Remote sensing mine automatic labeling method based on deep learning and shallow feature statistics

Abstract

The present application provides a remote sensing mining area automatic labeling method based on deep learning and shallow feature statistics, comprising: obtaining a remote sensing image sample set; taking the remote sensing sample image as the network source input, and the corresponding benchmark mining area module segmented image as the output; combining a preset loss function and evaluation index, applying each remote sensing sample image, training the target deep learning image segmentation model, and obtaining the benchmark mining area module segmentation model; obtaining the to-be-labeled remote sensing image, inputting the to-be-labeled remote sensing image into the benchmark mining area module segmentation model, and obtaining the mining area module segmented image; according to the mining area module segmented image, the object features of the mining area map patch are counted, and the corresponding ground object category is obtained; taking the mining area module segmented image as the basis, the mining area map patch with the ground object category being the mineral category is retained, and the remote sensing mining area labeling result is obtained. The present application introduces deep learning technology, innovatively combines the shallow feature statistical method, and provides strong technical support for resource management and environmental protection.

Description

Technical Field

[0001] This invention relates to the field of mining area identification technology, and in particular to an automatic labeling method for remote sensing mining areas based on deep learning and shallow feature statistics. Background Technology

[0002] Remote sensing technology has important applications in geological exploration and resource management, and the automatic identification and labeling of mining areas is a key task of remote sensing technology. Accurate labeling of mining areas is of great significance for resource management and environmental protection. Through accurate identification and monitoring of mining areas, governments and enterprises can better plan resource development and avoid resource waste and environmental damage caused by disorderly mining. At the same time, real-time monitoring of changes in mining areas can detect illegal mining activities and potential environmental risks early, allowing for timely countermeasures. Traditional mining area identification methods rely on manual labeling or simple machine learning algorithms, which often face problems of inefficiency and insufficient accuracy when dealing with complex and large-scale data. Summary of the Invention

[0003] Therefore, it is necessary to provide an automatic annotation method for remote sensing mining areas based on deep learning and shallow feature statistics to address the aforementioned technical problems.

[0004] An automatic annotation method for remote sensing mining areas based on deep learning and shallow feature statistics includes the following steps: Obtain a given set of remote sensing image samples; wherein, the set of remote sensing image samples includes: remote sensing sample images and segmented images of a benchmark mining area module; Using each remote sensing sample image as the network source input and the corresponding benchmark mining area module segmentation image as the output, combined with the preset loss function and evaluation index, the target deep learning image segmentation model is trained using each remote sensing sample image to obtain the benchmark mining area module segmentation model. A remote sensing image to be labeled is acquired, and the image is input into the baseline mining area module segmentation model to obtain a mining area module segmentation image; wherein, the mining area module segmentation image includes several mining area patches; Based on the object features of the segmented images of the mining area module, the corresponding land cover categories are obtained; wherein, the object features include: exponential features and geometric features; the land cover categories include: mineral categories and non-mineral categories; Based on the segmented image of the mining area module, mining area patches with land cover category of mineral are retained to obtain remote sensing mining area annotation results.

[0005] In one embodiment, the process further includes obtaining a given set of remote sensing image samples before: Acquire remote sensing images; By receiving the annotation information, the baseline mining area module segmentation image corresponding to the remote sensing image is obtained; The benchmark mining area module segmentation image is segmented according to a preset segmentation step size to obtain an effectively labeled image; The valid labeled images are filtered and their formats are standardized to obtain filtered labeled images; The selected and labeled images are randomly flipped, randomly cropped, and randomly blurred in color to obtain remote sensing sample images.

[0006] In one embodiment, the benchmark mining area module segmentation image is segmented according to a preset segmentation step size to obtain an effectively labeled image, including: The image segmentation of the benchmark mining area module is segmented using a preset cutting step size to obtain a segmented shp file; Based on the cut shapefile, coordinate transformation is performed using the following formula to obtain the effectively annotated image: , , in, To effectively label the x-coordinate of the image, This represents the x-coordinate of the shapefile being cut. This represents the x-coordinate of the top-left vertex of the image. Indicates the spatial resolution of pixels. To effectively label the ordinate of the image, The ordinate of the top left corner vertex of the image. This represents the vertical axis of the cut shapefile.

[0007] In one embodiment, the target deep learning image segmentation model includes: The system consists of a feature extraction backbone network, a neighbor feature aggregation module, a feature enhancement module, and a decoding module.

