Air-ground integrated mining area environmental monitoring method and system based on hyperspectral imaging

Abstract

The present invention provides an air-ground integrated mining area environment monitoring method and system based on hyperspectral imaging, which relates to the field of mining area hyperspectral imaging measurement technology, obtains hyperspectral image data and spectral data of the atmosphere, water body, soil and landform of the mining area and pre-processes them; interprets the pre-processed hyperspectral image data and spectral data of the atmosphere, water body, soil and landform respectively, and obtains hot spot feature vectors of the atmosphere, water body, soil and landform respectively; inputs the hot spot feature vectors of the atmosphere, water body, soil and landform into an abnormal monitoring model respectively, obtains environmental abnormality monitoring results, uses artificial intelligence to process data and predict and identify models of the atmosphere, water body, soil and landform of the mining area, uses customized transfer learning strategies to accelerate network training speed and improve model generalization; finally, uses on-site imaging spectral data to carry out mining area prediction, and realizes intelligent monitoring of the mining area environment.

Description

Technical Field

[0001] The present disclosure relates to the technical field of mine area hyperspectral imaging measurement, and in particular to an air-ground integrated mine area environment monitoring method and system based on hyperspectral imaging. Background Art

[0002] The statements in this section merely provide background information related to the present disclosure and do not necessarily constitute prior art.

[0003] Environmental monitoring in mining areas is extremely important. It is not only a line of defense to protect the ecological environment from damage caused by mining activities, but also an important cornerstone to ensure the sustainable development of the mining industry. By real-time monitoring of key indicators such as air quality, water quality, soil and geological stability, potential environmental risks can be discovered and responded to in a timely manner, ensuring the health and safety of surrounding residents and mining areas.

[0004] In-situ, real-time and large-scale environmental monitoring is of great significance. Mining area environmental monitoring includes: water quality status monitoring, soil environment monitoring, geological environment monitoring and atmospheric environment monitoring. Traditional mining area water quality monitoring, soil pollution monitoring and atmospheric environment monitoring include on-site manual sampling for laboratory analysis, mobile monitoring equipment or setting up automatic monitoring stations to achieve monitoring.

[0005] However, the traditional mining environment monitoring methods still have certain defects:

[0006] Traditional mining environmental monitoring relies on single-point sensors, which cannot fully reflect the environmental conditions of the mining area, resulting in deviations or omissions in the monitoring results; the monitoring cost is expensive, different monitoring objects require different monitoring equipment, different manpower and material resources, and the monitoring cost is relatively high; real-time performance cannot be ensured, data continuity is insufficient or limited to monitoring at fixed locations, making it difficult to detect and track mobile hazardous sources in real time. Summary of the invention

[0007] In order to solve the above problems, the present invention proposes an air-ground integrated mining area environment monitoring method and system based on hyperspectral imaging. Based on imaging hyperspectral technology, image spectral data of the atmosphere, dust, water, soil, and landforms in the mining area are obtained, and image information and spectral information are integrated to conduct all-round environmental monitoring and analysis of the mining area's air quality, surface water quality, soil heavy metals, geological structure, mining area lithology, and topography. The method and system can monitor the mining environment in situ, in real time, and over a large area.

[0008] According to some embodiments, the present disclosure adopts the following technical solutions:

[0009] The air-ground integrated mining area environmental monitoring method based on hyperspectral imaging includes:

[0010] Obtain and pre-process hyperspectral image data and spectral data of the atmosphere, water, soil and landforms in the mining area;

[0011] The pre-processed hyperspectral image data and spectral data of the atmosphere, water bodies, soil and landforms are interpreted respectively to obtain the hotspot feature vectors of the atmosphere, water bodies, soil and landforms;

[0012] The hot spot feature vectors of the atmosphere, water, soil and landform are respectively input into the anomaly monitoring model to obtain the environmental anomaly monitoring results;

[0013] Among them, the pre-processed hyperspectral image data and spectral data of the atmosphere are interpreted, including atmospheric concentration interpretation and dust concentration interpretation. The dust concentration interpretation process includes: based on the relationship between spectral data and dust fall, recursive feature elimination is used to screen the dominant bands and perform regression analysis on the dust fall, an inversion model of the dust fall in the mining area based on the reflectivity of the dominant band is established, based on the inversion model of the dust fall in the mining area, the dust in the air of the mining area is tracked, and a spatial distribution map of the dust concentration is generated, and the dust concentration feature vectors of different areas are displayed in the form of a heat map.

[0014] According to some embodiments, the present disclosure adopts the following technical solutions:

[0015] The air-ground integrated mining area environmental monitoring system based on hyperspectral imaging includes:

[0016] Data acquisition module, used to acquire and pre-process the hyperspectral image data and spectral data of the atmosphere, water, soil and landforms of the mining area;

[0017] The inversion module is used to interpret the pre-processed hyperspectral image data and spectral data of the atmosphere, water bodies, soil and landforms, and obtain the hotspot feature vectors of the atmosphere, water bodies, soil and landforms respectively;

[0018] The monitoring module is used to input the hot spot feature vectors of the atmosphere, water, soil and landform into the anomaly monitoring model to obtain the environmental anomaly monitoring results;

[0019] Among them, the pre-processed hyperspectral image data and spectral data of the atmosphere are interpreted, including atmospheric concentration interpretation and dust concentration interpretation. The dust concentration interpretation process includes: based on the relationship between spectral data and dust fall, recursive feature elimination is used to screen the dominant bands and perform regression analysis on the dust fall, an inversion model of the dust fall in the mining area based on the reflectivity of the dominant band is established, based on the inversion model of the dust fall in the mining area, the dust in the air of the mining area is tracked, and a spatial distribution map of the dust concentration is generated, and the dust concentration feature vectors of different areas are displayed in the form of a heat map.

[0020] According to some embodiments, the present disclosure adopts the following technical solutions:

[0021] A non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the air-ground integrated mining area environment monitoring method based on hyperspectral imaging is implemented.

[0022] According to some embodiments, the present disclosure adopts the following technical solutions:

[0023] An electronic device comprises: a processor, a memory and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device executes the air-ground integrated mining area environment monitoring method based on hyperspectral imaging.

