Atmospheric particulate matter identification feature recognition method and device based on scanning electron microscope technology

Abstract

The invention discloses an atmospheric particulate matter identification feature recognition method and device based on a scanning electron microscope technology, and the method comprises the steps: analyzing a sample of a set pollution source through a scanning electron microscope controlled by a computer, obtaining an image, component data and particle size parameters of particulate matters in the sample of the pollution source, and constructing a single particle data set of the sample of the pollution source; according to the residual network model and the extreme gradient boosting model, constructing a single particle emission source prediction model, and performing training; obtaining a probability value of a target single particle belonging to a set pollution source through the trained single particle emission source prediction model; and explaining the single particle emission source prediction model through an SHAP strategy, and taking the key features of the single particle adopted by the single particle emission source prediction model during decision making as the identification features of the single particle emitted by the set pollution source. According to the method, the information of the single particles discharged by different pollution sources can be fully excavated, and the identification characteristics of the particles discharged by the pollution sources are indicated.

Description

Technical Field

[0001] The present invention relates to the technical field of research on characteristics of atmospheric particulate matter emitted by pollution sources, and in particular to a method and device for identifying identification characteristics of atmospheric particulate matter based on scanning electron microscopy technology. Background Art

[0002] Atmospheric particulate matter pollution has a significant impact on climate, ecosystems and human health, and has attracted widespread attention from all walks of life in recent years. The main sources of atmospheric particulate matter are natural sources such as soil dust and anthropogenic sources such as biomass combustion and coal burning. In order to more accurately identify the sources of atmospheric particulate matter and quantify the source contribution, it is very important to study the characteristics of particulate matter emitted by pollution sources.

[0003] At present, most existing studies use offline chemical analysis methods, using filter membranes to collect particulate matter and analyze carbon components (OC, EC), water-soluble ions (such as SO42-, NO3-, Cl-, F-, etc.) and metal elements (Na, Mg, Al, Ca, Fe, etc.) in particulate matter. However, these methods analyze the entire sample, which may mask the subtle characteristics of individual particles emitted by different pollution sources.

[0004] Research on atmospheric particulate matter based on electron microscopy analysis has shown that particulate matter emitted from different pollution sources differs in single-particle morphology and chemical composition. However, there is currently no method for identifying the characteristics of particulate matter emitted from pollution sources based on single-particle morphology, particle size and chemical composition.

[0005] Therefore, how to invent a method for identifying the characteristic features of particulate matter emitted by air pollution sources, fully exploit the single particle information emitted by different pollution sources, and identify the characteristic features of particulate matter emitted by pollution sources has become an urgent problem to be solved. Summary of the invention

[0006] To this end, the present invention provides a method and device for identifying the signature features of atmospheric particulate matter based on scanning electron microscopy technology. The method can fully mine the single particle information emitted by different pollution sources, indicating the signature features of particulate matter emitted by pollution sources; it can also analyze the role of different features in the pollution source identification process, and can be used to further analyze the physical and chemical characteristics of particulate matter emitted by different atmospheric pollution sources and the source analysis of atmospheric particulate matter.

[0007] In order to achieve the above object, the present invention provides the following technical solution: a method for identifying the signature features of atmospheric particulate matter based on scanning electron microscopy technology, comprising:

[0008] By using a computer-controlled scanning electron microscope to analyze the collected samples of the set pollution source, images, composition data and particle size parameters of the particles in the pollution source samples are obtained, and a single particle data set of the pollution source samples is constructed;

[0009] The single particle data set is randomly divided into a first data set and a second data set according to a set ratio; the first data set is used for training a single particle emission source prediction model; and the second data set is used for verifying the single particle emission source prediction model;

[0010] A single particle emission source prediction model is constructed according to the residual network model and the extreme gradient boosting model; the single particle emission source prediction model is iteratively trained through the first data set; the prediction result of the single particle emission source prediction model is verified through the second data set to obtain the trained single particle emission source prediction model;

[0011] The image, composition data and particle size parameters of the target single particle are input into the trained single particle emission source prediction model for processing to obtain the probability value of the target single particle belonging to the set pollution source;

[0012] The single particle emission source prediction model is interpreted through the SHAP value strategy, and the key features of the target single particle used by the single particle emission source prediction model when making decisions are used as the identifying features of the target single particle emitted by the set pollution source.

[0013] As a preferred solution of the atmospheric particulate matter identification feature recognition method based on scanning electron microscopy technology, in the process of analyzing the collected samples of the set pollution sources by applying a computer-controlled scanning electron microscope, the set pollution sources include: soil dust, road dust, construction dust, coal smoke dust, biomass combustion dust and steel smelting dust.

