Mineral product law enforcement situation intelligent perception method based on multi-source remote sensing data

Abstract

The invention relates to the technical field of intelligent monitoring of illegal mining of mineral products, and discloses a mineral product law enforcement situation intelligent perception method based on multi-source remote sensing data, which comprises the following steps: S1, mining and extracting key driving factors of mineral product resource intelligent situation perception; s2, establishing a historical mineral product law enforcement record data set; s3, mineral product law enforcement situation prior probability modeling: constructing mineral product law enforcement situation prior probability modeling based on an artificial neural network ANN; s4, training and testing a mining area law enforcement situation awareness model; s5, early trend early warning based on prior probability updating; according to the invention, through application of the intelligent identification technology, machine learning and deep learning algorithms are utilized, the identification efficiency and accuracy of illegal mining behaviors of mineral products are improved, and real-time monitoring is realized. The problems that in the prior art, manual operation and experience judgment are relied on, and the intelligent degree is low are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent monitoring of illegal mining of minerals, and in particular to an intelligent perception method of mineral law enforcement situation based on multi-source remote sensing data. Background Art

[0002] Illegal mining of mineral resources not only damages the ecological environment, but also seriously affects the national economy and social stability. Traditional mining law enforcement mainly relies on manual inspections and reports, which have problems such as limited coverage, low efficiency and strong lag. With the development of remote sensing technology and big data analysis technology, using remote sensing data to monitor mineral mining activities has become a new research direction.

[0003] At present, there are mainly the following methods for mining law enforcement monitoring based on remote sensing technology: (1) Optical remote sensing monitoring: Using satellite optical images to monitor the surface, and identifying mining activities in mining areas through image analysis. However, this method is greatly affected by clouds and climate conditions, and the time to obtain high-quality images is limited. (2) Radar remote sensing monitoring: Radar remote sensing is not affected by weather and can monitor surface activities around the clock. Through synthetic aperture radar image analysis, surface deformation and mining activities can be identified. However, radar data processing is complex and requires high computing resources. (3) Multispectral and hyperspectral remote sensing monitoring: Using multispectral and hyperspectral data, the mineral types and mining activities in mining areas can be more accurately identified. However, this method requires complex image processing and spectral analysis technology, and has high requirements for acquisition equipment and professional knowledge. (4) UAV remote sensing monitoring: Using drones equipped with high-resolution cameras to conduct low-altitude monitoring of mining areas. Although this method is highly flexible, its coverage is limited.

[0004] Although existing technologies have improved the monitoring capabilities of mineral law enforcement to a certain extent, there are still many shortcomings. First, the processing of image data is complex. Whether it is optical images or radar data, the processing of multispectral or hyperspectral data requires complex algorithms and more computing resources, making it difficult to achieve real-time monitoring. Secondly, the data collection environment is highly dependent. Optical remote sensing monitoring is greatly affected by weather and lighting conditions. Although radar remote sensing is not affected by weather, data processing is complex and has high hardware requirements. In addition, the processing process is highly dependent on manpower. Although drone monitoring is flexible, it requires frequent patrol inspections and has limited coverage, making it difficult to effectively monitor multiple mining areas over a large area. Finally, the results analysis is not real-time enough. Traditional monitoring methods are mostly post-analysis, which makes it difficult to perceive the dynamic changes and future trends of mining areas in real time, resulting in delayed law enforcement.

[0005] In view of the shortcomings of the above-mentioned prior art, the present invention proposes an intelligent perception method of mineral law enforcement situation based on multi-source remote sensing data, aiming to achieve real-time and accurate monitoring of mineral mining activities through intelligent means, and improve law enforcement efficiency and coverage. Summary of the invention

[0006] The purpose of the present invention is to provide an intelligent perception method of mineral law enforcement situation based on multi-source remote sensing data to solve the problems raised in the above-mentioned background technology.

[0007] To achieve the above object, the present invention provides the following technical solution: a method for intelligent perception of mineral law enforcement situation based on multi-source remote sensing data, comprising the following steps:

[0008] S1: Mining and extracting key driving factors of intelligent situational awareness of mineral resources. In the process of mining and extracting intelligent situational awareness of mineral resources, key driving factors are divided into four categories: prerequisite factors, driving factors, inertia factors and regulatory factors. There are complex relationships between each type of factors. Through geographic information systems and remote sensing technology, these multi-source key driving factor data that affect mineral resource mining are extracted;

[0009] S2: Establish a historical mineral law enforcement record dataset: Based on the on-site investigation and historical records of law enforcement professionals, combined with satellite images through visual interpretation and correction, the vector boundaries of the illegal plots in the law enforcement monitoring area are delineated to obtain the classification labels of the mineral law enforcement situation. Then, the key driving factors extracted by S1 are used as features to establish a historical mineral law enforcement record dataset as training and verification data for the subsequent situation modeling model;

[0010] S3: Prior probability modeling of mineral law enforcement situation: In order to accurately model the mineral law enforcement situation, the prior probability modeling of mineral law enforcement situation is constructed based on artificial neural network ANN;

[0011] S4: Training and testing of mining area law enforcement situation awareness model: The mining area law enforcement situation awareness model constructed by S2 was trained and tested using the historical mineral law enforcement record dataset to evaluate the prediction accuracy and robustness of the mining area law enforcement situation awareness model;

[0012] S5: Early trend warning based on prior probability update: Use the mining area law enforcement situation awareness model trained in S4 to output the historical prior probability of law enforcement situation awareness, and solve the conditional probability, and update and warn the mineral situation based on Bayesian.