[0024] Compared with the prior art, the present invention has the following beneficial effects:

[0025] The present invention discloses an air-ground integrated mining area environment monitoring method based on hyperspectral imaging. Through non-contact, in-situ measurement, a three-dimensional data volume can be obtained by shooting once. While obtaining two-dimensional spatial information of the atmosphere, water, soil, and landforms in the mining area, dozens to hundreds of continuous bands of spectral data can be collected. It has the advantages of rich data acquisition and unified graphs and spectra. Different from the shortcomings of traditional single-point measurement, the present invention can monitor changes in the mining area environment over a large range and quickly.

[0026] The present invention discloses an air-ground integrated mining area environment monitoring method based on hyperspectral imaging, which uses artificial intelligence to process data of the mining area's atmosphere, water, soil and landforms, and then uses a customized transfer learning strategy to load parameters pre-trained based on laboratory data onto a deep learning transfer model based on on-site imaging spectral data, thereby expanding the mining area imaging spectral data set, accelerating network training speed, and improving model generalization; finally, the on-site imaging spectral data is used to carry out mining area prediction, thereby realizing intelligent monitoring of the mining area environment.

[0027] The present invention discloses an air-ground integrated mining environment monitoring method based on hyperspectral imaging. The air-ground integrated monitoring combines the flexibility of drones and the stability of ground systems, can perform efficient and accurate monitoring under complex conditions, and can monitor changes in the mineral environment for a long time and over a large range. Through long-term and continuous monitoring, abnormal changes in the mining environment can be discovered in a timely manner, providing strong support for safe production and environmental protection in mining areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The accompanying drawings constituting a part of the present disclosure are used to provide a further understanding of the present disclosure. The illustrative embodiments of the present disclosure and their descriptions are used to explain the present disclosure and do not constitute an improper limitation on the present disclosure.

[0029] Figure 1 This is a flow chart of the air-ground integrated mining area environment monitoring method based on hyperspectral imaging according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0030] The present disclosure is further described below in conjunction with the accompanying drawings and embodiments.

[0031] It should be noted that the following detailed descriptions are all illustrative and are intended to provide further explanation of the present disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present disclosure belongs.

[0032] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present disclosure. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0033] Example 1

[0034] In one embodiment of the present disclosure, an air-ground integrated mining area environmental monitoring method based on hyperspectral imaging is provided, which takes into account that the atmosphere, water, soil, and landform elements of the mining area are interrelated in the mining area environment and jointly affect the stability of the ecosystem, forming a dynamic system. For example, atmospheric pollutants are deposited into the soil through precipitation, and then enter the water body through surface runoff. Water pollution affects the soil through groundwater infiltration. Changes in landforms aggravate the pollution of water bodies or the atmosphere. These interactions show nonlinear and cumulative effects in space and time, resulting in significant changes in the ecosystem of the mining area. However, traditional single-element monitoring cannot fully grasp the changing trend of the mining area environment. Therefore, the present disclosure uses the atmosphere, water, soil, and landform of the mining area to perform integrated environmental detection. The method includes:

[0035] Obtain and pre-process hyperspectral image data and spectral data of the atmosphere, water, soil and landforms in the mining area;

[0036] The pre-processed hyperspectral image data and spectral data of the atmosphere, water bodies, soil and landforms are interpreted respectively to obtain the hotspot feature vectors of the atmosphere, water bodies, soil and landforms;

[0037] The hot spot feature vectors of the atmosphere, water, soil and landform are respectively input into the anomaly monitoring model to obtain the environmental anomaly monitoring results;

[0038] Among them, the pre-processed hyperspectral image data and spectral data of the atmosphere are interpreted, including atmospheric concentration interpretation and dust concentration interpretation. The dust concentration interpretation process includes: based on the relationship between spectral data and dust fall, recursive feature elimination is used to screen the dominant bands and perform regression analysis on the dust fall, an inversion model of the dust fall in the mining area based on the reflectivity of the dominant band is established, based on the inversion model of the dust fall in the mining area, the dust in the air of the mining area is tracked, and a spatial distribution map of the dust concentration is generated, and the dust concentration feature vectors of different areas are displayed in the form of a heat map.

[0039] As an embodiment, the air-ground integrated mining area environmental monitoring method based on hyperspectral imaging disclosed in the present invention obtains hyperspectral image data and spectral data of the atmosphere, water body, soil, and landform geology of the mining area based on imaging hyperspectral technology, interprets the pre-processed hyperspectral image data and spectral data of the atmosphere, water body, soil, and landform respectively, and obtains the hotspot feature vectors of the atmosphere, water body, soil, and landform respectively; inputs the hotspot feature vectors of the atmosphere, water body, soil, and landform respectively into the anomaly monitoring model to obtain the environmental anomaly monitoring result, and the specific implementation process is as follows:

[0040] Step 1: Interpret the atmosphere of the mining area to obtain the atmospheric concentration hotspot characteristic map and dust hotspot characteristic map in the atmosphere, including:

[0041] Step 1: Obtain and pre-process the hyperspectral image data and spectral data of the atmosphere in the mining area, including:

[0042] Step 101: First, select an air-based form to obtain hyperspectral image data of the mining area atmosphere based on a remote sensing platform. These data should cover the required observation period, and perform sensor calibration regularly to ensure the continuity and consistency of the data;

[0043] Step 102: performing preprocessing steps such as geometric correction, radiation correction, denoising, and baseline correction on the hyperspectral image data and spectral data of the above-mentioned mining area atmosphere;

[0044] Specifically, geometric correction is performed to correct the geometric distortion of remote sensing images to ensure that data at different times are spatially aligned;

[0045] Radiation correction: In order to eliminate the influence of external factors such as atmosphere and sensor response on remote sensing data, atmospheric correction is used to remove the influence of atmospheric scattering and absorption, restore the true reflectivity of the ground object, and calibrate the sensor using a standard radiation calibration plate or a ground object with known reflectivity.

[0046] Remove high-frequency noise through spectral filtering, such as Savitzky-Golay filter;

[0047] Finally, the absorption peak is highlighted by removing the background signal or baseline drift.

[0048] Step 2: Interpret the pre-processed hyperspectral image data and spectral data of the atmosphere, use spectral interpretation technology to identify and analyze various components in the atmosphere, and identify pollutants, gas components, etc. in the atmosphere by comparing with the known atmospheric spectral feature library; the interpretation process includes atmospheric concentration interpretation and dust concentration interpretation.

