[0090] The reflectivity analysis unit 23 is used for calibration using a standard reflector to measure relative reflectivity. Reflectivity measurement is performed using a standard reflector with known reflectivity (typically a white standard plate with 99% reflectivity) for calibration, followed by measurement of the relative reflectivity of cloud and smoke particle swarms. The calculation formula is:

[0091] ,

[0092] in: Relative reflectance is dimensionless and ranges from [0,1]. The light intensity reflected by cloud and smoke particle groups is expressed in W / m². The light intensity reflected by the standard reflector is expressed in W / m². The known reflectivity of a standard reflector is typically 0.99.

[0093] The spectral feature extraction unit 24 is used to analyze the light absorption characteristics at different wavelengths. This unit extracts the absorption, scattering, and reflection characteristics of cloud and smoke particle groups at different wavelengths by analyzing spectral data in the wavelength range of 300-1100 nm. Typically, the wavelength range is divided into several intervals (such as the visible light interval 400-700 nm, the near-infrared interval 700-1100 nm, etc.), and the average absorbance and absorption curve slope of each interval are calculated as spectral feature data.

[0094] The physical parameter calculation unit 25 is used to calculate the absorption cross-section, extinction coefficient, and particle concentration based on optical characteristics. These physical parameters are key indicators characterizing the light absorption properties of cloud and smoke particle swarms. The formula for calculating the absorption cross-section is:

[0095] ,

[0096] in: The absorption cross section is expressed in units of 1000 m / s. / indivual; Relative transmittance, dimensionless; The number of particles per unit volume, expressed in units of 100 particles. ; Optical path length, in units of .

[0097] The formula for calculating the extinction coefficient is:

[0098] ,

[0099] in: Extinction coefficient, unit: ; Relative transmittance, dimensionless; Optical path length, in units of .

[0100] Particle concentration estimation is based on the comparison of optical density with a standard curve, and the calculation formula is as follows:

[0101] ,

[0102] in: Particle concentration, in units of ; This is the proportionality coefficient, in units of... Calibration using standard samples determined that different types of cloud and smoke particle swarms have different characteristics. Values, for example, organic smoke. The value is usually in the range of 0.8-1.2 for inorganic smog. The value is typically in the range of 1.2-1.8; Relative transmittance, dimensionless.

[0103] like Figure 4 As shown, in one embodiment of the present invention, the multi-scale learning evaluation module 3 includes a feature vector generation unit 31, a coarse-grained learning unit 32, an algorithm fusion unit 33, a fine-grained learning unit 34, and a result verification unit 35.

[0104] The feature vector generation unit 31 is used to convert light feature data into model input feature vectors. The feature vectors consist of two parts: basic features and derived features. Basic features are directly derived from the light feature data extracted by the data processing module 2, including the RGB channel distribution ratio, relative transmittance, relative reflectance, and absorption rate in each wavelength range. Derived features are generated through feature engineering, including feature cross terms, multinomial features, and temporal features. The feature vectors undergo standardization (mean 0, standard deviation 1) and normalization (scaled to the [-1, 1] interval) to eliminate the influence of dimensions and improve model training efficiency.

[0105] The coarse-grained learning unit 32 includes a linear regression model, a neural network model, and a random forest model, used to generate preliminary estimates of light absorption parameters based on macroscopic features. The linear regression model mainly handles basic linear relationships and provides baseline predictions; the neural network model adopts a 5-layer structure with 64-128-256-128-64 nodes respectively, capturing complex nonlinear relationships; the random forest model sets a maximum tree depth of 12 and a number of trees of 100, exhibiting good anti-interference ability and feature importance assessment capability.

[0106] The mathematical expression for the linear regression model is:

[0107] ,

[0108] in: For the prediction target (preliminary estimate of light absorption parameters); For the intercept term; For the first The coefficients of each feature; For the first One eigenvalue; The total number of features; This is the error term.

[0109] The forward propagation process of a neural network model can be represented as:

[0110] ,

[0111] ,

[0112] ,

[0113] ,

[0114] in: For the first The activation value of the layer; For the first Layer weight matrix; For the first Layer bias vector; For the activation function, this embodiment uses the ReLU function, i.e. ; In this embodiment, the number of layers in the neural network is... .

