Ionic rare earth mining area monitoring and area estimation method and system

Abstract

The present invention provides a method and system for monitoring and estimating the area of ​​an ionic rare earth mining area, including: obtaining remote sensing image data containing a target detection area, and preprocessing the remote sensing image; making the preprocessed remote sensing image into a data set for identifying the characteristics of ionic rare earth mining activities for target detection; constructing a YOLOv8 target detection model and improving it; using the prepared data set to train the improved YOLOv8 target detection model, and using the trained YOLOv8 target detection model to identify the characteristic categories of ionic rare earth mining activities; according to the identification results, calculating the spatial position information of each mining activity characteristic category in the geographic coordinate system, so as to monitor and accurately locate the ionic rare earth mining behavior. The present invention solves the problems of scattered distribution of mining points and high labor cost consumption in the existing ionic rare earth mining activity monitoring, and can accurately identify the ionic rare earth mining behavior and interpret remote sensing images.

Description

Technical Field

[0001] The invention belongs to the technical field of mineral remote sensing detection application, and in particular relates to an ionic rare earth mining area monitoring and area estimation method and system based on remote sensing images. Background Art

[0002] Rare earth resources are known as "industrial gold" and "industrial vitamins" and are vital to contemporary science and technology and industry. Their excellent optical and electromagnetic properties make them performance enhancers in the metallurgical field, and they are used as efficient catalysts in petrochemicals. They are also widely used in glass, ceramics, new materials, new energy and other industries, greatly promoting technological innovation and development. As an engine driving national industrial upgrading and economic structural transformation, the strategic value of rare earth elements is immeasurable. They are not only a valuable asset in the process of industrialization, but also a strategic pillar of national security and global economic competitiveness.

[0003] On this basis, it is essential to monitor ionic rare earth mining areas and evaluate the mining intensity of mining areas. In view of the particularity of ionic rare earth mines, most mining areas are located in remote mountainous areas with high mountains and dense forests, scattered mining areas and numerous mining sites, which is a typical small-scale manual mining model. Traditional monitoring methods are mainly based on visual interpretation, and visual interpretation consumes too much time, which leads to a doubling of human and material costs. Although the accuracy is high, the efficiency is extremely low, which is far from meeting the urgent needs of real-time dynamic monitoring.

[0004] As artificial intelligence technology is changing with each passing day, deep learning, as the core of machine learning, makes it possible to automatically identify the mining characteristics of rare earth mining areas. It can realize the intelligent identification of mining characteristics, significantly shorten the monitoring cycle, improve efficiency, and show great potential in cost control. In recent years, my country's scientific research strength has leapt forward, accompanied by the emergence of a large amount of high-resolution domestic satellite remote sensing data, which provides a solid data foundation for image recognition of deep learning algorithms. Combining the wide-area coverage advantage of remote sensing technology with the powerful recognition ability of deep learning will not only deepen the research and application of remote sensing target recognition, but also bring revolutionary improvements to the detailed investigation of mining resources, environmental impact assessment and regulatory effectiveness.

[0005] Previous studies on ionic rare earth mining areas based on remote sensing technology have the following problems: First, the traditional method of identifying ionic rare earth mining areas based on remote sensing images is mainly based on manual visual interpretation and object-oriented recognition. Ionic rare earth mines are scattered in hilly and mountainous areas in the form of mining point groups, with a large distribution range. Visual interpretation requires a lot of manpower and material resources, high cost and long cycle. Object-oriented methods require manual rule-making, which is inefficient and subject to significant subjective influences, and it is difficult to meet the needs of dynamic supervision of rare earth mining areas; second, since ionic rare earth mines are typical small-scale manual mining models, the mining areas are scattered, there are many mining points, and the boundaries of the mining areas are blurred, the traditional method of using vegetation index and bare soil index to identify mining area boundaries cannot accurately locate rare earth mining points. Summary of the invention

[0006] One purpose of the present invention is to address the shortcomings of the prior art and provide a method for monitoring ionic rare earth mining areas based on remote sensing images. The method solves the problems of scattered mining points, blurred mining area boundaries, difficulty in accurate positioning, and high labor costs in existing ionic rare earth mining activity monitoring methods, and can accurately identify ionic rare earth mining activities and interpret remote sensing images.