[0008] In one embodiment, acquiring a remote sensing image to be labeled, and inputting the remote sensing image to be labeled into the baseline mining area module segmentation model to obtain a mining area module segmentation image includes: A remote sensing image to be labeled is acquired and input into the benchmark mining area module segmentation model. The benchmark mining area module segmentation model receives the input remote sensing image to be labeled and extracts preliminary image features through a feature extraction backbone network to obtain features at five stages with different depths and sizes. The features are input into the neighbor feature aggregation module, and the features of adjacent stages are aggregated by the neighbor feature aggregation module to obtain the first aggregated feature; The first aggregated feature is input into the feature enhancement module. The feature enhancement module obtains the superimposed features of different receptive fields through the superposition of dilated convolutions, and aggregates the superimposed features together through residual connections to obtain the second aggregated feature. The second aggregated feature is input into the decoding module, which restores the second aggregated feature to the size of the input remote sensing image to be labeled, thus obtaining the mining area module segmentation image.

[0009] In one embodiment, based on the statistical features of the segmented image from the mining area module, the corresponding land cover categories are obtained, including: Obtain mining area patches from the segmented image of the mining area module, and calculate the exponential and geometric features of each mining area patch; wherein, the exponential features include: normalized vegetation index, normalized water index, BG, BR, and GR; the geometric features include: area and aspect ratio; Based on the index features and geometric features, and according to the statistical results of the preset features, the corresponding land cover category is obtained.

[0010] In one embodiment, mining area patches are obtained from the segmented image of the mining area module, and the exponential and geometric features of each mining area patch are calculated; wherein, the exponential features include: normalized vegetation index, normalized water index, BG, BR, and GR; the geometric features include: area and aspect ratio, including: The normalized vegetation index is calculated using the following formula: , in, Represents the normalized vegetation index. This represents the near-infrared mean value of the mining area patch. Indicates the mean value of the red band; The normalized water index is calculated using the following formula: , in, Indicates the normalized water quality index. Indicates the mean value of the green band. This represents the near-infrared average value of a patch in the mining area. The exponential relationship between the blue and red bands in the mining area map is calculated using the formula shown below: , in, This indicates the exponential relationship between the blue and red bands in a mining area map. Indicates the blue band mean. Indicates the mean value of the red band; The exponential relationship between the green and red bands in the mining area map is calculated using the formula shown below: , in, This indicates the exponential relationship between the green and red bands in a mining area map. Indicates the mean value of the green band. Indicates the mean value of the red band; The exponential relationship between the blue and green bands in the mining area map is calculated using the formula shown below: , in, This indicates the exponential relationship between the blue and green bands in a mining area map. Indicates the blue band mean. Indicates the mean value of the green band; The area of ​​a mining zone patch is calculated using the following formula: , in, Indicates the area of ​​the mining area. Indicates the first The actual area of ​​a pixel n Indicates the number of pixels contained in a mining area patch; The aspect ratio of a mining area patch is calculated using the following formula: , in, Indicates the aspect ratio of the mining area patch. Indicates length, Indicates width, Indicates the area of ​​the mining area. This indicates the length of the bounding rectangle. This indicates the width of the rectangle surrounding the border.

[0011] An automatic annotation system for remote sensing mining areas based on deep learning and shallow feature statistics is provided to implement the automatic annotation method for remote sensing mining areas based on deep learning and shallow feature statistics as described above, including: An apparatus includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the remote sensing mining area automatic labeling method based on deep learning and shallow feature statistics described in the above embodiments.

[0012] A storage medium storing a computer program that, when executed by a processor, implements the steps of the remote sensing mining area automatic labeling method based on deep learning and shallow feature statistics described in the above embodiments.

[0013] Compared to existing technologies, the advantages and beneficial effects of this invention are as follows: By developing an automatic remote sensing mining area labeling technology based on deep learning and shallow feature statistics, this invention can achieve efficient and accurate monitoring of mining areas. This not only helps management departments to promptly grasp the dynamics of mining areas and scientifically formulate resource development and environmental protection policies, but also helps enterprises optimize production management, improve resource utilization efficiency, and reduce environmental damage. Furthermore, real-time mining area monitoring technology can promote public awareness and participation in mining development and environmental protection, jointly promoting the sustainable use of resources. Accurate mining area labeling and monitoring can provide important data support for environmental impact assessments, helping to evaluate the short-term and long-term environmental impacts of mining development and providing a scientific basis for the formulation and implementation of environmental protection measures. Simultaneously, real-time mining area monitoring technology can promptly detect and warn of environmental risks, such as pollution of water bodies and soil around mining areas, prompting relevant departments to take timely countermeasures to reduce environmental risks.