[0014] As a preferred solution of the method for identifying the characteristics of atmospheric particulate matter based on scanning electron microscopy technology, in the process of inputting the image, composition data and particle size parameters of the target single particle into the trained single particle emission source prediction model for processing, and obtaining the probability value that the target single particle belongs to the set pollution source, the image features of the target single particle are extracted through the residual network model, and the source of the target single particle is preliminarily predicted based on the image features of the target single particle, so as to obtain the preliminary probability value that the target single particle belongs to the set pollution source.

[0015] As a preferred solution of the method for identifying the characteristics of atmospheric particulate matter based on scanning electron microscopy technology, in the process of inputting the image, composition data and particle size parameters of the target single particle into the trained single particle emission source prediction model for processing, and obtaining the probability value of the target single particle belonging to the set pollution source, the preliminary probability value of the target single particle belonging to the set pollution source is merged with the particle size parameters and composition data of the target single particle, and input into the extreme gradient boosting model, and the probability value of the target single particle belonging to the set pollution source is obtained through prediction by the extreme gradient boosting model.

[0016] As a preferred solution of the atmospheric particulate matter identification feature recognition method based on scanning electron microscopy technology, in the process of interpreting the single particle emission source prediction model through the SHAP strategy, the expression of the SHAP value is:

[0017]

[0018] In the formula, φ j is the SHAP value; K is the set of all input variables; S is the subset without variable j. j >0, it means that the variable has a positive impact on the prediction; if φ j <0, it means that the variable has a negative impact on the prediction.

[0019] The present invention also provides an atmospheric particle identification feature recognition device based on scanning electron microscopy technology, which adopts the atmospheric particle identification feature recognition method based on scanning electron microscopy technology described above, comprising:

[0020] A single particle data set acquisition module is used to analyze the collected samples of the set pollution source by using a computer-controlled scanning electron microscope to obtain the image, composition data and particle size parameters of the particles in the pollution source samples, and to construct a single particle data set of the pollution source samples;

[0021] A single particle data set division module, used to randomly divide the single particle data set into a first data set and a second data set according to a set ratio; the first data set is used for training a single particle emission source prediction model; the second data set is used for verifying the single particle emission source prediction model;

[0022] A single particle emission source prediction model construction and training module is used to construct a single particle emission source prediction model based on a residual network model and an extreme gradient boosting model; iteratively train the single particle emission source prediction model through the first data set; and verify the prediction result of the single particle emission source prediction model through the second data set to obtain the trained single particle emission source prediction model;

[0023] A single particle pollution source probability acquisition module is used to input the image, component data and particle size parameters of the target single particle into the trained single particle emission source prediction model for processing, and obtain the probability value of the target single particle belonging to the set pollution source;

[0024] The single particle identification feature acquisition module is used to interpret the single particle emission source prediction model through the SHAP strategy, and use the key features of the target single particle used by the single particle emission source prediction model when making decisions as the identification features of the target single particle emitted by the set pollution source.

[0025] As a preferred solution of the atmospheric particulate matter identification feature recognition device based on scanning electron microscopy technology, in the single particle data set acquisition module, in the process of analyzing the collected samples of the set pollution source by applying a computer-controlled scanning electron microscope, the set pollution source includes: soil dust, road dust, construction dust, coal smoke dust, biomass combustion dust and steel smelting dust.

[0026] As a preferred solution of the atmospheric particulate matter identification feature recognition device based on scanning electron microscopy technology, in the probability acquisition module of the pollution source to which the single particle belongs, in the process of inputting the image, composition data and particle size parameters of the target single particle into the trained single particle emission source prediction model for processing, and obtaining the probability value of the target single particle belonging to the set pollution source, the image features of the target single particle are extracted through the residual network model, and the source of the target single particle is preliminarily predicted based on the image features of the target single particle, so as to obtain the preliminary probability value of the target single particle belonging to the set pollution source.

[0027] As a preferred solution of the atmospheric particulate matter identification feature recognition device based on scanning electron microscopy technology, in the probability acquisition module of the pollution source to which the single particle belongs, the image, component data and particle size parameters of the target single particle are input into the trained single particle emission source prediction model for processing to obtain the probability value of the target single particle belonging to the set pollution source. In the process, the preliminary probability value of the target single particle belonging to the set pollution source is merged with the particle size parameters and component data of the target single particle, and input into the extreme gradient boosting model. The probability value of the target single particle belonging to the set pollution source is obtained through prediction by the extreme gradient boosting model.

[0028] As a preferred solution of the atmospheric particulate identification feature recognition device based on scanning electron microscopy technology, in the single particle identification feature acquisition module, in the process of interpreting the single particle emission source prediction model through the SHAP strategy, the expression of the SHAP value is:

[0029]

[0030] In the formula, φ j is the SHAP value; K is the set of all input variables; S is the subset without variable j. j >0, it means that the variable has a positive impact on the prediction; if φ j <0, it means that the variable has a negative impact on the prediction.