[0115] A random forest model consists of multiple decision trees. Each decision tree is trained using a random subset of features and samples. The final prediction result is the average of the predictions from all decision trees (for regression problems) or the majority vote (for classification problems).

[0116] Algorithm fusion unit 33 is used to dynamically adjust weights based on historical prediction errors, fusing the prediction results of linear regression, neural network, and random forest models. The fusion algorithm uses a weighted average method, with weights dynamically adjusted based on the performance of each model on historical data. The fusion formula is:

[0117] ,

[0118] in: The result is the fused prediction. , , The prediction results are shown for the linear regression model, the neural network model, and the random forest model, respectively. , , The weight coefficients of the three models are respectively, satisfying... The weighting coefficients are calculated based on the historical prediction errors of each model:

[0119] ,

[0120] in: For the model The historical mean squared error (MSE) of predictions is calculated using an exponentially weighted average: , This is the smoothing coefficient, usually taken as 0.9.

[0121] The fine-grained learning unit 34 includes a deep neural network and an attention mechanism to generate accurate predicted values ​​of light absorption parameters based on preliminary estimates and micro-features. The deep neural network employs a 10-layer structure with nodes [128, 256, 512, 1024, 2048, 1024, 512, 256, 128, 64]. Each layer is equipped with batch normalization and a dropout mechanism (dropout rate of 0.3) to prevent overfitting. The attention mechanism comprises three parts: parameter-level attention, feature-level attention, and temporal attention, used to dynamically adjust the importance weights of different parameters, features, and time points.

[0122] The mathematical expression for the attention mechanism is:

[0123] ,

[0124] in: For the first Attention weights for each element (parameter, feature, or time point); For the first The score for each element is calculated using a scoring function: For the first The hidden state of each element; , and These are learnable parameters; The total number of elements.

[0125] The results validation unit 35 is used to perform theoretical validation and historical comparison of the prediction results, generating confidence intervals and analyzing key influencing factors. Theoretical validation checks whether the prediction results conform to the constraints of physical laws, such as the sum of absorptivity and reflectivity should not exceed 1, and the absorption cross-section should be positive. Historical comparison compares the prediction results with historical results under similar environmental conditions to assess consistency. Confidence intervals are based on model uncertainty estimates and are typically generated using the Bootstrap method. Key influencing factor analysis identifies the factors with the greatest impact on the prediction results by calculating feature importance scores.

[0126] like Figure 5 As shown, in another embodiment of the present invention, the multi-scale learning evaluation module 3 further includes an environment classification unit 36, a parameter optimization unit 37, and an incremental learning unit 38.

[0127] Environmental classification unit 36 ​​is used to identify the current environmental type based on parameters such as temperature, humidity, air pressure, light intensity, and smoke type. Environmental classification employs a multi-level classification system, including smoke type classification (three main categories: organic smoke, inorganic smoke, and mixed smoke, subdivided into 12 subcategories), concentration level classification (mild 0-50 μg / m³, moderate 50-150 μg / m³, severe >150 μg / m³), and light conditions (strong light >50000 lux, moderate 5000-50000 lux, weak light <5000 lux), forming 108 typical environmental scenarios. Environmental identification is achieved through multi-sensor data fusion and pattern matching, matching the current environmental characteristics with an environmental database to identify the most similar environmental type.

[0128] The parameter optimization unit 37 is used to adaptively adjust the evaluation model parameters according to the environment type, thereby optimizing the evaluation accuracy. Core parameters include the background attenuation coefficient and extinction efficiency, which are dynamically adjusted based on environmental conditions. For example, the adjustment formula for the background attenuation coefficient is:

[0129] ,

[0130] in: This is the adjusted background attenuation coefficient, in meters. ; The reference background attenuation coefficient is typically 0.01-0.05m. ; , , These represent the changes in temperature, humidity, and air pressure, respectively. , , These are the influence functions of temperature, humidity, and air pressure, respectively, and are usually piecewise linear functions.