[0014] This invention introduces deep learning technology and innovatively combines it with shallow feature statistical methods to improve the accuracy and efficiency of automatic labeling of remote sensing mining areas, thereby promoting the development of remote sensing technology, providing strong technical support for resource management and environmental protection, and promoting the sustainable use of resources. It not only has significant academic value but also remarkable social and economic implications. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating an automatic annotation method for remote sensing mining areas based on deep learning and shallow feature statistics in one embodiment. Figure 2 This is a comparison diagram of the actual open-pit mine area in one embodiment and the open-pit mine area referred to in this project; Figure 3 This is a schematic diagram illustrating the sample flipping effect in one embodiment; Figure 4 This is a schematic diagram illustrating the random cropping effect in one embodiment; Figure 5 This is a schematic diagram of a random color blur effect in one embodiment; Figure 6 This is a schematic diagram of a confusion matrix in one embodiment; Figure 7 This is a schematic diagram of the segmentation model structure of a benchmark mining area module in one embodiment; Figure 8 This is a schematic diagram of the neighbor feature aggregation module structure in one embodiment; Figure 9 This is a schematic diagram of the feature enhancement module structure in one embodiment; Figure 10 This is a schematic diagram of the decoding module structure in one embodiment; Figure 11A comparison chart showing the visualization results of a benchmark mining area module segmentation model in one embodiment; Figure 12 A comparison chart showing the visualization results of another benchmark mining area module segmentation model in one embodiment; Figure 13 This is a comparison chart of the results of a comparative experiment in one embodiment; Figure 14 This is a schematic diagram illustrating the segmentation visualization result of a benchmark mining area module segmentation model in one embodiment. Figure 15 This is a schematic diagram of the visualization results after the benchmark mining area module segmentation model is combined with shallow feature filtering in one embodiment. Figure 16 This is a schematic diagram of the structure of an automatic annotation system for remote sensing mining areas based on deep learning and shallow feature statistics in one embodiment. Figure 17 This is a schematic diagram of the internal structure of the device in one embodiment. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings.

[0017] It should be noted that, unless otherwise defined, the technical or scientific terms used in one or more embodiments of this specification should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in one or more embodiments of this specification do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word covers the element or object listed following the word and its equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0018] In one embodiment, such as Figure 1 As shown, an automatic annotation method for remote sensing mining areas based on deep learning and shallow feature statistics is provided, including the following steps: Step S101: Obtain a given remote sensing image sample set; wherein the remote sensing image sample set includes: remote sensing sample images and benchmark mining area module segmentation images.

[0019] Specifically, the constructed remote sensing image sample set is obtained, including remote sensing sample images and corresponding baseline mining area module segmentation images. For example, given a multi-source imagery open-pit mine basic sample library containing 3979 pairs of samples, each pair of samples contains a 256x256 remote sensing sample image and the binary mask image of the corresponding mining area (baseline mining area module segmentation image), providing rich baseline data reference for automated mine remote sensing identification.

[0020] Based on this, the process of obtaining a given remote sensing image sample set also includes: Acquire remote sensing images; By receiving the annotation information, the baseline mining area module segmentation image corresponding to the remote sensing image is obtained; The benchmark mining area module segmentation image is segmented according to a preset segmentation step size to obtain an effectively labeled image; The valid labeled images are filtered and their formats are standardized to obtain filtered labeled images; The selected and labeled images are randomly flipped, randomly cropped, and randomly blurred in color to obtain remote sensing sample images.

[0021] Specifically, because open-pit mines have low coverage in single-scene remote sensing images, meaning they are sparse in the images, directly using existing images for training would result in a high machine load and a low positive sample ratio (very low effective data ratio). Training with such imbalanced data would affect the accuracy of feature extraction and category classification. Therefore, a series of data preprocessing steps are needed to construct a remote sensing image sample set, including image annotation, effective region selection, mine area segmentation, and data augmentation.