[0031] The present invention has the following advantages: by using a computer-controlled scanning electron microscope to analyze the collected samples of the set pollution source, the image, component data and particle size parameters of the particulate matter in the pollution source sample are obtained, and a single particle data set of the pollution source sample is constructed; the single particle data set is randomly divided into a first data set and a second data set according to a set ratio; the first data set is used for training the single particle emission source prediction model; the second data set is used for verifying the single particle emission source prediction model; a single particle emission source prediction model is constructed according to a residual network model and an extreme gradient boosting model; the single particle emission source prediction model is iteratively trained by the first data set; the prediction result of the single particle emission source prediction model is verified by the second data set to obtain the trained single particle emission source prediction model; the image, component data and particle size parameters of the target single particle are input into the trained single particle emission source prediction model for processing to obtain the probability value of the target single particle belonging to the set pollution source; the single particle emission source prediction model is interpreted by the SHAP strategy, and the key features of the target single particle used by the single particle emission source prediction model in decision-making are used as the identification features of the set pollution source emitting the target single particle. The present invention can fully mine the single particle information emitted by different pollution sources and identify the particle identification characteristics of the pollution sources; it can also analyze the role of different characteristics in the pollution source identification process, and can be used to further analyze the physical and chemical characteristics of particulate matter emitted by different atmospheric pollution sources and the source analysis of atmospheric particulate matter. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the implementation methods of the present invention or the technical solutions in the prior art, the drawings required for the implementation methods or the prior art descriptions are briefly introduced below. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other implementation drawings can be derived from the provided drawings without creative work.

[0033] The structures, proportions, sizes, etc. illustrated in this specification are only used to match the contents disclosed in the specification so as to facilitate understanding and reading by persons familiar with the technology. They are not used to limit the conditions under which the present invention can be implemented, and therefore have no substantial technical significance. Any structural modification, change in proportion or adjustment of size shall still fall within the scope of the technical contents disclosed in the present invention without affecting the effects and purposes that can be achieved by the present invention.

[0034] Figure 1 A schematic flow chart of a method for identifying the signature characteristics of atmospheric particulate matter based on scanning electron microscopy technology provided in Example 1 of the present invention;

[0035] Figure 2A schematic diagram of specific implementation steps of the method for identifying the signature features of atmospheric particulate matter based on scanning electron microscopy technology provided in Example 1 of the present invention;

[0036] Figure 3 A schematic diagram of the SHAP analysis results of the ResNet model in the atmospheric particulate matter identification feature recognition method based on scanning electron microscopy technology provided in Example 1 of the present invention (taking a typical particle for each type of source as an example);

[0037] Figure 4 A schematic diagram of the average absolute SHAP value ranking of the input variables of the XGBoost model in the atmospheric particulate matter identification feature recognition method based on scanning electron microscopy technology provided in Example 1 of the present invention;

[0038] Figure 5 A schematic diagram summarizing the SHAP values ​​of the top 20 important variables in biomass combustion dust in the atmospheric particulate matter identification feature recognition method based on scanning electron microscopy technology provided in Example 1 of the present invention;

[0039] Figure 6 A schematic diagram summarizing the SHAP values ​​of the top 20 important variables in coal smoke in the atmospheric particulate matter identification feature recognition method based on scanning electron microscopy technology provided in Example 1 of the present invention;

[0040] Figure 7 A schematic diagram summarizing the SHAP values ​​of the top 20 important variables in building dust in the atmospheric particulate matter identification feature recognition method based on scanning electron microscopy technology provided in Example 1 of the present invention;

[0041] Figure 8 A schematic diagram summarizing the SHAP values ​​of the top 20 important variables in road dust in the method for identifying the signature characteristics of atmospheric particulate matter based on scanning electron microscopy technology provided in Example 1 of the present invention;

[0042] Fig. 9 A schematic diagram summarizing the SHAP values ​​of the top 20 important variables in soil dust in the atmospheric particulate matter identification feature recognition method based on scanning electron microscopy technology provided in Example 1 of the present invention;

[0043] Fig.10 A schematic diagram summarizing the SHAP values ​​of the top 20 important variables in the steel smelting source in the atmospheric particulate matter identification feature recognition method based on scanning electron microscopy technology provided in Example 1 of the present invention;

[0044] Fig.11 A schematic diagram of the architecture of an atmospheric particulate matter identification feature recognition device based on scanning electron microscopy technology provided in Example 2 of the present invention. DETAILED DESCRIPTION

[0045] The following is a description of the implementation of the present invention by specific embodiments. People familiar with the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0046] Example 1

[0047] See also Figure 1 and Figure 2 Embodiment 1 of the present invention provides a method for identifying the characteristics of atmospheric particulate matter based on scanning electron microscopy technology, comprising:

[0048] S1. Analyze the collected samples of the set pollution source by using a computer-controlled scanning electron microscope to obtain the image, composition data and particle size parameters of the particles in the pollution source samples, and construct a single particle data set of the pollution source samples;