[0013] The specific process of step S1 is as follows:

[0014] S101: Extract prerequisite factors, which are objective conditions for object transfer, mainly including accessibility, road convenience and industrial development. These factors are usually relatively fixed and change relatively slowly, but they have a direct impact on the mining and transportation of mineral resources. For example, areas with high accessibility are more likely to develop and transport mineral resources, while convenience and industrial development determine the economic benefits and sustainability of mineral resource development;

[0015] S102: Extract driving factors, which are the driving factors of object transfer, mainly including economic activities, population density and natural resource endowment. These factors are relatively dynamic and may be adjusted at any time due to changes in policies and market demand. Economic activities reflect the intensity of economic activities in a certain area, which can be represented by the intensity of night lights; population density affects labor supply and market demand; natural resource endowment determines the richness and development of mineral resources, including elevation, slope and soil structure;

[0016] S103: Extract inertia factors, which are the spatiotemporal correlations of the change categories, mainly including historical mining and spatial correlation. These factors reflect the impact of historical and geographical factors on mineral resource development. Regions that have been mined many times in history often have rich experience and infrastructure; spatial correlation reflects the mutual influence between different regions in mineral resource development;

[0017] S104: Extract regulatory factors, which are policy-driven regulatory efforts, mainly including atmospheric environment and soil and water conservation factors, including aerosols and air pollution particle concentrations, which directly affect the sustainable development of mineral resources and environmental protection.

[0018] The accessibility rate is calculated as follows:

[0019]

[0020] Among them, Road i is the road point closest to location x;

[0021] The road convenience is calculated as follows:

[0022]

[0023] Where r is the window radius, d k is the distance from the nearby POI point to the pixel point;

[0024] Industrial development is calculated as follows:

[0025]

[0026] Among them, d k Point to x for the nearest industry i The distance, c i is the distance x i The nearest industrial site.

[0027] The specific process of step S2 is as follows:

[0028] S201: Extraction of illegal mining information: Based on the on-site investigation and historical records of law enforcement professionals, remote sensing image interpretation of mines is carried out. The spectral characteristics, temporal characteristics and spatial characteristics of mine features are combined to gradually identify and appraise mine features. The relationship between mine features and mining rights is considered. The spatial location of the mining rights can basically determine the spatial location of the mine. Mining features are extracted within the scope of the mining rights or within a certain range around them.

[0029] S202: The extracted mining feature data is superimposed on the mining rights data, and spatial verification is performed using the ArcGIS platform. The vector boundaries of the land parcels in the law enforcement monitoring area are delineated through visual interpretation and correction;

[0030] S203: Collect the target time remote sensing images of the study area and the key driving factor data mentioned in S1, perform radiation calibration, atmospheric correction, geometric correction and spatial alignment, and extract the features of the corresponding spatial range according to the mining area vector boundary of S202;

[0031] S204: Integrate the processed mining area key driving factor data and law enforcement occurrence labels into a data set, and use random sampling to collect samples in a ratio of 7:3 to construct training sets and test sets, and finally obtain the historical mineral law enforcement record data set.

[0032] The specific process of step S3 is as follows:

[0033] S301: The input layer receives the multi-source characteristics of independent spatial variables, socio-economic variables and natural climate variables in the mining area and performs standardization processing;

[0034] S302: Build an artificial neural network ANN structure including input layer, hidden layer and output layer. Artificial neural network ANN is a machine learning model paradigm. Each layer contains multiple interconnected neurons, which communicate with each other through weights and activation functions. The advantage of artificial neural network ANN is that it can fit the complex relationship between input data and training targets through multiple iterations of learning and recall;

[0035] In the input layer, each neuron corresponds to an input variable, which is mathematically expressed as:

[0036] X(k,t)=[x1(k,t),x2(k,t),…,x n (k,t)] T

[0037] Among them, x i (k, t) is the i-th variable of grid cell k at simulation time t, and T is the transpose;

[0038] Before the data is input into the neural network, it is necessary to perform S301 standardization so that its values ​​are between the range of [0,1]. The standardization formula is:

[0039]

[0040] In the hidden layer, the signal received by neuron j at time t from all input neurons on grid cell p is estimated according to the following formula:

[0041]

[0042] Among them, net j (k, t) is the signal received by the jth hidden neuron, w i,j is the weight parameter between the input layer and the hidden layer; the hidden layer receives the signal input and generates a response value, and the connection between the hidden layer and the output layer is determined by the activation function; the activation function is the Sigmoid function, which is the most common response function, and its function is:

[0043]

[0044] Each neuron in the output layer corresponds to a specific classification type. The value of the mth neuron in the output layer will generate a value representing the probability of occurrence of the mth category in the grid cell. At training time t, the probability of category c occurring on grid cell k is denoted as P(k,c,t) and is estimated according to the following formula:

[0045]

[0046] Among them, w j,c is the adaptive weight between the hidden layer and the output layer, similar to w i,j , is calibrated during the training process;

[0047] S303: Using the data set established in S2 to train the model, the model is trained by adjusting appropriate training parameters, wherein the training parameters include: number of iterations, batch size, optimizer selection, initial learning rate setting, weight decay rate setting, and loss function selection, so that the model can accurately predict the mineral law enforcement situation;

[0048] S304: Input the remote sensing image of the mining area in the area to be estimated into the trained model to obtain a probability map of illegal events in the mining area, which is used to evaluate the possibility of mining mineral resources.

[0049] The specific process of step S4 is as follows:

[0050] S401: input the training set into the model, and through multiple iterations of training, enable the model to accurately predict the illegal mining situation in the mining area. During the training process, the model continuously adjusts the weight parameters to minimize the loss function value;

[0051] During the training process, the early stopping method is used, that is, when the performance of the model on the validation set no longer improves, the training is stopped in advance to prevent the model from overfitting;

[0052] S402: Use the test set to verify the trained model, and calculate the model's precision, recall, F1 score, and Kappa performance index by comparing the model prediction results with the true labels.

[0053] The specific process of step S5 is as follows:

[0054] S501: Apply the Bayesian update algorithm to process the time series of the historical mineral illegal situation map to improve the stability of the change trend;

[0055] Using the multi-source key driving factor data determined in step S1, the mining area law enforcement situation awareness model established through sufficient training in step 4 is applied to draw a mineral situation prediction map of the study site, and the Bayesian update algorithm is applied to the time series of historical mineral violation situation maps to improve the stability of the change trend by using time information;

[0056] The Bayesian update algorithm can continuously update the estimate of the probability of illegal occurrence (i.e., illegal behavior and normal mining in mining areas), which helps to make more reliable and consistent predictions. This time update can update the future mining behavior changes in mining areas based on historical illegal trends, and help eliminate classification errors and inconsistencies between historical observations that rely only on single-time predictions. The calculation formula for the updated law enforcement probability is shown in the following formula. When a new observation event is obtained in mining area c2 at time i+1, the posterior probability of the true illegal event D in mining area c1 is expressed as:

[0057]

[0058] Among them, c1 and c2 represent the categories of mining law enforcement events in the mining area, i.e., illegal or normal, i represents the time index of the event, P(D c ,i) represents the prior probability of violation at time i, and the conditional probability is calculated as follows:

[0059]

[0060] Among them, the conditional probability P(E c2,i+1 |E c1,i ) represents the probability that a pixel is marked as normal at time i+1, given that it is marked as illegal at time i, and P(E c2,i+1 ∩Ec1,i ) represents the joint probability that a pixel is classified as c1 at time i and as c2 at time i+1;

[0061] S502: Based on the probability update formula, the existing situation estimation probability can be updated through the historical mining area law enforcement event sequence, and the conditional probability of different historical times can be used to estimate the possibility of mining violations in combination with the latest observation data to evaluate the scope of hot and cold areas of mining safety risks;

[0062] S503: Use the predicted mining law enforcement situation images to assess the scope of hot and cold areas of mining safety risks and provide decision support.

[0063] Compared with the prior art, the present invention has the following beneficial effects:

[0064] The present invention establishes a logically reasonable driving factor system through the classification of multi-dimensional driving factors, extracts these multi-source data that affect the exploitation of mineral resources based on remote sensing technology and geographic information systems, and ensures the logical rationality and comprehensiveness of the driving factor system. At the same time, by combining static prerequisite factors, dynamic driving factors, historical inertia factors, and regulatory factors driven by policies, a scientific classification of factors affecting the exploitation of mineral resources is achieved;

[0065] By adopting high-precision data support, the monitoring accuracy of illegal mining situations has been significantly improved. Specifically, advanced remote sensing technology and data processing methods are used to provide data support with high temporal and spatial resolution, so as to achieve real-time monitoring and high-precision positioning of illegal mining activities in mines. This solves the problems of monitoring lag and insufficient precision caused by low data resolution in existing technologies. In addition, by constructing historical data sets, combining satellite images and on-site survey data, and using artificial neural networks to build a law enforcement situation awareness network framework, historical prior probability modeling and early trend warning are carried out. This not only improves the accuracy of monitoring illegal mining activities, but also provides situation prediction and warning based on historical data.