[0022] First, remote sensing images are acquired. With the development of satellite hardware and software technology, domestic satellite imagery has gradually replaced foreign satellite imagery, becoming the mainstream data source in the field of natural resource supervision. This embodiment uses remote sensing images provided by domestic satellites Gaofen-1, Gaofen-2, Gaofen-6, Gaofen-7, Ziyuan-1, Ziyuan-3, and CBN-4. All images have undergone correction, registration, and fusion, and include four bands: red, green, blue, and near-infrared, with resolutions of 2m, 0.8m, 2m, 0.7m, 2m, and 2m respectively. The data depth is 8 bits, and the data format is TIFF. Data information is shown in Table 1.

[0023] Table 1. Overview of Data Source Information Gaofen-1 R / G / B / NIR 4-band 2m 8-bit TIFF Gaofen-2 R / G / B / NIR 4-band 0.8m 8-bit TIFF Gaofen-6 R / G / B / NIR 4-band 2m 8-bit TIFF Gaofen-7 R / G / B / NIR 4-band 0.7m 8-bit TIFF Resource No. 1 R / G / B / NIR 4-band 2m 8-bit TIFF Resource No. 3 R / G / B / NIR 4-band 2m 8-bit TIFF China-Pakistan No. 4 R / G / B / NIR 4-band 2m 8-bit TIFF The experiment selected successfully captured remote sensing images of Chongqing Municipality between 2021 and 2023 as the data source. The most common limestone types in Chongqing (including limestone for decorative purposes, limestone for building materials, limestone for cement, etc.), sandstone types (including flagstone, sandstone for building, sand for cement mixing, etc.), shale types (including shale for bricks and tiles, shale for expanded clay aggregate, etc.), and other types (including fluorite, calcite, etc.) open-pit mines were selected as the mine identification targets.

[0024] It should be noted that in actual work, open-pit mine areas or mining activity patches typically include a collection of features such as mining areas, industrial sites, solid waste, mine buildings, and mine roads. Some mines may only contain one or a few of these features. Considering the complexity of the feature combinations and the impact on identification difficulty, and that the main objective of regulatory work is to focus on information related to "mining activity areas," the open-pit mine areas or mining activity patches in this embodiment refer only to the main carriers that effectively reflect the mining situation—the mining area, and the industrial site and solid waste integrated with the mining area, excluding mine buildings or separate industrial sites and solid waste. Figure 2 This shows the object area range of the open-pit mine or mining activity patch in this project.

[0025] Manual annotation of the mining area is performed using ArcGIS software. This generates specific annotation information, which is then received to obtain a segmented image of the benchmark mining area corresponding to the remote sensing image. In one embodiment, reference points for open-pit mines issued by Chongqing Municipality are compared with the original remote sensing image for manual screening. Areas clearly identified as mining areas are interpreted using ArcGIS software to generate vector annotations. These annotations are then categorized according to mineral type (limestone, sandstone, shale) and entered into the map attribute fields. All corner points of the vector polygons are based on CGCS2000 national geodetic coordinates.

[0026] Subsequently, considering the sparsity of the mining area patches, the mining area was selected for image segmentation. The segmentation was performed with a preset segmentation step size of 50 pixels to generate 256×256 TIFF format image data and labeled polygon shp files. The images were then converted to generate slice images (TIFF) and category binary mask images (TIFF) to obtain effectively labeled images.

[0027] Subsequently, considering the large sample size and high similarity among the segments, effective labeled images of the slices were sampled and filtered. Initially, a step size of 50 was set during image segmentation. This increased the representation of smaller mining areas in the dataset. However, for larger mining areas, a step size of 50 resulted in low image feature change rates and excessively increased the proportion of these areas. Therefore, data sampling and filtering were conducted based on mine category number and mining area size to achieve a relatively balanced proportion of different mine types.

[0028] Because the resolution of Gaofen-1, Gaofen-6, Ziyuan-1, Ziyuan-3, and China-Brazil-4 remote sensing images is 2m, while the resolution of Gaofen-2 imagery is 0.8m and Gaofen-7 imagery is 0.7m, the resolutions are inconsistent. To facilitate subsequent shallow feature extraction, statistical analysis, and deep learning experiments, it is necessary to standardize the data format of the images and binary masks. Therefore, the data from Gaofen-2 and Gaofen-7 were resampled to unify the pixel size to 2m, resulting in the selected and labeled images.