[0049] Step 201: interpreting the atmospheric concentration in the atmosphere, selecting a specific wavelength range based on the known absorption characteristics of the gas, matching the measured spectrum with the gas absorption spectrum in the standard spectrum library, determining the gas type, and inverting to obtain the atmospheric concentration value;

[0050] Specifically, the specific matching method uses the correlation coefficient method to normalize the spectrum to be measured and the standard spectrum, and calculate the Pearson correlation coefficient between the two:

[0051]

[0052] in, and The measured spectrum and the standard spectrum are The value at wavelength, and are their respective average values;

[0053] Determine the most matching standard spectrum according to the magnitude of the above correlation coefficient, and determine the gas type with the largest coefficient as the final matching gas type;

[0054] Furthermore, based on the Lambert-Beer law, a quantitative relationship between spectral absorption and gas concentration is established, and an atmospheric inversion model is constructed. The processed hyperspectral data is input into the atmospheric inversion model to calculate its concentration and distribution parameters, and the concentration distribution map of the atmosphere is obtained, thereby evaluating the environmental quality of the atmosphere in the mining area.

[0055] Among them, the infrared absorption spectrum of the target gas obtained by the thermal infrared hyperspectrometer is The radiation intensity under and background intensity The absorbance can be calculated:

[0056]

[0057] Then, based on the Lambert-Beer law, the gas concentration can be obtained:

[0058]

[0059] in, is the absorbance at a specific wavelength, is the molar absorption coefficient of the gas, is the optical path length. It can be seen that there is a relationship between the spectral absorption of thermal infrared hyperspectral data and the gas concentration.

[0060] Furthermore, a model of the relationship between absorbance and gas concentration is constructed. The specific inversion model establishment method is as follows: using the existing gas sample data, selecting the wavelength sensitive to the target gas as the input feature, selecting the combination of any two bands in the hyperspectral data, and constructing the hyperspectral gas index through three combinations of difference a, normalized difference b and ratio c:

[0061]

[0062] Among them, i and j are two bands randomly selected from the original spectrum, R i and R j are the spectral absorption of band i and band j respectively.

[0063] Furthermore, a linear model was established with the difference a, normalized difference b and ratio c as independent variables and the measured concentration;

[0064] Among them, the wavelength selection method for the most sensitive gas is that according to the Lambert-Beer law, the molar absorption coefficient It reflects the absorption intensity of the gas at a certain wavelength, so the sensitive wavelength is selected by calculating the absorption coefficient at different wavelengths;

[0065] Furthermore, the pre-processed air-based gas hyperspectral data of the mining area is input into the trained long short-term memory network model, and the long short-term memory network model outputs the concentration value of the gas;

[0066] Furthermore, a period of time is detected and data is monitored using time series to observe changes in atmospheric concentrations;

[0067] Specifically, a long short-term memory network is used to capture long-term dependencies in time series, identify long-term trends in atmospheric concentrations, conduct time series analysis on concentration distribution graphs at multiple time points, draw a line graph of concentration changes over time, intuitively display concentration change trends, and observe the changing trends and patterns of concentrations of hydrogen sulfide, ammonia, nitrogen oxides, methane, carbon monoxide, and sulfur dioxide;

[0068] Furthermore, by comparing the concentration distribution diagrams at different time points, hot spots and abnormal changes in concentration can be identified;

[0069] Specifically, within a given time window, the concentration trend, including the mean and standard deviation, was calculated;

[0070] Among them, the mean is: ;

[0071] The standard deviation is: ;

[0072] The dynamic threshold method is used to identify hot spots and abnormal change areas. The dynamic threshold is dynamically adjusted according to the mean and standard deviation of the data. The formula for the set benchmark threshold is:

[0073]

[0074] in, k is a user-defined coefficient that controls the sensitivity of the threshold (usually k The value is 2 or 3). At a certain moment, is the average concentration, is the standard deviation;

[0075] Furthermore, outliers in concentration changes can be detected by time series analysis and setting dynamic thresholds. Specifically, in the concentration distribution maps at different time points, areas with persistently high concentrations can be identified. These areas are long-term pollution sources and the most severely affected areas. By comparing the concentration distribution maps at different time points, areas with sudden increases in concentration can be identified. These areas are caused by the emergence of new pollution sources and environmental conditions. , it is judged as abnormally high and marked as a hot spot or potential risk point. , it is judged as abnormally low;

[0076] Specifically, the spatial clustering method (Moran's I) was used to further identify the spatial clustering characteristics of these abnormal change areas;

[0077] The specific process of using Moran's I index for spatial clustering is as follows: first, define the spatial weight matrix W, and the elements of the weight matrix are Represents the spatial relationship between location i and location j; then calculates the concentration value of each geographic location (pixel) Deviation:

[0078]

[0079] in, is the global average of the concentration values ​​at all locations; calculate the global spatial autocorrelation index Moran's I , (N is the total number of space units, is the spatial weight of location i and j, and are the concentration deviations at positions i and j, respectively, For all weights of and), when >0 indicates a positive correlation, indicating that spatially similar values ​​tend to cluster (hot spots); <0 indicates a negative correlation, indicating that spatially different values ​​tend to cluster (cold spots); =0 indicates random distribution with no obvious spatial correlation; the next step is to use local spatial autocorrelation to identify abnormal areas at specific locations ( It's location i The local Moran's I value indicates the degree of aggregation between the location and its neighboring areas; and is the deviation value between position i and adjacent position j); further, based on the calculated local Moran's I value, a spatial clustering map can be drawn, hot spots and cold spots can be marked with different colors, and areas with abnormal changes can be identified;

[0080] The processed airborne hyperspectral image data is visualized in three dimensions to intuitively display the spatial distribution and concentration changes of various components in the atmosphere;

[0081] Specifically, the XY plane corresponds to the spatial distribution of the hyperspectral image, that is, the geographic coordinates, and the Z axis is represented by the concentration of different gas components. The concentration value is represented by color, high concentration is represented by warm colors, and low concentration is represented by cold colors, generating a three-dimensional point cloud; gas layer display, if multiple gases are displayed, different layers can be generated for each gas, and different symbols are used to distinguish them; the three-dimensional data set (X, Y, concentration) is constructed into a volume data structure, and the hyperspectral data is rendered using volume rendering techniques such as Ray-casting and Marching Cubes to display the three-dimensional distribution of gas concentration; isosurface extraction, isosurfaces are calculated for different concentration levels, and the distribution of specific concentrations is displayed through three-dimensional surfaces, thereby obtaining a hot spot feature map of gas concentration.