[0066] Through the application of intelligent identification technology, machine learning and deep learning algorithms, the efficiency and accuracy of identifying illegal mining activities can be improved, and real-time monitoring can be achieved. This solves the problem of existing technologies that rely on manual operation and experience judgment and have a low level of intelligence. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 It is a schematic diagram of the process of the present invention;

[0068] Figure 2 It is a schematic diagram of the mining area law enforcement situation awareness model based on artificial neural network of the present invention;

[0069] Figure 3This is a time series mineral illegal situation perception map of Zhaoqing City from 2017 to 2022 in an embodiment of the present invention;

[0070] Figure 4 This is a perception map of illegal mineral activities in Huaiji County, Zhaoqing City in 2018 in the example of the present invention. DETAILED DESCRIPTION

[0071] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0072] Example

[0073] See also Figure 1-Figure 4 The present invention provides a technical solution: a method for intelligently sensing the situation of mineral law enforcement based on multi-source remote sensing data, comprising the following steps:

[0074] S1: Mining and extracting key driving factors for intelligent situational awareness of mineral resources. In the process of mining and extracting intelligent situational awareness of mineral resources, key driving factors are divided into four categories: prerequisite factors, driving factors, inertia factors and regulatory factors;

[0075] The specific process of step S1 is:

[0076] S101: Extract prerequisite factors, which are objective conditions for object transfer, mainly including accessibility, road convenience and industrial development. These factors are usually relatively fixed and change relatively slowly, and have a direct impact on the mining and transportation of mineral resources;

[0077] S102: Extract driving factors, which are the driving factors of object transfer, mainly including economic activities, population density and natural resource endowment. These factors are relatively dynamic and may be adjusted at any time due to changes in policies and market demand. Economic activities reflect the intensity of economic activities in a certain area, which can be represented by the intensity of night lights; population density affects labor supply and market demand; natural resource endowment determines the richness and development of mineral resources, including elevation, slope and soil structure;

[0078] S103: Extract inertia factors, which are the spatiotemporal correlations of the change categories, mainly including historical mining and spatial correlation. These factors reflect the impact of historical and geographical factors on mineral resource development. Regions that have been mined many times in history often have rich experience and infrastructure; spatial correlation reflects the mutual influence between different regions in mineral resource development;

[0079] S104: Extract regulatory factors, which are policy-driven regulatory efforts, mainly including atmospheric environment and soil and water conservation factors, including aerosols and air pollution particle concentrations, which directly affect the sustainable development of mineral resources and environmental protection.

[0080] The accessibility rate is calculated as follows:

[0081]

[0082] Among them, Road i is the road point closest to location x;

[0083] The road convenience is calculated as follows:

[0084]

[0085] Where r is the window radius, d k is the distance from the nearby POI point to the pixel point;

[0086] Industrial development is calculated as follows:

[0087]

[0088] Among them, d k Point to x for the nearest industry i The distance, c i is the distance x i The nearest industrial site;

[0089] S2: Establish a historical mineral law enforcement record dataset: Based on the on-site investigation and historical records of law enforcement professionals, combined with satellite images through visual interpretation and correction, the vector boundaries of the illegal plots in the law enforcement monitoring area are delineated to obtain the classification labels of the mineral law enforcement situation. Then, the key driving factors extracted by S1 are used as features to establish a historical mineral law enforcement record dataset as training and verification data for the subsequent situation modeling model;

[0090] The specific process of step S2 is:

[0091] S201: Extraction of illegal mining information: Based on the on-site investigation and historical records of law enforcement professionals, remote sensing image interpretation of mines is carried out. The spectral characteristics, temporal characteristics and spatial characteristics of mine features are combined to gradually identify and appraise mine features. The relationship between mine features and mining rights is considered. The spatial location of the mining rights can basically determine the spatial location of the mine. Mining features are extracted within the scope of the mining rights or within a certain range around them.

[0092] S202: The extracted mining feature data is superimposed on the mining rights data, and spatial verification is performed using the ArcGIS platform. The vector boundaries of the land parcels in the law enforcement monitoring area are delineated through visual interpretation and correction;

[0093] S203: Collect the target time remote sensing images of the study area and the key driving factor data mentioned in S1, perform radiation calibration, atmospheric correction, geometric correction and spatial alignment, and extract the features of the corresponding spatial range according to the mining area vector boundary of S202;

[0094] S204: Integrate the processed mining area key driving factor data and law enforcement occurrence labels into a data set, and use random sampling to collect samples in a ratio of 7:3 to construct training sets and test sets, and finally obtain a historical mineral law enforcement record data set;

[0095] S3: Prior probability modeling of mineral law enforcement situation: Constructing prior probability modeling of mineral law enforcement situation based on artificial neural network ANN;

[0096] The specific process of step S3 is:

[0097] S301: The input layer receives the multi-source characteristics of independent spatial variables, socio-economic variables and natural climate variables in the mining area and performs standardization processing;

[0098] S302: Building an artificial neural network ANN structure including an input layer, a hidden layer and an output layer. In the input layer, each neuron corresponds to an input variable, which is mathematically expressed as:

[0099] X(k,t)=[x1(k,t),x2(k,t),…,x n (k,t)] T

[0100] Among them, x i (k, t) is the i-th variable of grid cell k at simulation time t, and T is the transpose;