[0029] Finally, due to the difficulty in obtaining remote sensing samples, in order to enrich the remote sensing image sample set, the selected and labeled images were randomly flipped, randomly cropped, and randomly blurred in color.

[0030] 1) Random flipping By randomly flipping the data horizontally and vertically, four scenarios can be achieved: no flipping, horizontal flipping, vertical flipping, and simultaneous horizontal and vertical flipping. This means that the dataset can be enriched by four times through these operations. The specific flipping effect is shown below. Figure 3 As shown.

[0031] 2) Random cropping Set a scaling range [0.8-1.2], where the minimum scaling ratio is 0.8 (20% reduction) and the maximum scaling ratio is 1.2 (20% enlargement). 1 represents no scaling. The image scaling ratio is controlled by a random number.

[0032] When enlarging an image, a cropping box generated by random numbers is used to crop the edges of the original image. After cropping, bilinear interpolation is used to restore the cropped image to its original size. When shrinking, the original image is scaled down proportionally. After shrinking, the edges of the shrunk image are padded to match the input size, restoring it to the original image size. Specific cropping effects are shown below. Figure 4 As shown.

[0033] 3) Random color blur The original image is blurred using a Gaussian blur method. A randomly generated floating-point number between 0 and 1 is used as the radius of the Gaussian blur filter. The blur radius determines the intensity of the blur; the larger the value, the more pronounced the blur effect. Specific color blur effects are shown below. Figure 5 As shown.

[0034] Based on this, the segmented image of the benchmark mining area module is segmented according to a preset cutting step size to obtain an effectively labeled image, including: The image segmentation of the benchmark mining area module is segmented using a preset cutting step size to obtain a segmented shp file; Based on the cut shapefile, coordinate transformation is performed using the following formula to obtain the effectively annotated image: , , in, To effectively label the x-coordinate of the image, This represents the x-coordinate of the shapefile being cut. This represents the x-coordinate of the top-left vertex of the image. Indicates the spatial resolution of pixels. To effectively label the ordinate of the image, The ordinate of the top left corner vertex of the image. This represents the vertical axis of the cut shapefile.

[0035] Specifically, since the coordinate system of the shapefile of the mine sample is a spatial geographic coordinate system, while the coordinates of the mask nodes in the generated binary image data are graphic coordinates, a coordinate transformation needs to be performed using a formula: , , in, To effectively label the x-coordinate of the image, This represents the x-coordinate of the shapefile being cut. , The coordinates of the top-left vertex of the image can be obtained from the XML file. Indicates the spatial resolution of pixels. To effectively label the ordinate of the image, This represents the vertical axis of the cut shapefile.

[0036] Step S102: Using each remote sensing sample image as the network source input and the corresponding benchmark mining area module segmentation image as the output, the target deep learning image segmentation model is trained by combining the preset loss function and evaluation index with each remote sensing sample image to obtain the benchmark mining area module segmentation model.

[0037] Specifically, the target deep learning image segmentation model OMSegNet is a semantic segmentation model for remote sensing open-pit mine areas based on CNN and Transformer. Training the target deep learning image segmentation model yields a benchmark mine area module segmentation model.

[0038] The performance of the benchmark mining area module segmentation model was evaluated using evaluation metrics. Image semantic segmentation is the classification of images at the pixel level. The evaluation of recognition effectiveness can be accurate to a single pixel. Therefore, the evaluation metrics selected for this model are the common evaluation standards for most image segmentation algorithms, which calculate the accuracy (OP), precision (Precision), F1 score (F1), and recall of the pixel classification results, with each pixel as the basic unit.

[0039] like Figure 6 As shown, each remote sensing image corresponds to one mask, and each category uses a binary mask. Positive indicates a predicted pixel as a mine, and Negative indicates a non-mine pixel. TP represents the number of true positives (correct predictions of mine pixels), FP represents the number of false positives (incorrect predictions of non-mine pixels), TN represents the number of true negatives (correct predictions of non-mine pixels), and FN represents the number of false negatives (incorrect predictions of mine pixels as non-mine pixels). Any pixel must belong to one of these four categories.