[0082] Step 202: interpreting the dust concentration in the atmosphere, including: based on the relationship between spectral data and dust fall, using recursive feature elimination to screen the dominant band and perform regression analysis on the dust fall, establishing an inversion model for the dust fall in the mining area based on the reflectivity of the dominant band, tracking the dust in the air of the mining area based on the inversion model for the dust fall in the mining area, generating a spatial distribution map of the dust concentration, and displaying the dust concentration feature vectors in different areas in the form of a heat map.

[0083] Specifically, a ground-based hyperspectral camera is first used to obtain hyperspectral data of air with known dust fall. The above-mentioned ground-based spectral test data is used to perform spectrum-dust fall correlation analysis. Recursive feature elimination is used to screen the dominant bands and perform regression analysis on the dust fall. An inversion model for the dust fall in the mining area based on the reflectivity of the dominant band is established.

[0084] Furthermore, the recursive feature elimination method is used to calculate the weights of all bands. The specific process is:

[0085] First, all the bands are used as input to train the regression model. Then, the feature weight of each band is calculated through the model. Then, the bands with the smallest weight are gradually deleted. The model is retrained on the remaining bands, and the weight of each band is continued to be calculated. The above process is repeated continuously, and the least important bands are deleted until the set number of features is reached. The bands with larger weights are retained as the dominant bands.

[0086] Specifically, assuming that the sensitive bands of hyperspectral data x The vector is , the inversion model is:

[0087]

[0088] in, is the inversion model, which can be a linear function or a model constructed by a machine learning algorithm, is the model error. If a linear regression model is used, the inversion model can be expressed as:

[0089]

[0090] in, is the weight coefficient, reflecting the contribution of different bands to dustfall. The vector of the sensitive band is expressed as .

[0091] Based on the above-mentioned inversion model of dust fall, the dust in the air of the mining area is tracked to generate a spatial distribution map of dust concentration, and the dust concentration in different areas is displayed in the form of a heat map.

[0092] Specifically, the dust concentration data obtained by inversion is subjected to spatial interpolation processing to generate a spatial distribution map of dust concentration, and a heat map is created to obtain a hot spot feature map of dust.

[0093] Step 2: Interpret the water bodies in the mining area and obtain the hotspot feature map of the water body data, including:

[0094] Step 1: Collect water samples and perform hyperspectral imaging on the collected water samples through ground-based hyperspectral sensors to obtain spectral data covering different types of pollutants (such as suspended particulate matter, heavy metal pollution, etc.), and use laboratory methods to measure water quality parameters (such as suspended particulate matter concentration, heavy metal concentration, etc.), and use them as labels in supervised learning and source domain model construction;

[0095] Step 2: Perform preprocessing such as radiation correction, geometric correction and denoising on the above-mentioned ground-based water body hyperspectral image;

[0096] Step 3: Use machine learning algorithms to find suspended particles such as mineral particles, oil droplets, bubbles, and heavy metals such as lead, mercury, and cadmium in water bodies, identify the most sensitive spectral bands for pollutant identification, and extract relevant characteristic bands;

[0097] Specifically, the method for finding the most sensitive spectral band is to use the pollutant concentration corresponding to the hyperspectral data machine as input and use the random forest model for training. The random forest model will calculate the importance score (such as information gain) of each band according to the number of split nodes, sort them according to the importance score, and select the band with a higher score as the feature band;

[0098] In the random forest model, each tree splits nodes by features (spectral bands). The quality of each split is usually measured by information gain or Gini coefficient. Information gain is obtained by calculating the information entropy before and after the split, and the formula is:

[0099]

[0100] The calculation formula of entropy is:

[0101]

[0102] Among them, p i is the probability of class i.

[0103] Furthermore, all decision trees are aggregated to calculate the total information gain of each band as a split node in all trees. According to the importance score of each band, the band with a higher score is selected as the most sensitive feature band;

[0104] The quantitative relationship between spectral features and water quality parameters is used to establish a monitoring model as a water quality inversion pre-training model for subsequent transfer learning;

[0105] Specifically, a deep neural network was used to establish the relationship between characteristic bands and water quality parameters (suspended particles, heavy metal pollution);

[0106] Among them, suspended particles will show significant performance in the blue-green band (400-550nm), which is used as a characteristic band;

[0107] The deep neural network includes an input layer, a hidden layer, and an output layer with characteristic bands as input to predict the target water quality parameters. The input layer receives spectral characteristic band data, and the spectral data is converted into input layer features. Each band is used as an input node. The hidden layer includes multiple fully connected hidden layers, each layer contains several neurons, extracting complex features in the spectral data, and the output layer has neurons that match the number of water quality parameters.

[0108] Furthermore, the spectral features and water quality parameters in the training set are used to train the model and optimize the network weights so that the model can accurately map the quantitative relationship between the spectral features and the water quality parameters;

[0109] Furthermore, the trained model is saved as a pre-trained model for subsequent transfer learning, and the model is tested in different water areas and water quality conditions to improve the generalization ability of the model;

[0110] The hyperspectral image information of the mining area water body is collected in an air-based form as the target domain dataset, and then preprocessed, feature band extracted (same as the source domain, using the feature band selected in the pre-trained model), transferred learning and model fine-tuning are performed;

[0111] Specifically, transfer learning is to load the pre-trained model and make preliminary predictions on the spectral data of the target domain, use a small amount of labeled data of the target domain (i.e., water quality parameters measured in the target area) to fine-tune the model and adjust the model weights to adapt it to the data distribution of the target domain;

[0112] The model after transfer learning is used to monitor new waters in real time, predict water quality parameters, regularly obtain new water quality parameters and spectral data, and continuously update the model to update the prediction accuracy.