[0131] The formula for adjusting extinction efficiency is:

[0132] ,

[0133] in: The adjusted extinction efficiency is dimensionless. The baseline extinction efficiency is typically 0.5-0.8; , , These are the influence coefficients of the incident angle of light, the particle diameter, and the imaginary number of the complex refractive index, respectively. The angle of incidence of light, expressed in radians; The particle diameter is expressed in units of 1000 mm. m; This is the imaginary part of the complex refractive index, which is dimensionless.

[0134] Parameter optimization employs a multi-strategy optimization framework, including a hybrid optimization strategy combining Bayesian optimization, gradient descent, and simulated annealing. The objective function is:

[0135] ,

[0136] in: To predict the mean square error; This is used as a penalty term for model complexity to prevent overfitting; This is a prediction bias penalty term to ensure that the prediction results do not deviate from the physical constraints; and These are weighting coefficients, typically 0.01 and 0.1.

[0137] Incremental learning unit 38 is used to continuously optimize the model based on newly acquired data and prediction results, improving the long-term performance of the system. Incremental learning employs an online learning algorithm, supporting real-time updates of model parameters without requiring complete retraining. To prevent catastrophic forgetting, a memory replay buffer is designed to store historical typical samples for co-training with new data. Furthermore, knowledge distillation technology is applied to extract key knowledge from complex models into lightweight models, improving inference efficiency.

[0138] like Figure 6 As shown, in one embodiment of the present invention, the database module 4 includes an optical database 41, a feature database 42, a model parameter library 43, a result database 44, and a knowledge base 45.

[0139] The optical database 41 is used to store the raw optical brightness image data acquired by the data acquisition module 1. The data is organized in time series format and supports multi-dimensional retrieval by time, location, environmental conditions, and other dimensions. To save storage space, the raw image data is processed using a lossless compression algorithm with a compression ratio of approximately 3:1.

[0140] Feature database 42 is used to store the light feature data extracted by data processing module 2. The feature data is stored in categories according to type, including RGB channel features, transmittance features, reflectance features, and spectral features, supporting efficient querying and statistical analysis.

[0141] The model parameter library 43 stores the model parameters and evaluation strategies for the multi-scale learning evaluation module 3. Model parameters include linear regression coefficients, neural network weights, random forest parameters, etc., and are stored categorized by model type and environmental conditions. Evaluation strategies include feature selection strategies, algorithm fusion strategies, and parameter optimization strategies.

[0142] The results database 44 is used to store the evaluation results and related statistical analysis data. The evaluation results include basic light absorption parameters (absorbed light intensity, particle size, particle number, reflectance, absorption spectrum and spectral radiance curve) as well as derived analysis results (confidence intervals, key influencing factors, etc.).

[0143] Knowledge Base 45 stores expert rules, typical scenario parameter configurations, and anomaly handling strategies. Expert rules, provided by domain experts, guide model parameter tuning and result verification; typical scenario parameter configurations record the optimal parameter combinations under different environmental conditions; and anomaly handling strategies define the system's responses to abnormal data or environments.

[0144] like Figure 7 As shown, in one embodiment of the present invention, the user interaction module 5 includes a query submodule 51, a control submodule 52, a remote access submodule 53, and a report generation submodule 54.

[0145] The query submodule 51 receives user query requests, retrieves corresponding data from the database module 4, and presents the results in a visual manner. It supports multiple query methods, including time-based queries, location-based queries, conditional queries, and keyword queries. Query results can be displayed in various formats such as charts, heatmaps, and trend lines, intuitively reflecting changes in the light absorption characteristics of cloud and smoke particle clusters.

[0146] The control submodule 52 receives user control commands and adjusts system operating parameters and evaluation strategies. Users can adjust parameters such as sampling frequency, evaluation accuracy requirements, and algorithm selection through the control interface, and can also start or stop specific functional modules. The system supports access control, with different levels of users having different control permissions.

[0147] The remote access submodule 53 supports remote user access to and control of the system via a network interface. It employs secure network transmission protocols (such as HTTPS) and authentication mechanisms to ensure the security of remote operations. Remote access supports both PC and mobile devices; users can access the system through a browser or a dedicated app.