[0040] Precision represents the proportion of examples that are actually mines when they are classified as mines. It is used to evaluate the accuracy of pixels identified as mines in the generated results. Recall is a measure of coverage, measuring how many foreground pixels are found when multiple positive examples are classified as positive. Since precision and recall are interdependent, ideally both should be high. However, generally, high precision leads to low recall, and vice versa. To better compare network performance, the comprehensive performance evaluation metric F1 is introduced.

[0041] Based on this, target deep learning image segmentation models include: The system consists of a feature extraction backbone network, a neighbor feature aggregation module, a feature enhancement module, and a decoding module.

[0042] Specifically, the network structure of the target deep learning image segmentation model is as follows: Figure 7 As shown, it mainly consists of four modules: a feature extraction backbone network, a neighbor feature aggregation module, a feature enhancement module, and a decoding module.

[0043] The feature extraction backbone network extracts initial image features and, through continuous in-depth analysis, obtains features at five stages with different depths and sizes.

[0044] The neighbor feature aggregation module fuses the preliminary features extracted from the backbone network. In one embodiment, to achieve a lightweight design, the feature extraction backbone network uses MobileNet, which has significantly fewer parameters than mainstream backbone networks like ResNet and VGG. This module first unifies the dimensions of the three stages using pooling and bilinear interpolation, and then performs a concatenation operation to initially fuse the features from the three stages. The first aggregated feature then uses an attention mechanism to obtain its long-range dependencies, thus better aggregating the features from the three stages. The model structure is as follows: Figure 8 As shown.

[0045] The overall architecture of the feature enhancement module is as follows: Figure 9 As shown in the diagram, this module primarily uses dilated convolution to acquire features from different receptive fields, and then uses residual connections to fuse these features together to achieve feature enhancement.

[0046] In one embodiment, after receiving the first aggregated feature output by the neighboring feature enhancement module, the feature enhancement module first extracts features with a larger receptive field through a 3x3 dilated convolution with a dilation rate of 7. The output is then added to the features obtained from the original input features after a 1x1 convolution. Next, a 3x3 convolution with a dilation rate of 5 is performed, and this is fused with the features obtained from the original input features after a 1x1 convolution. This process is repeated to obtain the second aggregated feature. The dilation rate of each 3x3 dilated convolution is [7, 5, 3, 1].

[0047] After progressively deeper feature extraction, the first aggregated feature yields four second aggregated features at different scales. The shallower feature maps contain preliminary features such as lines, edges, and corners, while the deeper feature maps contain semantic features of the mining area. The decoding module fully integrates these second aggregated features at different scales, fusing them upwards from the deepest feature with the shallower ones, as shown in Figure 10. Accepting inputs from two different sets of second aggregated features, the deeper features are first upsampled to the size of the shallower features using bilinear interpolation, and then initially fused together through channel-level concatenation. Next, a 1x1 convolution is used to aggregate channel information, followed by a 7x7 depthwise separable convolution for spatial feature fusion. The fused results are processed using max pooling and average pooling at the channel level and max pooling and average pooling at the spatial level to obtain feature columns rich in channel information and feature surfaces rich in spatial information. After performing fully connected and convolutional operations on these features respectively, they are cross-multiplied to obtain two weight matrices. Finally, the two weight matrices are weighted and summed over the features at both scales to obtain a feature map that fully integrates feature information from two different scales. Then, the fusion operation is performed sequentially from bottom to top to obtain the final segmented binary image, i.e., the mining area module segmentation image.

[0048] Step S103: Obtain the remote sensing image to be labeled, and input the remote sensing image to be labeled into the benchmark mining area module segmentation model to obtain the mining area module segmentation image; wherein, the mining area module segmentation image includes several mining area patches.

[0049] Specifically, the remote sensing image to be labeled is input into the benchmark mining area module segmentation model to obtain a mining area module segmentation image that includes multiple mining area patches.

[0050] Based on this, a remote sensing image to be labeled is obtained, and the image is input into the baseline mining area module segmentation model to obtain a mining area module segmentation image, including: A remote sensing image to be labeled is acquired and input into the benchmark mining area module segmentation model. The benchmark mining area module segmentation model receives the input remote sensing image to be labeled and extracts preliminary image features through a feature extraction backbone network to obtain features at five stages with different depths and sizes. The features are input into the neighbor feature aggregation module, and the features of adjacent stages are aggregated by the neighbor feature aggregation module to obtain the first aggregated feature; The first aggregated feature is input into the feature enhancement module. The feature enhancement module obtains the superimposed features of different receptive fields through the superposition of dilated convolutions, and aggregates the superimposed features together through residual connections to obtain the second aggregated feature. The second aggregated feature is input into the decoding module, which restores the second aggregated feature to the size of the input remote sensing image to be labeled, thus obtaining the mining area module segmentation image.