[0007] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0008] A remote sensing image-based ionic rare earth mining area monitoring method, characterized in that it comprises the following steps:

[0009] Step 1: Obtain remote sensing image data containing the target detection area and preprocess the remote sensing image;

[0010] Step 2, preparing the remote sensing image preprocessed in step 1 into a data set of ionic rare earth mining activity feature identification for target detection;

[0011] Step 3: Build the YOLOv8 target detection model and improve it;

[0012] Step 4: Use the data set prepared in step 2 to train the improved YOLOv8 target detection model, and use the trained YOLOv8 target detection model to identify the characteristic categories of ionic rare earth mining activities;

[0013] Step 5: Based on the identification results of step 4, calculate the spatial position information of each mining activity characteristic category in the geographic coordinate system to monitor and accurately locate the ionic rare earth mining behavior.

[0014] Furthermore, the method for making a data set in step 2 is:

[0015] Convert remote sensing images from geographic coordinate system to plane coordinate system;

[0016] The remote sensing images are cut into image slices, and all sedimentation tanks and high-level tanks are manually marked on the slices and used as positive samples. At the same time, a certain number of non-sedimentation tanks and non-high-level tanks are marked and used as negative samples.

[0017] The data enhancement method is used to expand the positive and negative samples, and finally all samples are randomly divided into training set, validation set and test set according to a certain proportion.

[0018] Furthermore, the method for improving the YOLOv8 target detection model in step 3 is as follows:

[0019] (1) The Asymptotic Feature Pyramid Network structure is introduced to improve the neck network structure of the original YOLOv8 target detection model to integrate high-level semantic information with low-level spatial position information;

[0020] (2) Integrate the Convolutional Block Attention Module into the neck network structure of the original YOLOv8 target detection model;

[0021] (3) The Slicing Aided Hyper inference module is introduced in the model inference phase. This module divides the image to be inferred into M*N slices with overlapping parts, performs target detection on each slice, sets the intersection-over-union ratio threshold, uses non-maximum suppression to eliminate redundant overlapping detection frames, and finally merges the inference results back to the original image size.

[0022] Furthermore, the specific method of introducing the progressive feature pyramid network structure to improve the neck network structure of the original YOLOv8 target detection model includes:

[0023] The third CSPLayer_2Conv module and the last SPPF module in the feature extraction network structure are connected to multiple ASFF2 modules respectively, and then the multiple ASFF2 modules are connected to the upsampling structure including multiple consecutive CSPLayer_2Conv modules respectively, and at the same time, the second CSPLayer_2Conv module in the feature extraction network structure is connected to the last CSPLayer_2Conv module in the upsampling structure; then, the CSPLayer_2Conv module in the upsampling structure is connected to the ASFF3 module respectively, and finally, the multiple ASFF3 modules are connected to the downsampling structure including multiple consecutive CSPLayer_2Conv modules respectively.

[0024] Furthermore, the specific method of integrating the convolutional block attention module into the neck network structure of the original YOLOv8 target detection model includes:

[0025] The multiple CSPLayer_2Conv modules in the downsampling structure including multiple consecutive CSPLayer_2Conv modules are respectively connected to multiple CBMA attention modules, and then the multiple CBMA attention modules are respectively connected to the Dectect detection head modules for different target sizes.

[0026] Furthermore, the method for calculating the spatial location information of each mining activity feature category in the geographic coordinate system includes:

[0027] The pixel coordinate system coordinates of the center point of the identified object generated by the improved YOLOv8 target detection model are converted into the plane coordinate system coordinates representing the real space position based on the pixel coordinate system coordinates of the image slice, where the conversion formula between pixel coordinates and plane coordinates is:

[0028]

[0029]

[0030] In the formula, , Represents the pixel coordinates of the center point of the detection box; , Indicates the pixel coordinates of the upper left corner of the detection box; , Indicates the pixel coordinates of the lower right corner of the detection box; , express The pixel coordinates of the point; , express Plane coordinates of a point; , Represents the plane coordinates of the upper left corner of the slice; , Indicates the spatial resolution of the remote sensing image in the horizontal and vertical directions.

[0031] Another object of the present invention is to provide a method for estimating the area of ​​an ionic rare earth mining area based on remote sensing images, which is characterized in that, based on the recognition results of the above-mentioned ionic rare earth mining area monitoring method based on remote sensing images, a Segment Anything segmentation model is used to perform local segmentation based on a detection frame, and pixel-level segmentation is performed according to each mining activity feature category. Finally, based on the pixel-level segmentation results, the area of ​​each mining activity feature category is estimated.