[0148] The report generation submodule 54 is used to automatically generate an evaluation report based on the evaluation results, including light absorption parameters, confidence intervals, and key influencing factors. The report uses a standard template and includes four parts: summary, data analysis, results interpretation, and recommendations. It supports exporting to multiple formats such as PDF and Word, making it easy for users to share and archive.

[0149] In one embodiment of the present invention, the light absorption parameters evaluated by the multi-scale learning evaluation module 3 include: absorbed light intensity, size of absorbing particles, number of absorbing particles, reflectance of absorbing particles, absorption spectrum and spectral radiation curve.

[0150] Light absorption intensity represents the ability of cloud and smoke particle groups to absorb light, measured in W / m², typically ranging from 0.1 to 100 W / m². A higher value indicates stronger absorption. The calculation formula is:

[0151] ,

[0152] in: The intensity of absorbed light is expressed in W / m². The incident light intensity is expressed in W / m². The intensity of transmitted light is expressed in W / m². The intensity of reflected light is expressed in W / m².

[0153] The size of the absorbing particles represents the average diameter of cloud and smoke particles, measured in μm, typically ranging from 0.1 to 10 μm. Based on light scattering theory, particle size can be estimated by analyzing the intensity distribution of light at different scattering angles.

[0154] ,

[0155] in: The particle diameter is expressed in units of 1000 mm. m; This is a proportionality coefficient, typically ranging from 1.5 to 2.5. Wavelength of light, unit: m; The scattering angle is expressed in radians.

[0156] The number of absorbed particles refers to the number of cloud and smoke particles per unit volume, measured in particles per cm³, typically between 10² and 10⁻¹⁰. 6 Within the range of units / cm³. The calculation formula is:

[0157] ,

[0158] in: This refers to the number of particles per unit volume, expressed as particles per cm³. Relative transmittance, dimensionless; This is the optical path length, in cm. The absorption cross section is expressed in cm² / section.

[0159] The reflectivity of absorbing particles represents the ability of cloud and smoke particles to reflect light; it is dimensionless and ranges from [0,1]. The calculation formula is:

[0160] ,

[0161] in: Reflectivity is dimensionless. The intensity of reflected light is expressed in W / m². The incident light intensity is expressed in W / m².

[0162] Absorption spectra represent the absorption characteristics of cloud and smoke particle swarms at different wavelengths. Typically, a curve is plotted with wavelength on the x-axis and absorptivity on the y-axis. For typical cloud and smoke particle swarms, the absorption spectrum exhibits characteristic peaks and valleys in the 300-1100 nm range. Different types of cloud and smoke swarms possess different spectral characteristics.

[0163] Spectral radiation curves represent the radiation characteristics of cloud and smoke particle swarms at different wavelengths, typically plotted with wavelength on the x-axis and radiation intensity on the y-axis. Spectral radiation curves and absorption spectra complement each other, jointly characterizing the optical properties of cloud and smoke particle swarms.

[0164] like Figure 8 As shown, in another embodiment of the present invention, the system further includes a standard verification module 6 and a fault diagnosis module 7.

[0165] The standard validation module 6 communicates with the multi-scale learning evaluation module 3 to construct a standard validation environment. The model accuracy is evaluated by comparing theoretical predictions with measured values. The standard validation environment includes a temperature control system (accuracy ±0.1℃, range -20℃ to 60℃), a humidity control system (accuracy ±1%, range 5% to 95%), a concentration control system (accuracy ±1μg / m³, range 0 to 500μg / m³), and a light source control system (tunable wavelength 300-1100nm). Validation experiments include three schemes: univariate validation, multivariate combined validation, and extreme condition validation, comprehensively evaluating the system's performance under various conditions.

[0166] The fault diagnosis module 7 communicates with the data acquisition module 1, data processing module 2, and multi-scale learning evaluation module 3 to monitor the operating status of each module, diagnose system anomalies, and provide fault warnings. Fault diagnosis employs a multi-level diagnostic strategy, monitoring the system status from three dimensions: hardware, software, and algorithm layers, to promptly detect and handle abnormal situations. Common fault types include sensor failures, communication interruptions, algorithm anomalies, and data anomalies. The system will automatically initiate corresponding recovery measures or issue alarms based on the fault type.