[0049] S2. randomly divide the single particle data set into a first data set and a second data set according to a set ratio; the first data set is used for training a single particle emission source prediction model; and the second data set is used for verifying the single particle emission source prediction model;

[0050] S3. Construct a single particle emission source prediction model according to the residual network model and the extreme gradient boosting model; iteratively train the single particle emission source prediction model through the first data set; verify the prediction result of the single particle emission source prediction model through the second data set to obtain the trained single particle emission source prediction model;

[0051] S4, inputting the image, component data and particle size parameters of the target single particle into the trained single particle emission source prediction model for processing, and obtaining the probability value of the target single particle belonging to the set pollution source;

[0052] S5. The single particle emission source prediction model is interpreted through the SHAP strategy, and the key features of the target single particle used by the single particle emission source prediction model when making decisions are used as the identifying features of the target single particle emitted by the set pollution source.

[0053] In this embodiment, in step S1, in the process of analyzing the collected samples of the set pollution sources by applying a computer-controlled scanning electron microscope, the set pollution sources include: soil dust, road dust, construction dust, coal smoke dust, biomass combustion dust and steel smelting dust.

[0054] Specifically, in this embodiment, six types of pollution source samples were collected, including soil dust, road dust, construction dust, coal smoke dust, biomass combustion dust, and steel smelting dust. The pollution source samples were analyzed using the scanning electron microscope energy spectrum analysis (SEM-EDS) method to build a typical pollution source single particle database. The computer-controlled scanning electron microscope energy spectrum analysis (CCSEM-EDS) method was used to analyze at least 10,000 particles of each type of source sample, of which at least 2,000 particles were analyzed for each sample.

[0055] In this embodiment, in step S1, it also includes obtaining images, component data and particle size parameters of particulate matter in the pollution source sample;

[0056] Specifically, the data collected by CCSEM-EDS include a microscopic picture of particles, particle size parameters (such as maximum diameter, minimum diameter, average diameter, perpendicular diameter and equivalent circle diameter, etc.) and element data (including carbon (C), oxygen (O), sodium (Na), magnesium (Mg), aluminum (Al), silicon (Si), phosphorus (P), sulfur (S), chlorine (Cl), potassium (K), calcium (Ca), zinc (Zn), barium (Ba), vanadium (V), chromium (Cr), cobalt (Co), selenium (Se), tin (Sn), titanium (Ti), manganese (Mn), nickel (Ni), copper (Cu) and lead (Pb), etc.). The maximum, minimum and average diameters refer to the maximum, minimum and average diameters of the particles calculated by multiple rotation measurements, the perpendicular diameter refers to the diameter of the particles perpendicular to the maximum diameter, and the equivalent circle diameter refers to the diameter of a circle with the same area as the particle.

[0057] In this embodiment, in step S2, the single particle data set is randomly divided into a first data set and a second data set according to a set ratio; the first data set is used for training the single particle emission source prediction model; the second data set is used for verifying the single particle emission source prediction model;

[0058] Specifically, the single-particle data sets in the source samples obtained by analysis are divided into two sets. 80% of the single-particle data in each type of pollution source are randomly selected as the first data set, which is mainly used for training the single-particle emission source prediction model; the remaining 20% ​​of the single-particle data in each type of pollution source is the second data set, which is used to verify the accuracy of the prediction results of the single-particle emission source prediction model.

[0059] In this embodiment, in step S3, a single particle emission source prediction model is constructed according to the residual network model and the extreme gradient boosting model; the single particle emission source prediction model is iteratively trained through the first data set; the prediction result of the single particle emission source prediction model is verified through the second data set to obtain the trained single particle emission source prediction model;

[0060] Specifically, the single particle data in the first data set are input into the single particle emission source prediction model in batches, the pollution source of the single particle is predicted by the single particle emission source prediction model, the predicted result is compared with the actual pollution source, and the single particle emission source prediction model is continuously optimized.

[0061] The pollution source of the single particles in the second data set is predicted using the trained single-particle emission source prediction model. The probability value of the single particle belonging to the set pollution source is weighted and summed up by source category based on the mass or quantity of the single particle to obtain the mass or quantity proportion of atmospheric particulate matter emitted by the set pollution source.

[0062] In this embodiment, in step S4, while inputting the image, component data and particle size parameters of the target single particle into the trained single particle emission source prediction model for processing, and obtaining the probability value that the target single particle belongs to the set pollution source, the image features of the target single particle are extracted through the residual network model, and the source of the target single particle is preliminarily predicted based on the image features of the target single particle, so as to obtain the preliminary probability value that the target single particle belongs to the set pollution source.