[0032] Furthermore, the pixel level segmentation method includes:

[0033] Crop the remote sensing image into image slices, input the image slices into the Segment Anything segmentation model, and for each image slice, read the detection box and label of the object in the image slice;

[0034] The Segment Anything segmentation model is used to segment the image slices based on the detection box of the object. Each segmented region is then converted into a binary mask, which is vectorized into polygons using the vectorize_mask function. Each polygon is assigned a label based on the detection result.

[0035] Compile the vectorized and labeled polygons into a GeoDataFrame, including the original coordinate system of the image tiles, and save it.

[0036] Further, the area estimation method is as follows:

[0037] Based on the pixel-level segmentation results, the segmented image is converted from BGR format to RGB format. For each segmented identified object, its pixel area is calculated and converted into the actual area through the given spatial resolution. The area calculation formula is:

[0038]

[0039] In the formula, is the area of ​​each identified object in the detection frame; is the number of pixels of the segmentation result within the detection frame; , Correspondingly, it represents the spatial resolution of the remote sensing image in the horizontal and vertical directions;

[0040] After calculating the area of ​​each identified object in the detection frame, the areas are accumulated and counted according to the categories of the identified objects to obtain the total area of ​​each characteristic category of mining activities in the study area. Then, the pastemask2img function is used to superimpose the segmentation mask on the original image and mark the detection frame and calculated area of ​​each identified object.

[0041] The present invention also provides a system for implementing the above-mentioned ionic rare earth mining area monitoring method, characterized in that it includes:

[0042] A data acquisition and processing module is used to acquire remote sensing image data including the target detection area and pre-process the remote sensing image;

[0043] A data set making module is used to make the remote sensing images preprocessed by the data acquisition processing module into a characteristic identification data set of ionic rare earth mining activities for target detection;

[0044] The model building and training module is used to build a YOLOv8 target detection model, and improve the YOLOv8 target detection model based on the characteristics of ionic rare earth mining activities. The improved YOLOv8 target detection model is trained using the ionic rare earth mining activity feature identification dataset produced by the dataset production module, and the trained YOLOv8 target detection model is used to identify the feature categories of ionic rare earth mining activities.

[0045] The monitoring and positioning module is used to calculate the spatial position information of each mining activity characteristic category in the geographic coordinate system based on the recognition results of the model construction and training modules, so as to monitor and accurately locate the ionic rare earth mining behavior.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] 1. The present invention aims at the problems of scattered distribution of mining points, fuzzy boundaries of mining areas, difficulty in accurate positioning, and high labor cost consumption in existing ionic rare earth mining activity monitoring methods. The present invention cleverly selects sedimentation tanks and high-level tanks, typical structures in ionic rare earth mining areas, as characteristic marks of ionic rare earth mining activities and remote sensing interpretation marks of rare earth mining points. As important links in the ionic rare earth mining process and unique artificial structures in ionic rare earth mining areas, most of them have regular shapes and are significantly different from other landforms in nature. They present special colors and reflect water characteristics in different sensor band combinations. They have high recognition in remote sensing images and can be accurately used as characteristic marks of ionic rare earth mining behavior and remote sensing image interpretation marks;

[0048] 2. The present invention effectively realizes the rapid identification and precise positioning of sedimentation tanks and high-level tanks in ionic rare earth mining areas; in view of the problem that existing methods are difficult to quickly and accurately realize the identification and coordinate positioning of sedimentation tanks and high-level tanks in ionic rare earth mining areas, the present invention improves the YOLOv8 target detection model based on the characteristics of sedimentation tanks and high-level tanks in high-resolution remote sensing images, which have a small area, regular shape, and obvious boundaries, so that it is easier to capture and identify sedimentation tanks and high-level tanks as small target objects in remote sensing images, and can output the coordinates of the recognition results in the pixel coordinate system of the image slice, and then combine with the coordinate conversion formula to calculate the spatial position information of the sedimentation tank and the high-level tank in the geographic coordinate system, so as to realize the monitoring and accurate positioning of ionic rare earth mining behavior; based on long-time series high-resolution remote sensing satellite images, it can realize continuous monitoring of ionic rare earth mining activities, and trace back the evolution process and development trend of the mining area;