[0101] Before the data is input into the neural network, it is necessary to perform S301 standardization so that its values ​​are between the range of [0,1]. The standardization formula is:

[0102]

[0103] In the hidden layer, the signal received by neuron j at time t from all input neurons on grid cell p is estimated according to the following formula:

[0104]

[0105] Among them, net j(k, t) is the signal received by the jth hidden neuron, w i,j is the weight parameter between the input layer and the hidden layer; the hidden layer receives the signal input and generates a response value, and the connection between the hidden layer and the output layer is determined by the activation function; the activation function is the Sigmoid function, which is the most common response function, and its function is:

[0106]

[0107] Each neuron in the output layer corresponds to a specific classification type. The value of the mth neuron in the output layer will generate a value representing the probability of occurrence of the mth category in the grid cell. At training time t, the probability of category c occurring on grid cell k is denoted as P(k,c,t) and is estimated according to the following formula:

[0108]

[0109] Among them, w j,c is the adaptive weight between the hidden layer and the output layer, similar to w i,j , is calibrated during the training process;

[0110] S303: Using the data set established in S2 to train the model, the model is trained by adjusting appropriate training parameters, wherein the training parameters include: number of iterations, batch size, optimizer selection, initial learning rate setting, weight decay rate setting, and loss function selection, so that the model can accurately predict the mineral law enforcement situation;

[0111] S304: inputting the remote sensing image of the mining area of ​​the area to be estimated into the trained model to obtain a probability map of illegal events in the mining area, which is used to evaluate the possibility of mining mineral resources;

[0112] S4: Training and testing of mining area law enforcement situation awareness model: The mining area law enforcement situation awareness model constructed by S2 was trained and tested using the historical mineral law enforcement record dataset to evaluate the prediction accuracy and robustness of the mining area law enforcement situation awareness model;

[0113] The specific process of step S4 is:

[0114] S401: input the training set into the model, and through multiple iterations of training, enable the model to accurately predict the illegal mining situation in the mining area. During the training process, the model continuously adjusts the weight parameters to minimize the loss function value;

[0115] During the training process, the early stopping method is used, that is, when the performance of the model on the validation set no longer improves, the training is stopped in advance to prevent the model from overfitting;

[0116] S402: Use the test set to verify the trained model, and calculate the model's precision, recall, F1 score, and Kappa performance index by comparing the model's prediction results with the true labels;

[0117] S5: Early trend warning based on prior probability update: Use the mining area law enforcement situation awareness model trained in S4 to output the historical prior probability of law enforcement situation awareness, and solve the conditional probability, and update and warn the mineral situation based on Bayesian;

[0118] The specific process of step S5 is:

[0119] S501: Apply the Bayesian update algorithm to process the time series of the historical mineral illegal situation map to improve the stability of the change trend;

[0120] Using the multi-source key driving factor data determined in step S1, the mining area law enforcement situation awareness model established through sufficient training in step 4 is applied to draw a mineral situation prediction map of the study site, and the Bayesian update algorithm is applied to the time series of historical mineral violation situation maps to improve the stability of the change trend by using time information;

[0121] The Bayesian update algorithm can continuously update the estimate of the probability of violation, which helps to make more reliable and consistent predictions. It can update the future mining behavior changes based on historical violation trends, and help eliminate classification errors and inconsistencies between historical observations that rely only on single-time predictions. The calculation formula for the updated law enforcement probability is shown in the following formula. When a new observation event is obtained in mining area c2 at time i+1, the posterior probability of the actual violation event D in mining area c1 is expressed as:

[0122]

[0123] Among them, c1 and c2 represent the categories of mining law enforcement events in the mining area, i.e., illegal or normal, i represents the time index of the event, P(D c ,i) represents the prior probability of violation at time i, and the conditional probability is calculated as follows:

[0124]

[0125] Among them, the conditional probability P(E c2,i+1 |E c1,i ) represents the probability that a pixel is marked as normal at time i+1, given that it is marked as illegal at time i, and P(E c2,i+1 ∩E c1,i ) represents the joint probability that a pixel is classified as c1 at time i and as c2 at time i+1;

[0126] S502: Based on the probability update formula, the existing situation estimation probability can be updated through the historical mining area law enforcement event sequence, and the conditional probability of different historical times can be used to estimate the possibility of mining violations in combination with the latest observation data to evaluate the scope of hot and cold areas of mining safety risks;

[0127] S503: Use the predicted mining law enforcement situation images to assess the scope of hot and cold areas of mining safety risks and provide decision support.