[0063] Specifically, data augmentation techniques are applied to enhance the diversity of the data set. Random rotation, flipping, and scaling of single-particle micrographs are used to reduce the impact of the variability of particle position and size on the prediction results. At the same time, the image brightness, contrast, saturation, and hue are randomly adjusted to reduce the differences in micrographs of different samples caused by differences in CCSEM-EDS analysis conditions. A ResNet model based on transfer learning is used to extract image features and predict the source of particles based on image features. The initial weight parameters of the model are obtained by training on the large ImageNet dataset. During the training process, the Adam optimizer is used, the learning rate is set to 1e-2, the step size is set to 7, the decay rate is set to 0.1, and the negative log-likelihood loss function is used to optimize the model performance. Finally, the probability value of each particle coming from different sources is recorded.

[0064] In this embodiment, in step S4, it also includes inputting the image, component data and particle size parameters of the target single particle into the trained single particle emission source prediction model for processing to obtain the probability value of the target single particle belonging to the set pollution source, merging the preliminary probability value of the target single particle belonging to the set pollution source with the particle size parameters and component data of the target single particle, and inputting them into the extreme gradient boosting model, and predicting through the extreme gradient boosting model to obtain the probability value of the target single particle belonging to the set pollution source.

[0065] Specifically, the preliminary probability value of the single particle belonging to the set pollution source predicted by the residual network model is combined with the particle size parameters and composition data of the single particle as the input of the extreme gradient boosting (XGBoost) model, and the particle source label is the dependent variable. The extreme gradient boosting model training is completed, and the probability of each particle belonging to a different pollution source calculated by the extreme gradient boosting model is output. In the optimization process of the extreme gradient boosting model, grid search is used to select the most appropriate hyperparameters, and five-fold cross validation is used to evaluate the performance of the XGBoost model.

[0066] In this embodiment, in step S5, the single particle emission source prediction model is interpreted by the SHAP strategy, and the key features of the target single particle used by the single particle emission source prediction model when making decisions are used as the identifying features of the target single particle emitted by the set pollution source.

[0067] Specifically, the interpretable machine learning strategy, SHAP strategy (SHapley AdditiveexPlanation Approach), is applied to explain the model. The SHAP strategy is based on the joint game theory to distribute the total gain to each participant. In short, the difference between using a variable (such as j) for model prediction and not using j for prediction is attributed to the marginal contribution of variable j. Considering the interaction effect between variables, the difference is calculated for each possible variable subset combination for each sample. For each prediction sample x i , generate predicted values With K variables, the explanatory model g can be described as a linear function of feature attribution:

[0068]

[0069] In the formula, φ j (g,x i ) is the SHAP value, which means that variable j has an input of x i The impact of model g on prediction; φ0(g,x) is the base value, which represents the expected value of the prediction when considering all possible combinations of input factors.

[0070] φ j For all possible combinations of variable subsets, φ j Weighted average of values:

[0071]

[0072] Where K is the set of all input variables; S is the subset without variable j. j >0, it means that the variable has a positive impact on the prediction; if φ j <0, it means that the variable has a negative impact on the prediction.

[0073] The SHAP value represents the importance and impact of each input feature in the dataset on the prediction of an individual instance. It is worth noting that the SHAP value of each input feature is not fixed, but varies according to the specific evaluation example. A positive SHAP value means that the input variable increases the output result, while a negative SHAP value means that the input variable decreases the output result. The absolute SHAP value (|SHAP|) indicates the importance of each factor in affecting the prediction value. Therefore, a higher positive SHAP value indicates a higher positive SHAP value in increasing the prediction value. , while lower negative SHAP values ​​indicate that is of higher importance.

[0074] In this embodiment, the single particle emission source prediction model is composed of the coupling of ResNet and XGBoost models, and the SHAP method is used to interpret the two parts respectively, and the key features of the image, particle size and chemical elements of the particles emitted by different pollution sources are mined respectively.

[0075] like Figure 3 The following are the SHAP analysis results of the ResNet model. Each of the six types of sources is illustrated by a typical particle, and the true source label is indicated in the first column. The importance of different pollution sources to the model prediction results decreases from the second column to the seventh column. The SHAP value reflects the importance of the pixel to the model's prediction of the corresponding source (the label above the small red and blue shaded image). The higher the SHAP value of a pixel, the greater the contribution of the model to classifying the particle as that specific source. In this embodiment, the image identification features of the particles are well stepped (especially Figure 1 The identification features are consistent with manual experience, such as the irregular shape of soil particles, the stacked spheres of coal particles, and the shiny rod-like structure of steelmaking particles.