[0049] 3. The present invention effectively realizes the refined extraction and area estimation of sedimentation tanks and high-level tanks in ionic rare earth mining areas; in view of the problem that existing methods are difficult to quickly and accurately realize the area calculation of sedimentation tanks and high-level tanks in ionic rare earth mining areas, the present invention adopts a deep learning strategy that combines object-level target detection and pixel-level segmentation. First, the improved YOLOv8 target detection model is used to detect and locate sedimentation tanks and high-level tanks. Secondly, the relevant parameters of the Segment Anything segmentation model are modified according to the detection results, and the main objects inside the detection frame are segmented at the pixel level and the corresponding category information is assigned. The area information of the sedimentation tank and the high-level tank is obtained by combining the area calculation formula, so as to realize the refined extraction and area estimation of the sedimentation tank and the high-level tank that mark the ionic rare earth mining activities. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A flow chart of a method for detecting and estimating the area of ​​a sedimentation pond in a rare earth mining area based on remote sensing provided by an embodiment of the present invention;

[0051] Figure 2 This is a diagram of the improved YOLOv8 target detection network structure according to an embodiment of the present invention. DETAILED DESCRIPTION

[0052] The technical solutions in the embodiments of the present invention will be described clearly and completely below in combination with the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0053] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0054] The present invention will be further described below in conjunction with specific embodiments, but the present invention is not limited thereto.

[0055] like Figure 1 As shown, an embodiment of the present invention provides a remote sensing-based ionic rare earth mining area sedimentation tank detection method. The research object of this embodiment is a typical ionic rare earth mining area in 2013, 2016, 2018, 2021 and 2023. This embodiment aims to identify the characteristic features of the mining area through remote sensing and deep learning technology, monitor and track the mining activity points and their changing trends in the rare earth mining area in each year, and realize the visualization of the long time series and the whole process of mining area development, further analyze the changes and development trends of mining behavior in the rare earth mining area over the past 10 years, and then realize the quantitative evaluation of the development degree and resource reserves of the mining area, so as to assist relevant departments in more scientific and reasonable planning and development of rare earth mining areas. It includes the following steps:

[0056] Step 1: Obtain remote sensing image data containing the target detection area and preprocess the remote sensing image;

[0057] In this embodiment, the Google Earth high-resolution satellite map base map of the study area in 2013, 2016, 2018, 2021 and 2023 is downloaded based on the Google Earth platform, and the image spatial resolution is 2.68e-06°. The image is preprocessed, which includes de-clouding the remote sensing image, and then geometrically correcting, radiometrically calibrating, and atmospherically correcting the image, and then the remote sensing image is cropped using the study area vector file.

[0058] Step 2: The remote sensing image pre-processed in step 1 is made into an ionic rare earth mining activity feature identification data set for target detection; in the embodiment, the preparation method is as follows:

[0059] The Google Earth satellite map base map of all study areas is converted from the WGS_1984 geographic coordinate system to the WGS_1984_UTM_zone_47N plane coordinate system, and the image spatial resolution is 0.284m; the Google Earth satellite map base map of the study area in 2023 after the coordinate system conversion is cut into image slices according to the size of 512pixel*512pixel, and all sedimentation tanks and high-level tanks are manually marked on the slices using Labelme software to make positive samples. At the same time, a certain number of non-sedimentation tanks and high-level tanks in the study area, such as buildings and farmland, are marked as negative samples; data enhancement methods such as rotation, mirroring, brightness, chromaticity, mosaic, and Gaussian blur are used to expand the samples, and all samples are randomly divided into training sets, validation sets, and test sets at a ratio of 8:1:1 (or other ratios) for the sedimentation tank and high-level tank detection model to train. Step 3. Build a YOLOv8 target detection model and improve it;

[0060] In this embodiment, the improved YOLOv8 target detection model structure is as follows: Figure 2 As shown, the improvement method is as follows:

[0061] (1) Introduce the Asymptotic Feature Pyramid Network (AFPN) structure to improve the original neck network structure and realize the mutual fusion of high-level semantic information and low-level spatial position information; specifically, connect the third CSPLayer_2Conv module and the last SPPF module in the feature extraction network structure to two ASFF2 modules; then connect these two ASFF2 modules to the upsampling structure composed of three consecutive CSPLayer_2Conv modules, and at the same time, connect the second CSPLayer_2Conv module in the feature extraction network structure to the last CSPLayer_2Conv module in the upsampling structure. Then connect the three CSPLayer_2Conv modules in the upsampling structure to three ASFF3 modules. Then connect these three ASFF3 modules to the downsampling structure composed of three consecutive CSPLayer_2Conv modules;

[0062] Through the above improvements, the network starts to fuse adjacent low-level features and gradually introduces high-level features, achieving more comprehensive and rich information transmission and fusion between different levels, while alleviating the inconsistency of multi-object information through adaptive spatial fusion operations;

[0063] (2) Use the Convolutional Block Attention Module (CBAM) to integrate into the original neck network structure to improve the detection performance of the network. Specifically, the three CSPLayer_2Conv modules in the downsampling structure composed of three consecutive CSPLayer_2Conv modules are connected to three CBAM attention modules respectively, and then the three CBAM attention modules are connected to the Dectect detection head modules for different target sizes respectively;

[0064] Through the above improvements, the CBMA attention module infers the attention map from the channel and spatial dimensions in turn, and then multiplies the obtained attention map with the input feature map to perform adaptive feature refinement;

[0065] (3) The Slicing Aided Hyper inference (SAHI) module is introduced in the model reasoning phase. This module is designed specifically for small object detection. The module first divides the image to be inferred into M*N slices with overlapping parts, performs target detection on each slice, sets the intersection-over-union ratio threshold, uses non-maximum suppression to eliminate redundant overlapping detection frames, and finally merges the inference results back to the original image size.

[0066] Using SAHI for auxiliary reasoning can effectively improve the model's detection effect on small targets and help capture smaller sedimentation ponds in remote sensing images.

[0067] Step 4: Use the data set prepared in step 2 to train the improved YOLOv8 target detection model, and use the trained YOLOv8 target detection model to identify the characteristic categories of ionic rare earth mining activities;

[0068] In the embodiment, the improved YOLOv8 target detection model is first trained, and the specific method is: the operating system is Windows 10, the processor is Intel (R) Core (TM) i9-14900K 3.20 GHz, the GPU is NVIDIA GeForceRTX 4090, the video memory is 64G, and MMYOLO is used as the deep learning framework; in terms of parameter setting, the initial learning rate is set to 0.0001, the learning rate decay coefficient is 0.1, the batch size is set to 16, a total of 300 iterations, and the confidence threshold is set to 0.7. Based on the above hardware configuration and software, the YOLOv8 target detection model is trained using the training set and verified using the validation set, and finally the test set is used to test to obtain the YOLOv8 target detection model with the best performance.

[0069] After obtaining the trained YOLOv8 target detection model, the Google Earth satellite map base maps of the study area in 2013, 2016, 2018, 2021, and 2023 were cropped into image slices according to the size of 512pixel*512pixel, and the above slices were put into the trained YOLOv8 target detection model for recognition. When the YOLOv8 target detection model recognizes that there are sedimentation ponds or high-level ponds in the image slices, the identified object (i.e., sedimentation ponds or high-level ponds) will be marked with a rectangular detection frame at the corresponding position of the image slice. At the same time, the pixel coordinates of the center point of the identified object based on the image slice pixel coordinate system and the pixel coordinates of the upper left and lower right corners of the rectangular detection frame, as well as the category label corresponding to the identified object are output in text file format.

[0070] Step 5: According to the recognition result of step 4, the spatial position information of each mining activity characteristic category in the geographic coordinate system is calculated to monitor and accurately locate the ionic rare earth mining behavior;

[0071] In this step, according to the identification results of the sedimentation tank and the high-level tank obtained in step 4, the spatial position information of the sedimentation tank and the high-level tank in the geographic coordinate system is calculated, wherein the coordinate calculation method of the sedimentation tank and the high-level tank in the geographic coordinate system is as follows:

[0072] The pixel coordinates of the center points of the identified sedimentation pool and high-level pool generated by the improved YOLOv8 target detection model are converted into the WGS_1984_UTM_zone_47N plane coordinate system coordinates representing the real space position based on the pixel coordinate system of the image slice; the conversion formula between pixel coordinates and plane coordinates is:

[0073]

[0074]

[0075] In the formula, , Represents the pixel coordinates of the center point of the detection box; , Indicates the pixel coordinates of the upper left corner of the detection box; , Indicates the pixel coordinates of the lower right corner of the detection box; , express The pixel coordinates of the point; , express The WGS_1984_UTM_zone_47N plane coordinates of the point; , Represents the WGS_1984_UTM_zone_47N plane coordinate of the upper left corner of the slice; , It represents the spatial resolution of the remote sensing image in the horizontal and vertical directions, which is 0.284 (unit: m) in this embodiment.