[0128] Take the Zhaoqing City Mineral Law Enforcement Situation Awareness Project as an example:

[0129] Step 1: Collect remote sensing image data and multi-source driving factors of Zhaoqing City;

[0130] In the Zhaoqing City Mineral Law Enforcement Situation Awareness Project, collecting remote sensing image data and mining rights and mining area data is the basis for the implementation of the entire project. Zhaoqing City is located in the central and western part of Guangdong Province and is an important node city in the Guangdong-Hong Kong-Macao Greater Bay Area. It has three urban districts (Duanzhou District, Dinghu District, Gaoyao District), four counties (Guangning County, Deqing County, Fengkai County, Huaiji County), and one county-level city (Sihui City) under its jurisdiction, with a total land area of ​​14,900 km2. We selected Sentinel-2 and Google high-resolution image data to ensure that the remote sensing images of Zhaoqing City from 2017 to 2022 are covered. In order to ensure the timeliness and coverage of the data, multi-phase images are obtained every year, especially in the dry and rainy seasons, so as to capture the changes in the mining area in different seasons. After the data is acquired, a series of pre-processing such as projection transformation, geometric correction, resampling, cropping, and normalization are performed on the data, providing a solid data foundation for subsequent mining area identification and change detection.

[0131] In terms of multi-source factor data collection, the prerequisite factors such as traffic convenience, accessibility, and industrial development are extracted through distance analysis using Amap road network data and POI data, and the driving factor data corresponding to the study area are clipped through global data sets such as NOAA's VIIRS night lights, NASA EOSDIS, and WorldPop. Inertial factor data such as historical mining data can be obtained from the Zhaoqing Natural Resources Bureau and remote sensing visual interpretation. In terms of regulatory factors, aerosol and air quality data provided by the ESAClimate Office and NASA Air Quality Data can be used.

[0132] Step 2: Extract information on illegal mining and establish a mineral resources law enforcement monitoring database;

[0133] The city's mining rights and mining area data were obtained from the Zhaoqing Municipal Natural Resources Bureau, including detailed information such as the location, area, type of mineral, and mining period of the mining area. In addition, detailed information on illegal mining in mines was extracted based on remote sensing images.

[0134] When identifying and appraising mining features, we use the spectral, temporal and spatial characteristics of remote sensing images to identify and appraise mining features. We determine the location and scope of mining features by combining visual interpretation with automated classification. We use the ArcGIS platform to perform spatial verification by overlaying the extracted mining feature data with the mining rights data. We draw the vector boundaries of the plots within the law enforcement monitoring area through visual interpretation correction. The purpose of this step is to ensure that the extracted mining features are consistent with the spatial location of the actual mining rights, thereby improving the accuracy and reliability of the data.

[0135] Next, the processed multi-source feature data of the mining area and the law enforcement occurrence labels were integrated into a data set. The data set was divided into a training set and a test set in a ratio of 7:3 by random sampling. The training set was used for model training, and the test set was used for model verification. Finally, a mineral law enforcement situation monitoring database was constructed, which provided basic data support for the construction of subsequent models.

[0136] Step 3: Construct a mining area law enforcement situation awareness model based on artificial neural network;

[0137] A mining area law enforcement situation awareness model was constructed using ANN. The input layer receives independent spatial variables, socioeconomic variables, natural climate variables and other multi-source characteristics of the mining area, and performs standardization so that the values ​​of all input variables are between 0 and 1. The network includes an input layer, a hidden layer and an output layer. The input layer size is the number of input multi-source variables, which is 16 in this application, the number of intermediate layers is 32, and the number of output layers is 2.

[0138] The model is trained using the constructed mineral law enforcement situation dataset. The training parameters include the number of iterations, batch size, optimizer selection, initial learning rate setting, weight decay rate setting, and loss function selection. Parameter selection is crucial to the training effect of the model. We set the number of training to 1000, select the Adam optimizer, the initial learning rate to 0.01, the training batch size to 4, the momentum to 0.9, and the weight decay to 0.0001. The cross-validation method is used to ensure the generalization ability of the model.

[0139] Step 4: Training and testing of mining area law enforcement situation awareness model;

[0140] The constructed artificial neural network model was trained and tested, and the training set was input into the model. Through multiple iterations of training, the model was able to accurately predict the illegal mining situation in the mining area. During the training process, the model continuously adjusted the weight parameters to minimize the loss function value. We adopted the early stopping method, that is, when the performance of the model on the validation set no longer improved, the training was stopped in advance to prevent the model from overfitting.

[0141] The trained model was verified using the test set to evaluate the model's prediction accuracy and robustness. By comparing the model's prediction results with the true labels, the model's performance indicators such as precision, recall, F1 score, and Kappa were calculated. Specifically, the accuracy of all districts and counties in Zhaoqing City was around 80%, and the recall rate was close to 90%, proving that the monitoring model performed well. Fengkai County performed best in all indicators, with an F1 score of 0.85 and a Kappa coefficient of 0.75, followed by Guangning County and Dinghu District. The accuracy of each district and county was not much different, indicating that the model can take into account the impact of spatial heterogeneity of illegal situations. Dinghu District, Huaiji County, and Deqing County performed relatively consistently, with similar indicators, especially the recall rate and F1 score of 0.88 and 0.82.