[0076] like Figure 4 As shown in the figure, it is the average absolute SHAP value ranking (from large to small) of the input variables of the XGBoost model. RD , CD , CC , BB and St are the probability value variables of particulate matter corresponding to soil dust, road dust, construction dust, coal smoke, biomass combustion and steelmaking, respectively, predicted by the ResNet model based on microscopic images. max , D min and D avg are the maximum, minimum and average diameters of the particles obtained from multiple rotation measurements, D perpis the vertical diameter (the diameter of the particle perpendicular to the maximum diameter), and Dcirc is the equivalent circle diameter (the diameter of a circle with the same area as the particle). In the process of single particle emission source prediction, the image features play the most important role in identification. Ca, Al, Si and Cl are the four most critical chemical elements, and D circ The particle size is the most critical parameter.

[0077] The summary diagrams of the SHAP values ​​of the top 20 important variables in the six sources of biomass combustion dust, coal smoke dust, construction dust, road dust, soil dust and steel smelting are as follows: Figure 5-Figure 10 These key variables are the identification features (variables) of different pollution sources.

[0078] The present invention can understand the identification effect of different features (variables). For example, the higher the proportion of some elements in the particles, the more likely the particles will be identified as a specific source class, such as Ca mainly helps to identify construction ash, elements such as Al and Ca are essential for identifying soil particles, and P, Si and K are essential for identifying biomass burning sources. However, although some elements are abundant in certain sources, the increase in the content of single particles may reduce the possibility of the model predicting them as their respective sources. For example, Ca in road dust and coal smoke, and Al in construction dust, because they are more abundant in other pollution sources. On the contrary, elements with lower content in specific sources, such as S in soil particles, P in construction dust, K in coal combustion particles, and Ca in biomass combustion particles, also play an important identification role. Their scarcity in these sources means that higher element content reduces the possibility of particles coming from these specific sources.

[0079] In summary, the present invention uses a computer-controlled scanning electron microscope to analyze the collected samples of the set pollution source, obtains the image, component data and particle size parameters of the particulate matter in the pollution source sample, and constructs a single particle data set of the pollution source sample; randomly divides the single particle data set into a first data set and a second data set according to a set ratio; the first data set is used for training the single particle emission source prediction model; the second data set is used for verifying the single particle emission source prediction model; constructs a single particle emission source prediction model according to the residual network model and the extreme gradient boosting model; iteratively trains the single particle emission source prediction model through the first data set; verifies the prediction result of the single particle emission source prediction model through the second data set to obtain the trained single particle emission source prediction model; inputs the image, component data and particle size parameters of the target single particle into the trained single particle emission source prediction model for processing to obtain the probability value of the target single particle belonging to the set pollution source; interprets the single particle emission source prediction model through the SHAP strategy, and uses the key features of the target single particle used by the single particle emission source prediction model in decision-making as the identification features of the set pollution source emitting the target single particle. The present invention can fully mine the single particle information emitted by different pollution sources and identify the particle identification characteristics of the pollution sources; it can also analyze the role of different characteristics in the pollution source identification process, and can be used to further analyze the physical and chemical characteristics of particulate matter emitted by different atmospheric pollution sources and the source analysis of atmospheric particulate matter.

[0080] It should be noted that the method of the embodiment of the present disclosure can be performed by a single device, such as a computer or a server. The method of the present embodiment can also be applied in a distributed scenario and completed by multiple devices cooperating with each other. In the case of such a distributed scenario, one of the multiple devices can only perform one or more steps in the method of the embodiment of the present disclosure, and the multiple devices will interact with each other to complete the described method.

[0081] It should be noted that the above describes some embodiments of the present disclosure. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the above embodiments and still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0082] Example 2

[0083] See also Fig.11 Embodiment 2 of the present invention provides an atmospheric particulate matter identification feature recognition device based on scanning electron microscopy technology, comprising:

[0084] The single particle data set acquisition module 001 is used to analyze the collected samples of the set pollution source by using a scanning electron microscope controlled by a computer, obtain the image, composition data and particle size parameters of the particles in the pollution source samples, and construct a single particle data set of the pollution source samples;

[0085] The single particle data set division module 002 is used to randomly divide the single particle data set into a first data set and a second data set according to a set ratio; the first data set is used for training the single particle emission source prediction model; the second data set is used for verifying the single particle emission source prediction model;

[0086] The single particle emission source prediction model construction and training module 003 is used to construct a single particle emission source prediction model according to the residual network model and the extreme gradient boosting model; iteratively train the single particle emission source prediction model through the first data set; verify the prediction result of the single particle emission source prediction model through the second data set to obtain the trained single particle emission source prediction model;

[0087] The single particle pollution source probability acquisition module 004 is used to input the image, component data and particle size parameters of the target single particle into the trained single particle emission source prediction model for processing, and obtain the probability value of the target single particle belonging to the set pollution source;

[0088] The single particle identification feature acquisition module 005 is used to interpret the single particle emission source prediction model through the SHAP value strategy, and use the key features of the target single particle used by the single particle emission source prediction model when making decisions as the identification features of the target single particle emitted by the set pollution source.