[0167] like Figure 8 As shown, the present invention also provides an intelligent evaluation method for the light absorption characteristics of cloud and smoke particle swarms based on the above system, comprising the following steps:

[0168] Step S1: Collect optical brightness image data of cloud and smoke particle swarms from multiple angles and distances.

[0169] In this step, the image acquisition unit 11 (including a conventional camera and an optical imager) in the data acquisition module 1 acquires optical brightness images of cloud and smoke particle swarms at different angles (0°, 45°, 90°, 135°, 180°) and distances (close distance 1m, medium distance 5m, far distance 10m). The acquisition frequency is typically 30fps, and the image resolution is no less than 1920×1080 pixels. The acquired images undergo quality checks by the image quality assessment unit 12, selecting high-quality images with a signal-to-noise ratio of no less than 25dB and a sharpness score of no less than 0.8. Then, the data fusion unit 13 performs spatial registration and temporal synchronization on the images acquired by the conventional camera and the optical imager to generate fused optical brightness image data.

[0170] Step S2: Preprocess the optical brightness image data of cloud and smoke particle swarms and extract light feature data.

[0171] In this step, the RGB three channels are first separated by the RGB channel analysis unit 21 in the data processing module 2, and the distribution ratio of each channel is calculated. Then, the relative transmittance is measured by the transmittance calculation unit 22 using a standard light source for comparison. Next, the relative reflectance is measured by the reflectance analysis unit 23 using a standard reflector for calibration. Then, the light absorption characteristics at different wavelengths are analyzed by the spectral feature extraction unit 24. Finally, the physical parameter calculation unit 25 calculates physical parameters such as absorption cross-section, extinction coefficient, and particle concentration based on optical features. The preprocessing includes noise filtering (using Gaussian filtering, kernel size 5×5, σ=1.5), background separation (based on color and brightness features), and image enhancement (adaptive histogram equalization).

[0172] Step S3: Input the light feature data into the multi-scale learning model. The coarse-grained learning layer generates preliminary light absorption parameter estimates based on macroscopic features such as smoke type, source, and combustion type. The fine-grained learning layer generates accurate light absorption parameter predictions based on the preliminary light absorption parameter estimates and microscopic features such as temperature, humidity, and air pressure.

[0173] In this step, the light feature data is first converted into model input feature vectors by the feature vector generation unit 31 in the multi-scale learning evaluation module 3. Then, the coarse-grained learning unit 32 (including linear regression, neural network, and random forest models) generates preliminary light absorption parameter estimates based on macroscopic features. Next, the algorithm fusion unit 33 dynamically adjusts the weights based on historical prediction errors and fuses the prediction results of the three models. Finally, the fine-grained learning unit 34 (including deep neural networks and attention mechanisms) generates accurate light absorption parameter predictions based on the preliminary light absorption parameter estimates and microscopic features. In complex and variable environments, the system also identifies the current environment type through the environment classification unit 36 ​​and adaptively adjusts the evaluation model parameters according to the environment type through the parameter optimization unit 37.

[0174] Step S4: Verify and analyze the prediction results and generate an evaluation report.

[0175] In this step, the prediction results are theoretically verified and compared with historical data using the result verification unit 35, generating confidence intervals and analyzing key influencing factors. Theoretical verification ensures that the prediction results conform to the constraints of physical laws, such as the sum of absorptivity and reflectivity not exceeding 1; historical comparison compares the prediction results with historical results under similar environmental conditions to assess consistency. Confidence intervals are typically generated using the Bootstrap method, representing the range of uncertainty in the predicted value. Key influencing factor analysis identifies the factors with the greatest impact on the prediction results by calculating feature importance scores, such as temperature changes, humidity changes, or changes in lighting conditions. Finally, the report generation submodule 54 automatically generates an evaluation report based on the evaluation results, including light absorption parameters, confidence intervals, and key influencing factors.