[0142] Step 5: Application and analysis of mining illegal situation awareness guided by prior probability;

[0143] Finally, based on the trained model, the application and analysis of the perception of illegal mining situations guided by prior probability was carried out. The historical prior probability of law enforcement situation perception output by the model was used, and the conditional probability was calculated in combination with the historical experience mining data. Based on the Bayesian formula, the illegal mining situation of minerals was updated and warned. The Bayesian formula calculates the posterior probability through the prior probability and likelihood function, thereby realizing the prediction of future situations. According to the changes in conditional probability, the illegal mining trend in the mining area can be discovered in time and early warning can be issued. By comparing the remote sensing images of different time periods, the changes in the mining area are analyzed to identify potential illegal mining behaviors.

[0144] Specifically, during the period of 2017-2022, the overall situation of illegal mining in Zhaoqing City showed a certain fluctuation trend. 2017 was the peak period of illegal activities, while 2019 and 2021 were the trough periods, indicating the effectiveness of law enforcement and governance measures. Dinghu District, Deqing County, Guangning County and Huaiji County are the key areas of illegal activities, and more targeted law enforcement and monitoring measures are needed.

[0145] Obviously, the above embodiments of the present invention are merely examples for clearly explaining the present invention, and are not intended to limit the implementation methods of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all implementation methods here. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the claims of the present invention.

[0146] Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative work shall fall within the scope of protection of the present invention.

Claims

1. An intelligent perception method for mineral law enforcement situation based on multi-source remote sensing data, characterized in that: The following steps are involved: S1: Mining and extracting key driving factors for intelligent situational awareness of mineral resources. In the process of mining and extracting intelligent situational awareness of mineral resources, key driving factors are divided into four categories: prerequisite factors, driving factors, inertia factors and regulatory factors; S2: Establish a historical mineral law enforcement record dataset: Based on the on-site investigation and historical records of law enforcement professionals, combined with satellite images through visual interpretation and correction, the vector boundaries of the illegal plots in the law enforcement monitoring area are delineated to obtain the classification labels of the mineral law enforcement situation. Then, the key driving factors extracted by S1 are used as features to establish a historical mineral law enforcement record dataset as training and verification data for the subsequent situation modeling model; S3: Prior probability modeling of mineral law enforcement situation: Constructing prior probability modeling of mineral law enforcement situation based on artificial neural network ANN; S4: Training and testing of mining area law enforcement situation awareness model: The mining area law enforcement situation awareness model constructed by S2 was trained and tested using the historical mineral law enforcement record dataset to evaluate the prediction accuracy and robustness of the mining area law enforcement situation awareness model; S5: Early trend warning based on prior probability update: Use the mining area law enforcement situation awareness model trained in S4 to output the historical prior probability of law enforcement situation awareness, and solve the conditional probability, and update and warn the mineral situation based on Bayesian.

2. The method for intelligent perception of mineral law enforcement situation based on multi-source remote sensing data according to claim 1 is characterized by: The specific process of step S1 is: S101: Extract prerequisite factors, which are objective conditions for object transfer, mainly including accessibility, road convenience and industrial development. These factors are usually relatively fixed and change relatively slowly, and have a direct impact on the mining and transportation of mineral resources; S102: Extract driving factors, which are the driving factors of object transfer, mainly including economic activities, population density and natural resource endowment. These factors are relatively dynamic and may be adjusted at any time due to changes in policies and market demand. Economic activities reflect the intensity of economic activities in a certain area, which can be represented by the intensity of night lights; population density affects labor supply and market demand; natural resource endowment determines the richness and development of mineral resources, including elevation, slope and soil structure; S103: Extract inertia factors, which are the spatiotemporal correlations of the change categories, mainly including historical mining and spatial correlation. These factors reflect the impact of historical and geographical factors on mineral resource development. Regions that have been mined many times in history often have rich experience and infrastructure; spatial correlation reflects the mutual influence between different regions in mineral resource development; S104: Extract regulatory factors, which are policy-driven regulatory efforts, mainly including atmospheric environment and soil and water conservation factors, including aerosols and air pollution particle concentrations, which directly affect the sustainable development of mineral resources and environmental protection.

3. The method for intelligent perception of mineral law enforcement situation based on multi-source remote sensing data according to claim 2 is characterized by: The accessibility rate is calculated as follows: Among them, Road i is the road point closest to location x; The road convenience is calculated as follows: Where r is the window radius, d k is the distance from the nearby POI point to the pixel point; Industrial development is calculated as follows: Among them, d k Point to x for the nearest industry i The distance, c i is the distance x i The nearest industrial site.

4. The method for intelligent perception of mineral law enforcement situation based on multi-source remote sensing data according to claim 1 is characterized by: The specific process of step S2 is: S201: Extraction of illegal mining information: Based on the on-site investigation and historical records of law enforcement professionals, remote sensing image interpretation of mines is carried out. The spectral characteristics, temporal characteristics and spatial characteristics of mine features are combined to gradually identify and appraise mine features. The relationship between mine features and mining rights is considered. The spatial location of the mining rights can basically determine the spatial location of the mine. Mining features are extracted within the scope of the mining rights or within a certain range around them. S202: The extracted mining feature data is superimposed on the mining rights data, and spatial verification is performed using the ArcGIS platform. The vector boundaries of the land parcels in the law enforcement monitoring area are delineated through visual interpretation and correction; S203: Collect the target time remote sensing images of the study area and the key driving factor data mentioned in S1, perform radiation calibration, atmospheric correction, geometric correction and spatial alignment, and extract the features of the corresponding spatial range according to the mining area vector boundary of S202; S204: Integrate the processed mining area key driving factor data and law enforcement occurrence labels into a data set, and use random sampling to collect samples in a ratio of 7:3 to construct training sets and test sets, and finally obtain the historical mineral law enforcement record data set.