[0076] By obtaining the coordinates of the sedimentation tank and the high-level tank in real space through the above method, the ionic rare earth mining behavior can be monitored and accurately located.

[0077] The embodiment of the present invention also provides an ionic mining area estimation method based on remote sensing images. The method uses a Segment Anything segmentation model to perform local segmentation of the recognition result based on the detection frame based on the recognition result obtained by the above positioning, so as to perform pixel-level segmentation on each mining activity feature category, and finally, based on the pixel-level segmentation result, the area of ​​each mining activity feature category is estimated;

[0078] In the embodiment, based on the sedimentation pool and high-level pool identified by the above-mentioned YOLOv8 target detection model, the sedimentation pool and high-level pool inside the detection frame are segmented at the pixel level, and the method is as follows:

[0079] 1) Load a specific segmentation model from the "Segment Anything" model registry based on a predefined or user-specified model type and checkpoint path;

[0080] 2) The input of the model is a directory containing image slices and corresponding detection result text files. For each image slice, the corresponding detection result is read from the text file, including the detection box and label of the identified object in the image slice;

[0081] 3) Use the "Segment Anything" model to segment the image slices within the detection box of the identified object; each segmented region is then converted into a binary mask, which is vectorized into polygons using the vectorize_mask function, which uses raster geometry to convert pixel-based masks into vectorized shapes. Each polygon is assigned a label based on the detection result, creating a labeled vector dataset;

[0082] 4) Use the GeoPandas package to compile the vectorized and labeled polygons into a GeoDataFrame, which includes the original coordinate system of the image slices, namely the WGS_1984_UTM_zone_47N plane coordinate system; then save it as a Shapefile file in the specified path, making the vector data accessible to GIS applications;

[0083] Through the above operations, the Segment Anything segmentation model is implemented to perform pixel-level segmentation on the main recognition object only within the detection frame, and the corresponding category label is assigned to the segmentation result according to the category label to which the detection frame belongs, and then the pixel-level segmentation results of the sedimentation pool and the high-level pool are output in .shp format.

[0084] After pixel-level segmentation of the sedimentation tank and the high-level pool, the area of ​​the sedimentation tank and the high-level pool is estimated as follows:

[0085] Load a specific segmentation model from the "Segment Anything" model registry based on predefined or user-specified model type and checkpoint path;

[0086] Read the image slices and use OpenCV to convert the image from BGR format to RGB format to adapt to the model input requirements. Use the "Segment Anything" model to segment the image slices within the detection box of the identified object. For each segmented object, calculate its pixel area and convert it to the actual area using the given spatial resolution. The area calculation formula is:

[0087] ;

[0088] In the formula, It is the area of ​​sedimentation tank or high-level tank in the detection frame; is the number of pixels of the segmentation result in the detection frame; , It represents the spatial resolution of the remote sensing image in the horizontal and vertical directions, which is 0.284 (unit: m) in this embodiment.

[0089] After calculating the area of ​​the sedimentation tank or high-level pool within the detection frame, the area is accumulated and counted according to the category of the identified object to obtain the total area of ​​the sedimentation tank and the total area of ​​the high-level pool in the study area. Then, the pastemask2img function is used to superimpose the segmentation mask on the original image and mark the detection frame and calculated area of ​​each identified object. Through the above operations, the refined extraction of sedimentation tanks and high-level pools and the visual display of area estimation results are achieved.

[0090] The embodiment of the present invention further provides a system for implementing the above-mentioned ionic rare earth mining area monitoring method based on remote sensing images, comprising:

[0091] A data acquisition and processing module is used to acquire remote sensing image data including the target detection area and pre-process the remote sensing image;

[0092] A data set making module is used to make the remote sensing images preprocessed by the data acquisition processing module into a characteristic identification data set of ionic rare earth mining activities for target detection;

[0093] The model building and training module is used to build a YOLOv8 target detection model, and improve the YOLOv8 target detection model based on the characteristics of ionic rare earth mining activities. The improved YOLOv8 target detection model is trained using the ionic rare earth mining activity feature identification dataset produced by the dataset production module, and the trained YOLOv8 target detection model is used to identify the feature categories of ionic rare earth mining activities.