[0089] In this embodiment, in the single particle data set acquisition module 001, in the process of analyzing the collected samples of the set pollution source by applying a computer-controlled scanning electron microscope, the set pollution source includes: soil dust, road dust, construction dust, coal smoke dust, biomass combustion dust and steel smelting dust.

[0090] In this embodiment, in the single particle belonging to pollution source probability acquisition module 004, while inputting the image, component data and particle size parameters of the target single particle into the trained single particle emission source prediction model for processing, and obtaining the probability value of the target single particle belonging to the set pollution source, the image features of the target single particle are extracted through the residual network model, and the source of the target single particle is preliminarily predicted based on the image features of the target single particle, so as to obtain the preliminary probability value of the target single particle belonging to the set pollution source.

[0091] In this embodiment, in the single particle belonging to pollution source probability acquisition module 004, while inputting the image, component data and particle size parameters of the target single particle into the trained single particle emission source prediction model for processing, in the process of obtaining the probability value of the target single particle belonging to the set pollution source, the preliminary probability value of the target single particle belonging to the set pollution source is merged with the particle size parameters and component data of the target single particle, and input into the extreme gradient boosting model, and the probability value of the target single particle belonging to the set pollution source is obtained through prediction by the extreme gradient boosting model.

[0092] In this embodiment, in the single particle identification feature acquisition module 005, in the process of interpreting the single particle emission source prediction model by using the SHAP value strategy, the expression of the SHAP value is:

[0093]

[0094] In the formula, φ j is the SHAP value; K is the set of all input variables; S is the subset without variable j. j >0, it means that the variable has a positive impact on the prediction; if φ j <0, it means that the variable has a negative impact on the prediction.

[0095] It should be noted that the information interaction, execution process, etc. between the modules / units of the above-mentioned system are based on the same concept as the method embodiment in Example 1 of the present application, and the technical effects they bring are the same as those of the method embodiment of the present application. For specific contents, please refer to the description in the method embodiment shown in the previous part of the present application, and will not be repeated here.

[0096] Example 3

[0097] Embodiment 3 of the present invention provides a non-transitory computer-readable storage medium, in which a program code for a method for identifying characteristics of atmospheric particulate matter based on scanning electron microscopy technology is stored. The program code includes instructions for executing the method for identifying characteristics of atmospheric particulate matter based on scanning electron microscopy technology of embodiment 1 or any possible implementation thereof.

[0098] The computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server or a data center that includes one or more available media. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)).

[0099] Example 4

[0100] Embodiment 4 of the present invention provides an electronic device, including: a memory and a processor;

[0101] The processor and the memory communicate with each other via a bus; the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to execute the atmospheric particulate identification feature recognition method based on scanning electron microscopy technology of Example 1 or any possible implementation thereof.

[0102] Specifically, the processor can be implemented by hardware or by software. When implemented by hardware, the processor can be a logic circuit, an integrated circuit, etc.; when implemented by software, the processor can be a general-purpose processor implemented by reading software codes stored in a memory. The memory can be integrated in the processor or can be located outside the processor and exist independently.

[0103] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions can be transmitted from a website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center.

[0104] Obviously, those skilled in the art should understand that the above modules or steps of the present invention can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, and optionally, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order than here, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0105] Although the present invention has been described in detail above by general description and specific embodiments, it is obvious to those skilled in the art that some modifications or improvements can be made to the present invention. Therefore, these modifications or improvements made without departing from the spirit of the present invention all belong to the scope of protection claimed by the present invention.

Claims

1. A method for identifying the signature features of atmospheric particles based on scanning electron microscopy technology, characterized in that: include: By using a computer-controlled scanning electron microscope to analyze the collected samples of the set pollution source, images, composition data and particle size parameters of the particles in the pollution source samples are obtained, and a single particle data set of the pollution source samples is constructed; The single particle data set is randomly divided into a first data set and a second data set according to a set ratio; the first data set is used for training a single particle emission source prediction model; and the second data set is used for verifying the single particle emission source prediction model; A single particle emission source prediction model is constructed according to the residual network model and the extreme gradient boosting model; the single particle emission source prediction model is iteratively trained through the first data set; the prediction result of the single particle emission source prediction model is verified through the second data set to obtain the trained single particle emission source prediction model; The image, composition data and particle size parameters of the target single particle are input into the trained single particle emission source prediction model for processing to obtain the probability value of the target single particle belonging to the set pollution source; The single particle emission source prediction model is interpreted through the SHAP strategy, and the key features of the target single particle used by the single particle emission source prediction model when making decisions are used as the identifying features of the target single particle emitted by the set pollution source.

2. The method for identifying the signature features of atmospheric particles based on scanning electron microscopy technology according to claim 1 is characterized in that: In the process of analyzing the collected samples of the set pollution sources by applying a computer-controlled scanning electron microscope, the set pollution sources include: soil dust, road dust, construction dust, coal smoke dust, biomass combustion dust and steel smelting dust.