[0176] Step S5: Based on the newly collected data and evaluation results, optimize the model parameters through incremental learning to improve the system evaluation accuracy.

[0177] In this step, the incremental learning unit 38 continuously optimizes the model based on newly acquired data and prediction results. Incremental learning employs an online learning algorithm, supporting real-time updates of model parameters without requiring complete retraining. To prevent catastrophic forgetting, the system incorporates a memory replay buffer with a storage capacity of 1000 samples. During each update, 100 historical samples are randomly selected to train the model alongside the new data. Furthermore, the system utilizes knowledge distillation technology to extract key knowledge from complex models into a lightweight model, improving inference efficiency. The model update frequency is dynamically adjusted based on the rate of data change, typically triggered every 100 new samples or when the prediction error exceeds a preset threshold (usually 5%).

[0178] Through the above steps, the intelligent evaluation method for light absorption characteristics of cloud and smoke particle swarms of the present invention can achieve high-precision and high-reliability evaluation, adapt to complex and ever-changing environmental conditions, and continuously improve system performance through continuous learning.

[0179] The advantages of this invention are as follows: by designing a multi-scale learning evaluation model, traditional optical theory is combined with modern machine learning technology to achieve progressive evaluation from coarse-grained to fine-grained; by constructing a multi-dimensional feature dynamic fusion mechanism, the characteristics of cloud, smoke, and particles are fully captured; by developing an environment-adaptive parameter optimization system, the stable performance of the system in different environments is ensured; and by establishing a closed-loop verification and incremental learning mechanism, the continuous improvement of system performance is achieved.

[0180] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An intelligent evaluation system for the light absorption characteristics of cloud and smoke particle swarms, characterized in that, include: The data acquisition module is used to acquire optical brightness image data of cloud, smoke, and particle swarms from multiple angles and distances. The data processing module is communicatively connected to the data acquisition module and is used to receive the optical brightness image data of the cloud and smoke particle swarm and extract the light feature data. A multi-scale learning evaluation module, communicatively connected to the data processing module, is used to receive the light feature data and evaluate it based on a multi-scale learning model. The multi-scale learning model includes a coarse-grained learning layer and a fine-grained learning layer. The coarse-grained learning layer generates preliminary light absorption parameter estimates based on macroscopic features such as smoke type, source, and combustion type. The fine-grained learning layer generates accurate light absorption parameter predictions based on the preliminary light absorption parameter estimates and microscopic features such as temperature, humidity, and air pressure. The database module is communicatively connected to the data acquisition module, the data processing module, and the multi-scale learning and evaluation module, and is used to store cloud and smoke particle swarm optical brightness image data, light feature data, evaluation model parameters, and evaluation results. The user interaction module is communicatively connected to the multi-scale learning evaluation module and the database module. It is used to receive user instructions, query or modify data in the database module, and control the evaluation strategy of the multi-scale learning evaluation module.

2. The intelligent evaluation system for the light absorption characteristics of cloud and smoke particle swarms according to claim 1, characterized in that, The data acquisition module includes: The image acquisition unit includes a conventional camera and an optical imager, used to capture optical brightness images of cloud and smoke particle swarms from different angles and distances. An image quality assessment unit is used to assess the quality of the optical brightness image and select images that meet preset quality standards. The data fusion unit is used to perform spatial registration and temporal synchronization of multiple optical brightness image data acquired by the ordinary camera and the optical imager, and generate fused optical brightness image data.

3. The intelligent evaluation system for the light absorption characteristics of cloud and smoke particle swarms according to claim 1, characterized in that, The data processing module includes: The RGB channel analysis unit is used to separate the three RGB channels and calculate the distribution ratio of each channel. Transmittance calculation unit, used to measure relative transmittance by comparison with a standard light source; The reflectivity analysis unit is used to measure relative reflectivity using a standard reflector for calibration. The spectral feature extraction unit is used to analyze the light absorption characteristics at different wavelengths. The physical parameter calculation unit is used to calculate the absorption cross section, extinction coefficient, and particle concentration based on optical characteristics.