[0094] The monitoring and positioning module is used to calculate the spatial position information of each mining activity characteristic category in the geographic coordinate system based on the recognition results of the model construction and training modules, so as to monitor and accurately locate the ionic rare earth mining behavior.

[0095] Compared with the prior art, the present invention cleverly selects sedimentation ponds and high-level ponds, which are characteristic features of ionic rare earth mining areas, as mining activity identifiers and remote sensing image feature interpretation marks in ionic rare earth mining areas, and uses high-resolution remote sensing images to produce a dataset of characteristic identifiers of ionic rare earth mining activities. According to the characteristics of sedimentation ponds and high-level ponds in remote sensing images, the YOLOv8 target detection model is improved in a targeted manner to improve the model's detection performance for targets. The trained model is used to identify sedimentation ponds and high-level ponds in the study area, and their spatial position information is calculated to achieve monitoring and accurate positioning of ionic rare earth mining behavior. A deep learning strategy combining object-level target detection and pixel-level segmentation is adopted, and the Segment Anything model is used to perform pixel-level segmentation on the main objects inside the detection frame, so as to achieve refined extraction and area estimation of sedimentation ponds and high-level ponds. It is helpful to continuously monitor the mining activities in ionic rare earth mining areas and trace back the evolution process and development trend of mining areas.

[0096] The above are only preferred embodiments of the present invention, and are not intended to limit the implementation methods and protection scope of the present invention. Those skilled in the art should be aware that all solutions obtained by equivalent substitutions and obvious changes made using the contents of the specification of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for monitoring ionic rare earth mining areas based on remote sensing images, characterized in that: The steps include: Step 1: Obtain remote sensing image data containing the target detection area and preprocess the remote sensing image; Step 2, preparing the remote sensing image preprocessed in step 1 into a data set of ionic rare earth mining activity feature identification for target detection; Step 3: Build the YOLOv8 target detection model and improve it; Step 4: Use the data set prepared in step 2 to train the improved YOLOv8 target detection model, and use the trained YOLOv8 target detection model to identify the characteristic categories of ionic rare earth mining activities; Step 5: According to the recognition result of step 4, the spatial position information of each mining activity characteristic category in the geographic coordinate system is calculated to monitor and accurately locate the ionic rare earth mining behavior; The method for improving the YOLOv8 target detection model in step 3 is as follows: (1) The Asymptotic Feature Pyramid Network structure is introduced to improve the neck network structure of the original YOLOv8 target detection model to integrate high-level semantic information with low-level spatial position information; (2) Integrate the Convolutional Block Attention Module into the neck network structure of the original YOLOv8 target detection model; (3) The Slicing Aided Hyper inference module is introduced in the model inference phase. This module divides the image to be inferred into M*N slices with overlapping parts, performs target detection on each slice, sets the intersection-over-union ratio threshold, uses non-maximum suppression to eliminate redundant overlapping detection frames, and finally merges the inference results back to the original image size; The specific methods of introducing the progressive feature pyramid network structure to improve the neck network structure of the original YOLOv8 target detection model include: Connect the third CSPLayer_2Conv module and the last SPPF module in the feature extraction network structure to multiple ASFF2 modules, and then connect the multiple ASFF2 modules to the upsampling structure including multiple consecutive CSPLayer_2Conv modules respectively, and connect the second CSPLayer_2Conv module in the feature extraction network structure to the last CSPLayer_2Conv module in the upsampling structure; then connect the CSPLayer_2Conv module in the upsampling structure to the ASFF3 module, and finally connect the multiple ASFF3 modules to the downsampling structure including multiple consecutive CSPLayer_2Conv modules respectively; The specific method of integrating the convolutional block attention module into the neck network structure of the original YOLOv8 target detection model includes: Connecting multiple CSPLayer_2Conv modules in a downsampling structure including multiple consecutive CSPLayer_2Conv modules to multiple CBMA attention modules respectively, and then connecting the multiple CBMA attention modules to Dectect detection head modules for different target sizes respectively; The method for calculating the spatial location information of each mining activity characteristic category in the geographic coordinate system includes: The pixel coordinate system coordinates of the center point of the identified object generated by the improved YOLOv8 target detection model are converted into the plane coordinate system coordinates representing the real space position based on the pixel coordinate system coordinates of the image slice, where the conversion formula between pixel coordinates and plane coordinates is: In the formula, , Represents the pixel coordinates of the center point of the detection box; , Indicates the pixel coordinates of the upper left corner of the detection box; , Indicates the pixel coordinates of the lower right corner of the detection box; , express The pixel coordinates of the point; , express Plane coordinates of a point; , Represents the plane coordinates of the upper left corner of the slice; , Indicates the spatial resolution of the remote sensing image in the horizontal and vertical directions.