3. The method for identifying the signature features of atmospheric particles based on scanning electron microscopy technology according to claim 2 is characterized in that: In the process of inputting the image, component data and particle size parameters of the target single particle into the trained single particle emission source prediction model for processing to obtain the probability value of the target single particle belonging to the set pollution source, the image features of the target single particle are extracted through the residual network model, and the source of the target single particle is preliminarily predicted based on the image features of the target single particle to obtain the preliminary probability value of the target single particle belonging to the set pollution source.

4. The method for identifying the signature features of atmospheric particles based on scanning electron microscopy technology according to claim 3 is characterized in that: In the process of inputting the image, component data and particle size parameters of the target single particle into the trained single particle emission source prediction model for processing to obtain the probability value of the target single particle belonging to the set pollution source, the preliminary probability value of the target single particle belonging to the set pollution source is merged with the particle size parameters and component data of the target single particle, and inputted into the extreme gradient boosting model, and the probability value of the target single particle belonging to the set pollution source is obtained through prediction by the extreme gradient boosting model.

5. The method for identifying the signature features of atmospheric particles based on scanning electron microscopy technology according to claim 4 is characterized in that: In the process of interpreting the single particle emission source prediction model through the SHAP strategy, the expression of the SHAP value is: In the formula, φ j is the SHAP value; K is the set of all input variables; S is the subset without variable j; if φ j >0, it means that the variable has a positive impact on the prediction; if φ j <0, it means that the variable has a negative impact on the prediction.

6. The atmospheric particle identification feature recognition device based on scanning electron microscopy technology adopts the atmospheric particle identification feature recognition method based on scanning electron microscopy technology described in claims 1-5 above, characterized in that: include: A single particle data set acquisition module is used to analyze the collected samples of the set pollution source by using a computer-controlled scanning electron microscope to obtain the image, composition data and particle size parameters of the particles in the pollution source samples, and to construct a single particle data set of the pollution source samples; A single particle data set division module, used to randomly divide the single particle data set into a first data set and a second data set according to a set ratio; the first data set is used for training a single particle emission source prediction model; the second data set is used for verifying the single particle emission source prediction model; A single particle emission source prediction model construction and training module is used to construct a single particle emission source prediction model based on a residual network model and an extreme gradient boosting model; iteratively train the single particle emission source prediction model through the first data set; and verify the prediction result of the single particle emission source prediction model through the second data set to obtain the trained single particle emission source prediction model; A single particle pollution source probability acquisition module is used to input the image, component data and particle size parameters of the target single particle into the trained single particle emission source prediction model for processing, and obtain the probability value of the target single particle belonging to the set pollution source; The single particle identification feature acquisition module is used to interpret the single particle emission source prediction model through the SHAP value strategy, and use the key features of the target single particle used by the single particle emission source prediction model when making decisions as the identification features of the target single particle emitted by the set pollution source.

7. The atmospheric particulate matter identification feature recognition device based on scanning electron microscope technology according to claim 6 is characterized in that: In the single particle data set acquisition module, in the process of analyzing the collected samples of the set pollution sources by applying a computer-controlled scanning electron microscope, the set pollution sources include: soil dust, road dust, construction dust, coal smoke dust, biomass combustion dust and steel smelting dust.

8. The atmospheric particulate matter identification feature recognition device based on scanning electron microscope technology according to claim 7 is characterized in that: In the single particle pollution source probability acquisition module, the image, component data and particle size parameters of the target single particle are input into the trained single particle emission source prediction model for processing to obtain the probability value of the target single particle belonging to the set pollution source. The image features of the target single particle are extracted through the residual network model, and the source of the target single particle is preliminarily predicted based on the image features of the target single particle to obtain the preliminary probability value of the target single particle belonging to the set pollution source.

9. The atmospheric particulate matter identification feature recognition device based on scanning electron microscope technology according to claim 8 is characterized in that: In the single particle belonging pollution source probability acquisition module, the image, component data and particle size parameters of the target single particle are input into the trained single particle emission source prediction model for processing to obtain the probability value of the target single particle belonging to the set pollution source. In the process, the preliminary probability value of the target single particle belonging to the set pollution source is merged with the particle size parameters and component data of the target single particle, and input into the extreme gradient boosting model. The probability value of the target single particle belonging to the set pollution source is obtained through prediction by the extreme gradient boosting model.

10. The atmospheric particulate matter identification feature recognition device based on scanning electron microscope technology according to claim 9 is characterized in that: In the single particle identification feature acquisition module, in the process of interpreting the single particle emission source prediction model through the SHAP strategy, the expression of the SHAP value is: In the formula, φ j is the SHAP value; K is the set of all input variables; S is the subset without variable j; If φ j >0, it means that the variable has a positive impact on the prediction; if φ j <0, it means that the variable has a negative impact on the prediction.

Images