[0113] Among them, the input of the ground-based hyperspectral water quality parameter inversion (source domain) pre-training model is:

[0114]

[0115] Among them, each Represents the reflectance of a spectral band in the source domain, and the hidden layer is: , the water quality parameters predicted by the output layer are:

[0116]

[0117] In the process of transfer learning, the structure of the pre-trained model remains unchanged, but the weights are fine-tuned using the target domain data: the input of the model is the feature bands of the target domain: , the pre-trained weights of the source domain is used as initialization, but during fine-tuning, the weights are updated to , so that the model better adapts to the data distribution of the target domain: ,

[0118] The water quality parameters predicted by the output layer after fine-tuning are: ;

[0119] Use the gradient descent method to minimize the loss function and update the model weights: ;

[0120] Thus, the hot spot characteristic map of suspended particle concentration and heavy metal concentration in the water body is obtained according to the water quality parameters predicted by the output layer;

[0121] Step 3: Interpret the soil in the mining area and obtain the hot spot feature map of the soil. The specific process includes:

[0122] Step 1: Select the foundation form to collect soil hyperspectral image information in the mining area, and use laboratory analysis methods to analyze the pollution content of heavy metals chromium (Cr), nickel (Ni), arsenic (As) and cadmium (Cd) in the soil;

[0123] Step 2: After the spectral data has been preprocessed in the same way as above, the characteristic sensitive bands of different heavy metals in the soil are found. Combined with the chemical properties of heavy metals, the spectral absorption characteristics in specific bands are analyzed, and a content inversion model is established based on the characteristic sensitive bands as a pre-inversion model for soil heavy metal content.

[0124] Specifically, based on the improved Mustang optimization algorithm, the feature sensitive bands are selected. First, the population is initialized, and then the fitness function is calculated:

[0125]

[0126] in, represents the classification accuracy of band subset x, Indicates the length of the band subset, and is a weight factor that adjusts the trade-off between accuracy and simplicity,

[0127] Then, the behavior of the wild horse group is used to update the position. The algorithm updates the individuals in the population through multiple iterations, gradually approaching the optimal band combination. After each iteration, the individuals update their positions according to the fitness value and retain the optimal band subset. When the preset number of iterations is reached or the fitness function converges, the algorithm terminates and outputs the best feature band subset.

[0128] The extreme learning machine is used to establish the inversion model of heavy metal content. The characteristic sensitive bands extracted above are used as input data, the number of hidden layer nodes is set, and the weights and biases of the hidden layer nodes are randomly initialized. Then, the input data is mapped to the hidden layer through the activation function, and the hidden layer output matrix is ​​calculated. Then, the weight of the output layer is solved by the least squares method to minimize the error between the predicted value and the true value. Finally, the performance of the model is evaluated by cross-validation and other methods, and the number of hidden layer nodes and other hyperparameters are adjusted.

[0129] Step 3: Select the air-based form to collect soil hyperspectral image information of the mining area over a large area;

[0130] Step 4: The airborne soil hyperspectral image is preprocessed, the characteristic bands are screened (selected according to the results of the ground-based research), and the heavy metal content is inverted;

[0131] Step 5: Reveal the spatial distribution characteristics of heavy metal elements, draw the spatial distribution map of heavy metals in the mining area, obtain the hot spot characteristic map of heavy metal concentrations, display the concentration distribution of heavy metal elements in the study area, and use the cumulative index method to evaluate the pollution status and potential ecological risks of the soil in the study area, and analyze the sources of heavy metal pollution.

[0132] Step 4: Interpret the topography of the mining area and obtain the hot spot feature map of the topography. The specific process includes:

[0133] Step 1: Select the air-based form to collect large-scale hyperspectral image information of the mining area's geological and geomorphic features, and analyze the spatial distribution and morphological characteristics of geological elements such as rock layers, faults, and folds in the mining area based on image features and spectral features;

[0134] Step 2: Preprocess and extract features of hyperspectral images;

[0135] Step 3: Extract the spectral information of rocks and minerals from the spectral dimension of the hyperspectral image, and extract the geological texture information from the image dimension, wherein the texture features of the image are calculated, including contrast, correlation, energy, and homogeneity, to reveal the texture information of the geological structure;

[0136] Step 4: Embed the spectral features and texture features into the input sequence of the Transformer respectively, learn the complex relationship between these features through the self-attention mechanism, and perform fusion processing at the output layer;

[0137] Specifically, the extracted spectral features are converted into a vector sequence suitable for inputting the Transformer model. Each spectral feature can be regarded as an element of an input sequence, and the entire spectral feature sequence is embedded to form a multi-dimensional input matrix.

[0138] Furthermore, feature embedding is performed to convert the extracted texture features into vector sequences in the same way. Texture features contain spatial relationship information, so different spatial regions (such as different image blocks) are embedded to form corresponding sequence inputs.

[0139] Furthermore, using the Transformer input, the embedded sequences of spectral features and texture features are concatenated or combined to form a complete input sequence. This input sequence will contain both spectral and texture information, and each element represents a feature vector of a time step. In order to make the Transformer model perceive the order information of the sequence, positional encoding can be added to the input sequence to learn the temporal and spatial information in the hyperspectral image.

[0140] Furthermore, the self-attention mechanism is integrated. For spectral and texture features, in order to learn the complex relationship between them and dynamically adjust the importance of features, at each layer of the Transformer, the self-attention mechanism calculates the relationship between each element in the input sequence and all other elements (i.e., the attention weight), and then performs the weighted summation to obtain a new feature representation.

[0141] At the same time, through the stacking of multiple Transformer layers, the self-attention mechanism can gradually fuse spectral features and texture features to obtain global feature representation. During the model training process, through the update of weights, the model will gradually learn the best combination of spectral and texture features.

[0142] Furthermore, the output layer is fused. The output features after multiple Transformer layers are usually reduced in dimension and fused through one or more linear layers to map the high-dimensional features to a target space.