2. The remote sensing image-based ionic rare earth mining area monitoring method according to claim 1, characterized in that: The method for making the data set in step 2 is: Convert remote sensing images from geographic coordinate system to plane coordinate system; The remote sensing images are cut into image slices, and all sedimentation tanks and high-level tanks are manually marked on the slices and used as positive samples. At the same time, a certain number of non-sedimentation tanks and non-high-level tanks are marked and used as negative samples. The data enhancement method is used to expand the positive and negative samples, and finally all samples are randomly divided into training set, validation set and test set according to a certain proportion.

3. A method for estimating the area of ​​ionic rare earth mining areas based on remote sensing images, characterized in that: Based on the recognition results of the remote sensing image-based ionic rare earth mining area monitoring method described in any one of claims 1-2, the SegmentAnything segmentation model is used to perform local segmentation based on the detection frame to perform pixel-level segmentation on each mining activity feature category, and finally, based on the pixel-level segmentation results, the area of ​​each mining activity feature category is estimated.

4. The method for estimating the area of ​​ionic rare earth mining areas based on remote sensing images according to claim 3 is characterized in that: Pixel-level segmentation methods include: Crop the remote sensing image into image slices, input the image slices into the Segment Anything segmentation model, and for each image slice, read the detection box and label of the identified object in the image slice; Use the Segment Anything segmentation model to segment the image slices based on the object's detection box, convert each segmented area into a binary mask, and vectorize the binary mask into polygons through the vectorize_mask function. Each polygon is assigned a label based on the detection result. Compile the vectorized and labeled polygons into a GeoDataFrame, including the original coordinate system of the image tiles, and save it.

5. The method for estimating the area of ​​ionic rare earth mining areas based on remote sensing images according to claim 3 is characterized in that: The area estimation method is as follows: Based on the pixel-level segmentation results, the segmented image is converted from BGR format to RGB format. For each segmented identified object, its pixel area is calculated and converted into the actual area through the given spatial resolution. The area calculation formula is: In the formula, is the area of ​​each identified object in the detection frame; is the number of pixels of the segmentation result within the detection frame; , Correspondingly, it represents the spatial resolution of the remote sensing image in the horizontal and vertical directions; After calculating the area of ​​each identified object in the detection frame, the areas are accumulated and counted according to the categories of the identified objects to obtain the total area of ​​each characteristic category of mining activities in the study area. Then, the pastemask2img function is used to superimpose the segmentation mask on the original image and mark the detection frame and calculated area of ​​each identified object.

6. A system for implementing the remote sensing image-based ionic rare earth mining area monitoring method according to any one of claims 1 to 2, characterized in that: include: A data acquisition and processing module is used to acquire remote sensing image data including the target detection area and pre-process the remote sensing image; A data set making module, used to make the remote sensing images pre-processed by the data acquisition processing module into a characteristic identification data set of ionic rare earth mining activities for target detection; The model building and training module is used to build a YOLOv8 target detection model, and improve the YOLOv8 target detection model based on the characteristics of ionic rare earth mining activities. The improved YOLOv8 target detection model is trained using the ionic rare earth mining activity feature identification dataset produced by the dataset production module, and the trained YOLOv8 target detection model is used to identify the feature categories of ionic rare earth mining activities. The monitoring and positioning module is used to calculate the spatial position information of each mining activity characteristic category in the geographic coordinate system based on the recognition results of the model construction and training modules, so as to monitor and accurately locate the ionic rare earth mining behavior.

Images