5. The method for intelligent perception of mineral law enforcement situation based on multi-source remote sensing data according to claim 1 is characterized by: The specific process of step S3 is: S301: The input layer receives the multi-source characteristics of independent spatial variables, socio-economic variables and natural climate variables in the mining area and performs standardization processing; S302: Building an artificial neural network ANN structure including an input layer, a hidden layer and an output layer. In the input layer, each neuron corresponds to an input variable, which is mathematically expressed as: X(k,t)=[x1(k,t),x2(k,t),…,x n (k,t)] T Among them, x i (k, t) is the i-th variable of grid cell k at simulation time t, and T is the transpose; Before the data is input into the neural network, it is necessary to perform S301 standardization so that its values ​​are between the range of [0,1]. The standardization formula is: In the hidden layer, the signal received by neuron j at time t from all input neurons on grid cell p is estimated according to the following formula: Among them, net j (k, t) is the signal received by the jth hidden neuron, w i,j is the weight parameter between the input layer and the hidden layer; the hidden layer receives the signal input and generates a response value, and the connection between the hidden layer and the output layer is determined by the activation function; the activation function is the Sigmoid function, which is the most common response function, and its function is: Each neuron in the output layer corresponds to a specific classification type. The value of the mth neuron in the output layer will generate a value representing the probability of occurrence of the mth category in the grid cell. At training time t, the probability of category c occurring on grid cell k is denoted as P(k,c,t) and is estimated according to the following formula: Among them, w j,c is the adaptive weight between the hidden layer and the output layer, similar to w i,j , is calibrated during the training process; S303: Using the data set established in S2 to train the model, the model is trained by adjusting appropriate training parameters, wherein the training parameters include: number of iterations, batch size, optimizer selection, initial learning rate setting, weight decay rate setting, and loss function selection, so that the model can accurately predict the mineral law enforcement situation; S304: Input the remote sensing image of the mining area in the area to be estimated into the trained model to obtain a probability map of illegal events in the mining area, which is used to evaluate the possibility of mining mineral resources.

6. The method for intelligent perception of mineral law enforcement situation based on multi-source remote sensing data according to claim 1 is characterized by: The specific process of step S4 is: S401: input the training set into the model, and through multiple iterations of training, enable the model to accurately predict the illegal mining situation in the mining area. During the training process, the model continuously adjusts the weight parameters to minimize the loss function value; During the training process, the early stopping method is used, that is, when the performance of the model on the validation set no longer improves, the training is stopped in advance to prevent the model from overfitting; S402: Use the test set to verify the trained model, and calculate the model's precision, recall, F1 score, and Kappa performance index by comparing the model prediction results with the true labels.

7. The method for intelligent perception of mineral law enforcement situation based on multi-source remote sensing data according to claim 1 is characterized by: The specific process of step S5 is: S501: Apply the Bayesian update algorithm to process the time series of the historical mineral illegal situation map to improve the stability of the change trend; Using the multi-source key driving factor data determined in step S1, the mining area law enforcement situation awareness model established through sufficient training in step 4 is applied to draw a mineral situation prediction map of the study site, and the Bayesian update algorithm is applied to the time series of historical mineral violation situation maps to improve the stability of the change trend by using time information; The Bayesian update algorithm can continuously update the estimate of the probability of violation, which helps to make more reliable and consistent predictions. It can update the future mining behavior changes based on historical violation trends, and help eliminate classification errors and inconsistencies between historical observations that rely only on single-time predictions. The calculation formula for the updated law enforcement probability is shown in the following formula. When a new observation event is obtained in mining area c2 at time i+1, the posterior probability of the actual violation event D in mining area c1 is expressed as: Among them, c1 and c2 represent the categories of mining law enforcement events in the mining area, i.e., illegal or normal, i represents the time index of the event, P(D c ,i) represents the prior probability of violation at time i, and the conditional probability is calculated as follows: Among them, the conditional probability P(E c2,i+1 |E c1,i ) represents the probability that a pixel is marked as normal at time i+1, given that it is marked as illegal at time i, and P(E c2,i+1 ∩E c1,i ) represents the joint probability that a pixel is classified as c1 at time i and as c2 at time i+1; S502: Based on the probability update formula, the existing situation estimation probability can be updated through the historical mining area law enforcement event sequence, and the conditional probability of different historical times can be used to estimate the possibility of mining violations in combination with the latest observation data to evaluate the scope of hot and cold areas of mining safety risks; S503: Use the predicted mining law enforcement situation images to assess the scope of hot and cold areas of mining safety risks and provide decision support.

Images