[0143] Step 5: Construct a mining area landform monitoring model based on the reduced dimension and fused features;

[0144] Specifically, the output layer is used to predict the specific geological elements in the mining area, such as the spatial distribution and morphological characteristics of rock layers, faults and folds;

[0145] Step 6: The main output of the model is the distribution map of geological elements in the mining area, showing the spatial location and morphological characteristics of geological structures such as rock layers, faults, and folds;

[0146] Step 7: Use wavelet transform and spatial sparse modeling of deformation signals to assess changes in mining area geomorphology, such as fault activity and rock subsidence;

[0147] Specifically, wavelet transform decomposes time series data in both the time domain and the frequency domain, capturing the changing characteristics at different scales;

[0148] Among them, based on the sparse representation theory, the model identifies and reconstructs the surface changes that occur locally in the mining area, improving the spatial resolution of monitoring and the sensitivity to abnormal events;

[0149] Step 8: According to the distribution and morphology of geological elements and other geomorphological changes, the risk of geological disasters in the mining area, such as landslides and ground subsidence, can be evaluated; specifically, by estimating the confidence interval of the time series prediction results, the model can provide decision makers with more reliable predictions, and the confidence interval of deformation prediction can be evaluated by using methods such as Bayesian inference, and then the risk analysis of geological disasters can be carried out. In this way, the geomorphological hot spot feature map in the geomorphological data can be obtained, including rock layers, faults and fold information.

[0150] Step 5: Process the hot spot feature maps of the atmosphere, water, soil and landforms, extract feature vectors and input them into the anomaly monitoring model, and obtain the dynamic monitoring results of environmental anomalies through feature-level fusion and time series analysis;

[0151] Step 1: Use a multi-scale convolutional neural network to extract hotspot features of different spatial scales in the atmosphere, water bodies, soil, and landforms, capture the correlation between them, and use feature-level fusion technology to fuse the pollutant concentration in the atmosphere, the distribution of atmospheric gases, the distribution of heavy metals in the soil, the distribution of heavy metals in the water, and the topographic change characteristics of the landforms;

[0152] Specifically, convolution kernels (filters) of different scales are designed to extract features in different spaces. Each convolution layer captures spatial information of different scales through convolution kernels of different sizes. The 3*3 convolution kernel extracts local hotspot information and captures smaller features, such as fine gas distribution, diffusion trend of atmospheric particulate matter, and local pollutants in water bodies; the 5*5 convolution kernel captures medium-scale regional hotspot features, such as larger soil pollution distribution; the 7*7 convolution kernel captures global and cross-regional features, such as large-scale landform changes;

[0153] Furthermore, each convolution branch performs a convolution operation on the input hotspot feature vector to extract multi-scale features, capture the changes in the distribution of different target concentrations in different regions, the trend of local point pollution expanding to the global distribution, the local concentration points of pollution and the diffusion behavior of pollutants, and the local undulation and sedimentation of the terrain, and directly superimpose the convolution outputs of each scale to form a comprehensive feature vector, which retains the hotspot information at all scales;

[0154] Furthermore, the feature vectors of each medium are concatenated to form a unified feature representation, and the joint feature vector = [atmospheric feature vector, water feature vector, soil feature vector, landform feature vector];

[0155] Furthermore, the attention mechanism is introduced to automatically learn the relative importance of each medium in different time and space, so as to selectively strengthen the characteristics of certain media, complete the feature fusion strategy, and finally generate a comprehensive feature vector with multi-scale and multi-media information. This feature vector contains multi-scale spatial information of different media (atmosphere, water, soil, landform);

[0156] Step 2: Introduce multi-temporal inputs into the multi-scale convolutional neural network to track the evolution of pollution hotspots over time, identify the dynamic changes of pollution diffusion, and realize dynamic monitoring of the mining environment;

[0157] Specifically, for each time point, the pollution hotspot feature data of the atmosphere, water, soil, and landforms are collected. The data at each time point contains multiple channels (different pollutant concentration maps, distribution changes, etc.). These feature maps are stacked in chronological order and expanded through the time dimension to form multi-phase input. Through multi-phase input, the convolutional network can capture the changes of pollution hotspots over time, especially the diffusion trend of pollutants, the movement and expansion of local hotspots;

[0158] Step 3: Use the deep learning-based time series prediction model to predict the trend of pollutant temporal changes, thereby predicting the future pollution evolution trend;

[0159] Specifically, the time series data of pollutant concentrations in multiple time phases are input into the long short-term memory network deep learning model. The model extracts time series features through its hidden layer, uses memory and forgetting mechanisms to capture the long-term and short-term change patterns of pollutants, and predicts pollutant concentrations at the next time point or multiple time points in the future.

[0160] Example 2

[0161] In one embodiment of the present disclosure, a system for monitoring an air-ground integrated mining environment based on hyperspectral imaging is provided, comprising:

[0162] Data acquisition module, used to acquire and pre-process the hyperspectral image data and spectral data of the atmosphere, water, soil and landforms of the mining area;

[0163] The inversion module is used to interpret the pre-processed hyperspectral image data and spectral data of the atmosphere, water bodies, soil and landforms, and obtain the hotspot feature vectors of the atmosphere, water bodies, soil and landforms respectively;

[0164] The monitoring module is used to input the hot spot feature vectors of the atmosphere, water, soil and landform into the anomaly monitoring model to obtain the environmental anomaly monitoring results;

[0165] Among them, the pre-processed hyperspectral image data and spectral data of the atmosphere are interpreted, including atmospheric concentration interpretation and dust concentration interpretation. The dust concentration interpretation process includes: based on the relationship between spectral data and dust fall, recursive feature elimination is used to screen the dominant bands and perform regression analysis on the dust fall, an inversion model of the dust fall in the mining area based on the reflectivity of the dominant band is established, based on the inversion model of the dust fall in the mining area, the dust in the air of the mining area is tracked, and a spatial distribution map of the dust concentration is generated, and the dust concentration feature vectors of different areas are displayed in the form of a heat map.

[0166] Furthermore, the data acquisition module includes an air-based submodule and a ground-based submodule, wherein the air-based submodule includes a drone and an imaging hyperspectral camera. The drone is used to carry the imaging hyperspectral camera to collect large-area, high-altitude hyperspectral data, covering the mining area and its surrounding environment. The imaging hyperspectral camera installed on the drone can capture spectral data of the atmosphere, water bodies and landforms with high resolution for subsequent analysis;

[0167] The ground-based submodule includes a tripod, an imaging hyperspectral camera, and a halogen light source. The tripod is equipped with a gimbal to achieve relative movement with the observed object. The imaging hyperspectral camera is fixed above the gimbal to stably collect hyperspectral image information of the mining area. The imaging hyperspectral camera is the same as the air-based submodule, but is used for ground collection to capture hyperspectral data such as soil near the ground. The halogen light source can provide a stable light source to ensure that consistent spectral data can be obtained under different lighting conditions, which is very important especially at night or in low light conditions.