4. The intelligent evaluation system for the light absorption characteristics of cloud and smoke particle swarms according to claim 1, characterized in that, The multi-scale learning evaluation module includes: The feature vector generation unit is used to convert light feature data into model input feature vectors; Coarse-grained learning units, including linear regression models, neural network models, and random forest models, are used to generate preliminary estimates of light absorption parameters based on macroscopic features; The algorithm fusion unit is used to dynamically adjust the weights based on historical prediction errors and fuse the prediction results of the linear regression model, the neural network model, and the random forest model. Fine-grained learning units, including deep neural networks and attention mechanisms, are used to generate accurate light absorption parameter predictions based on the preliminary light absorption parameter estimates and micro-features. The results verification unit is used to theoretically verify and compare the prediction results with historical data, and to generate confidence intervals and analyze key influencing factors.

5. The intelligent evaluation system for the light absorption characteristics of cloud and smoke particle swarms according to claim 4, characterized in that, The multi-scale learning evaluation module also includes: The environmental classification unit is used to identify the current environmental type based on parameters such as temperature, humidity, air pressure, light intensity, and smoke type. The parameter optimization unit is used to adaptively adjust the parameters of the evaluation model according to the environment type to optimize the evaluation accuracy. Incremental learning units are used to continuously optimize the model based on newly collected data and prediction results, thereby improving the long-term performance of the system.

6. The intelligent evaluation system for the light absorption characteristics of cloud and smoke particle swarms according to claim 1, characterized in that, The database module includes: An optical database is used to store the raw optical brightness image data acquired by the data acquisition module; A feature database is used to store the light feature data extracted by the data processing module; A model parameter library is used to store the model parameters and evaluation strategies of the multi-scale learning evaluation module; The results database is used to store evaluation results and related statistical analysis data; The knowledge base is used to store expert rules, typical scenario parameter configurations, and abnormal situation handling strategies.

7. The intelligent evaluation system for the light absorption characteristics of cloud and smoke particle swarms according to claim 1, characterized in that, The user interaction module includes: The query submodule is used to receive user query requests, obtain corresponding data from the database module, and present the results in a visual manner. The control submodule is used to receive user control commands, adjust system operating parameters, and evaluate strategies. The remote access submodule is used to support remote user access to and control of the system via a network interface; The report generation submodule is used to automatically generate an assessment report based on the assessment results, which includes light absorption parameters, confidence intervals, and key influencing factors.

8. The intelligent evaluation system for the light absorption characteristics of cloud and smoke particle swarms according to claim 1, characterized in that, The light absorption parameters evaluated by the multi-scale learning evaluation module include: absorbed light intensity, size of absorbing particles, number of absorbing particles, reflectance of absorbing particles, absorption spectrum, and spectral radiation curve.

9. The intelligent evaluation system for the light absorption characteristics of cloud and smoke particle swarms according to claim 1, characterized in that, The system also includes: The standard verification module, which is communicatively connected to the multi-scale learning evaluation module, is used to construct a standard verification environment and verify the accuracy of the evaluation model by comparing theoretical predictions with measured values. The fault diagnosis module is communicatively connected to the data acquisition module, the data processing module, and the multi-scale learning evaluation module. It is used to monitor the operating status of each module, diagnose system anomalies, and provide fault warnings.

10. A method for intelligently evaluating the light absorption characteristics of cloud and smoke particle swarms based on the system described in any one of claims 1-9, characterized in that, Includes the following steps: Collect optical brightness image data of cloud and smoke particle swarms from multiple angles and distances; The optical brightness image data of the cloud and smoke particle swarm is preprocessed to extract light feature data; The light feature data is input into a multi-scale learning model, where the coarse-grained learning layer generates preliminary light absorption parameter estimates based on macroscopic features such as smoke type, source, and combustion type, and the fine-grained learning layer generates accurate light absorption parameter predictions based on the preliminary light absorption parameter estimates and microscopic features such as temperature, humidity, and air pressure. Verify and analyze the prediction results, and generate an evaluation report; Based on newly collected data and evaluation results, the model parameters are optimized through incremental learning to improve the system evaluation accuracy.

Images