[0168] Furthermore, the inversion module includes a gas processing submodule, a water quality processing submodule, a soil processing submodule and a landform processing submodule, which are used to analyze the gas, water quality, soil pollutants and landforms in the mining area; the gas processing submodule analyzes the hyperspectral image data of the gas above the mining area, extracts the spatial distribution and concentration changes of polluting gases such as sulfides and nitrogen oxides, and detects the dynamic behavior of dust particles in the air. The hyperspectral image of the atmosphere above the mining area is preprocessed, the characteristic band is selected, the concentration inversion model is constructed, and the types and concentrations of gases and particles are output;

[0169] The water quality treatment submodule analyzes suspended particles in the water of the mining area, including mineral particles, oil droplets, bubbles, etc., and heavy metal pollution such as lead, mercury, and cadmium, based on transfer learning between the laboratory and the mining site. It is responsible for migrating the results of laboratory analysis to field data, and outputs the types and concentrations of water pollutants through preprocessing, characteristic band selection, inversion model construction, and output;

[0170] The soil processing submodule analyzes the heavy metal pollution of chromium (Cr), nickel (Ni), arsenic (As) and cadmium (Cd) in the soil based on the transfer learning between the laboratory and the mining site. It is responsible for migrating the soil pollutant results analyzed in the laboratory to the on-site soil data, and outputs the distribution of heavy metals in the soil based on preprocessing, characteristic band selection and inversion model;

[0171] The landform processing submodule analyzes the spatial distribution and morphological characteristics of geological elements such as rock layers, faults, and folds in the mining area based on the spectral information fusion of hyperspectral images. The spectrum and texture information are fused through the self-attention mechanism of Transformer to form the comprehensive characteristics of landform monitoring. Finally, based on the fused characteristics, a monitoring model is constructed to analyze the spatiotemporal distribution of geological elements.

[0172] Furthermore, the monitoring module includes a result display module and an anomaly delineation module. The result display module has three functions: spatial distribution observation, threshold adjustment, and time series viewing. It can observe the spatial distribution of gas concentration from different angles, and select slices on the X, Y, or Z plane to observe the concentration distribution on a specific plane. By adjusting the displayed concentration threshold, high-concentration areas can be highlighted or low-concentration areas can be ignored. The atmospheric composition distribution and water pollution changes at different times can be viewed by sliding time.

[0173] The anomaly delineation module can integrate the information of the gas processing submodule, water quality processing submodule, soil processing submodule and landform processing submodule of the data processing module, set thresholds according to environmental monitoring standards, and delineate abnormal information.

[0174] Example 3

[0175] In one embodiment of the present disclosure, a non-transitory computer-readable storage medium is provided, wherein the non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the air-ground integrated mining area environment monitoring method based on hyperspectral imaging is implemented.

[0176] Example 4

[0177] In one embodiment of the present disclosure, an electronic device is provided, including: a processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device executes the air-ground integrated mining area environment monitoring method based on hyperspectral imaging.

[0178] The present disclosure is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present disclosure. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0179] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0180] Although the above describes the specific implementation methods of the present disclosure in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present disclosure. Technical personnel in the relevant field should understand that on the basis of the technical solution of the present disclosure, various modifications or variations that can be made by those skilled in the art without creative work are still within the scope of protection of the present disclosure.

Claims

1. An air-ground integrated mining area environmental monitoring method based on hyperspectral imaging, characterized in that: include: Obtain and pre-process hyperspectral image data and spectral data of the atmosphere, water, soil and landforms in the mining area; The pre-processed hyperspectral image data and spectral data of the atmosphere, water bodies, soil and landforms are interpreted respectively to obtain the hotspot feature vectors of the atmosphere, water bodies, soil and landforms; The hotspot feature vectors of the atmosphere, water, soil and landform are input into the anomaly monitoring model to obtain the environmental anomaly monitoring results, including: Step 1: Use a multi-scale convolutional neural network to extract hotspot features of different spatial scales in the atmosphere, water bodies, soil, and landforms, capture the correlation between them, and use feature-level fusion technology to fuse the pollutant concentration in the atmosphere, the distribution of atmospheric gases, the distribution of heavy metals in the soil, the distribution of heavy metals in the water, and the topographic change characteristics of the landforms; Specifically, convolution kernels of different scales are designed to extract features in different spaces. Each convolution branch performs a convolution operation on the input hotspot feature vector to extract multi-scale features, capture the changes in the concentration distribution of different targets in different regions, the trend of local point pollution expanding to the global distribution, the diffusion behavior of local concentration points of pollution and pollutants, and the local undulation and sedimentation characteristics of the terrain. The convolution outputs of each scale are directly superimposed to form a comprehensive feature vector, which retains the hotspot information at all scales. Furthermore, the feature vectors of each medium are concatenated to form a unified feature representation, and the joint feature vector = [atmospheric feature vector, water feature vector, soil feature vector, landform feature vector]; Furthermore, the attention mechanism is introduced to automatically learn the relative importance of each medium in different time and space, so as to selectively strengthen the features of certain media, complete the feature fusion strategy, and finally generate a comprehensive feature vector with multi-scale and multi-media information; Step 2: Introduce multi-temporal inputs into the multi-scale convolutional neural network to track the evolution of pollution hotspots over time, identify the dynamic changes of pollution diffusion, and realize dynamic monitoring of the mining environment; Specifically, for each time point, the pollution hotspot feature data of the atmosphere, water, soil, and landforms are collected. The data at each time point contains multiple channels. These feature maps are stacked in chronological order and expanded through the time dimension to form multi-phase input. Through multi-phase input, the convolutional network can capture the changes of pollution hotspots over time, including the diffusion trend of pollutants, the movement and expansion of local hotspots; Step 3: The time series prediction model based on deep learning is used to predict the trend of the temporal changes of pollutants, so as to predict the future pollution evolution trend; specifically, the time series data of pollutant concentrations in multiple time phases are input into the long short-term memory network deep learning model. The model extracts temporal features through its hidden layer, and uses the memory and forgetting mechanism to capture the long-term and short-term change patterns of pollutants, and predict the pollutant concentration at the next time point or multiple time points in the future.

2. The air-ground integrated mining area environment monitoring method based on hyperspectral imaging according to claim 1 is characterized in that: The pre-processed hyperspectral image data and spectral data of the atmosphere are interpreted, including atmospheric concentration interpretation and dust concentration interpretation. The dust concentration interpretation process includes: based on the relationship between spectral data and dust fall, using recursive feature elimination to screen the dominant band and perform regression analysis on the dust fall, establishing an inversion model of the dust fall in the mining area based on the reflectivity of the dominant band, tracking the dust in the air of the mining area based on the inversion model of the dust fall in the mining area, generating a spatial distribution map of the dust concentration, and displaying the dust concentration feature vectors of different areas in the form of a heat map; The atmospheric concentration interpretation process includes: preprocessing the mining area atmospheric hyperspectral image data and spectral data with geometric correction, radiation correction, denoising and baseline correction; performing spectral interpretation, comparing the known atmospheric spectral feature library based on the correlation coefficient method, and identifying pollutants and gas components in the atmosphere; establishing a quantitative relationship between spectral absorption and gas concentration based on the Lambert-Beer law, and constructing an atmospheric inversion model, inputting the processed spectral data into the atmospheric inversion model, calculating the atmospheric concentration and distribution parameters, and obtaining the atmospheric concentration characteristic vector.

3. The air-ground integrated mining area environmental monitoring method based on hyperspectral imaging according to claim 2 is characterized in that: According to the infrared absorption spectrum of the gas components in the atmosphere, the absorbance is calculated by the radiation intensity and background intensity at each wavelength, and then the gas concentration is calculated based on the Lambert-Beer law to obtain the quantitative relationship between the spectral absorption and the gas concentration. The gas concentration is calculated based on the Lambert-Beer law as follows: in, is the absorbance at a specific wavelength, is the molar absorption coefficient of the gas, is the optical path length.

4. The air-ground integrated mining area environment monitoring method based on hyperspectral imaging according to claim 2 is characterized in that: Use recursive feature elimination to screen out advantageous bands, including: First, all the spectral bands of dust are taken as input to train the regression model. Then, the feature weight of each band is calculated through the model. Then, the bands with the smallest weight are gradually deleted. The model is retrained on the remaining bands and the weight of each band is continued to be calculated. The above process is repeated continuously, and the least important bands are deleted until the set number of features is reached, and the bands with larger weights are retained as the dominant bands.

5. The method for air-ground integrated mining area environmental monitoring based on hyperspectral imaging according to claim 2, characterized in that: An inversion model for dust fall in mining areas based on the reflectivity of the dominant band is established, including: The vector of the dominant band of the spectral data is , the inversion model is ,in is the inversion model, is a linear function or a model constructed by a machine learning algorithm, is the model error. If the linear regression model is used, the inversion model is expressed as: in, is the weight coefficient, reflecting the contribution of different bands to dust fall. It is the dominant band.

6. The air-ground integrated mining area environment monitoring method based on hyperspectral imaging according to claim 1 is characterized in that: Interpret the pre-processed water body hyperspectral image data and spectral data, including: The hyperspectral images of water bodies are subjected to radiation correction, geometric correction and denoising preprocessing, and a machine learning algorithm is used to find the most sensitive spectral bands for identifying mineral particles, oil droplets, bubble suspended particles, and heavy metal pollutants such as lead, mercury, and cadmium in water bodies, and extract relevant feature bands. Specifically, the random forest model is used to extract feature bands. The random forest model calculates the importance score of each band based on the number of split nodes, sorts them according to the importance scores, and selects bands with higher scores as feature bands.

7. The air-ground integrated mining area environment monitoring method based on hyperspectral imaging according to claim 1 is characterized in that: Interpretation of pre-processed soil hyperspectral image data and spectral data, including: After the spectral data of the soil is preprocessed, the characteristic sensitive bands are selected based on the improved Mustang optimization algorithm to find the characteristic sensitive bands of different heavy metals in the soil. Combined with the chemical properties of heavy metals, their spectral absorption characteristics in specific bands are analyzed. The extreme learning machine model is selected to establish a heavy metal content inversion model based on the characteristic sensitive bands as a preliminary inversion model for soil heavy metal content.

8. An air-ground integrated mining area environmental monitoring system based on hyperspectral imaging, which implements the air-ground integrated mining area environmental monitoring method based on hyperspectral imaging according to any one of claims 1 to 7, characterized in that: include: Data acquisition module, used to acquire and pre-process the hyperspectral image data and spectral data of the atmosphere, water, soil and landforms of the mining area; The inversion module is used to interpret the pre-processed hyperspectral image data and spectral data of the atmosphere, water bodies, soil and landforms, and obtain the hotspot feature vectors of the atmosphere, water bodies, soil and landforms respectively; The monitoring module is used to input the hot spot feature vectors of the atmosphere, water, soil and landform into the anomaly monitoring model to obtain the environmental anomaly monitoring results.

9. The air-ground integrated mining area environment monitoring system based on hyperspectral imaging according to claim 8, characterized in that: The pre-processed hyperspectral image data and spectral data of the atmosphere are interpreted, including atmospheric concentration interpretation and dust concentration interpretation. The dust concentration interpretation process includes: based on the relationship between spectral data and dust fall, recursive feature elimination is used to screen the dominant band and perform regression analysis on the dust fall, an inversion model of the dust fall in the mining area based on the reflectivity of the dominant band is established, based on the inversion model of the dust fall in the mining area, the dust in the air of the mining area is tracked, a spatial distribution map of the dust concentration is generated, and the dust concentration feature vectors of different areas are displayed in the form of a heat map.

10. A non-transitory computer-readable storage medium, characterized in that: The non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by the processor, the air-ground integrated mining area environment monitoring method based on hyperspectral imaging as described in any one of claims 1-7 is implemented.

11. An electronic device, characterized in that: include: A processor, a memory and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory so that the electronic device executes the air-ground integrated mining area environment monitoring method based on hyperspectral imaging as described in any one of claims 1-7.

Images