[0051] Specifically, after receiving an input remote sensing image to be labeled, the model extracts preliminary image features through a feature extraction backbone network (MobileNet), and continuously refines the extraction process to obtain features at five different depths and sizes. Then, a neighbor feature aggregation module aggregates features from adjacent stages to achieve preliminary feature enhancement. These enhanced features are then input into a feature enhancement module, which uses dilated convolutions to obtain features from different receptive fields and aggregates them together through residual connections. Finally, a decoding module restores the image to the size of the input remote sensing image to obtain the final segmentation result, resulting in the segmented image of the mining area module.

[0052] Step S104: Based on the object features of the segmented image of the mining area module, the land cover category is obtained; wherein, the object features include: exponential features and geometric features; the land cover category includes: mineral category and non-mineral category.

[0053] Specifically, taking mining area patches as the object and their characteristics as the core, the calculation of these characteristics identifies the land cover category to which the object belongs. Object characteristics include index features and geometric features; land cover categories include mineral categories and non-mineral categories. Index features mainly refer to custom-defined relevant feature indices reflected in high-resolution imagery, used to describe related land cover, and are highly specific and reflective of designated land cover. Examples include the Normalized Difference Vegetation Index (NDVI), which is sensitive to green land cover and is typically used to reflect vegetation cover; the Normalized Difference Water Index (NDWI), which is sensitive to bluish-green land cover and is typically used to reflect water cover; and the Normalized Building Shape Index (NHSI), which is sensitive to regular shapes and is typically used to reflect buildings, etc.

[0054] Based on this, according to the statistical features of the segmented images in the mining area module, the corresponding land cover categories are obtained as follows: Obtain mining area patches from the segmented image of the mining area module, and calculate the exponential and geometric features of each mining area patch; wherein, the exponential features include: normalized vegetation index, normalized water index, BG, BR, and GR; the geometric features include: area and aspect ratio; Based on the index features and geometric features, and according to the statistical results of the preset features, the corresponding land cover category is obtained.

[0055] Specifically, the index features of image targets describe the relevant characteristics of a specific feature through spectral, geometric, and other features, and are often used as the main features for extracting the corresponding feature. Commonly used index features include Normalized Difference Vegetation Index (NDVI), Normalized Difference Water Index (NDWI), and Normalized Building Shape Index (NHSI). Since vegetation and water bodies are the most common features near mining areas, while buildings are less common, this embodiment selects Normalized Difference Vegetation Index (NDVI), Normalized Difference Water Index (NDWI), BG, BR, and GR as index feature quantification indicators.

[0056] Geometric features reflect the geometric and morphological information of ground features as they appear on an image. An image object is a set of adjacent pixels with similar gray levels, which naturally reflects the geometric shape characteristics of the ground feature on a two-dimensional image.

[0057] Based on the knowledge base of shallow features of common open-pit mining activity patches, 15 categories of shallow feature values ​​(spectral, textural, index, and geometric features) were statistically analyzed according to mineral type. This yielded the screening ranges for 15 categories of shallow features for 28 mineral types, resulting in preset feature statistics. These preset feature statistics serve as the basis for screening mining area patches. Specifically, the index feature screening range for all mineral types is the average range of NDVI, NDWI, BG, BR, and GR in the shallow feature screening table. The geometric feature screening range is uniformly set according to area A≥300 and aspect ratio 0.2≤LW≤5. The other 12 feature categories are subdivided for different mineral types, as detailed in the attached table. Table 2 shows the screening range settings for the 12 categories of shallow features for some mineral types (limestone and sandstone).

[0058] Table 2 Screening Table of Shallow Features of Limestone and Sandstone Patches

[0059] Based on this, mining area patches are obtained from the segmented image of the mining area module, and the exponential and geometric features of each mining area patch are calculated; wherein, the exponential features include: normalized vegetation index, normalized water index, BG, BR, and GR; the geometric features include: area and aspect ratio, including: The normalized vegetation index is calculated using the following formula: , in, Represents the normalized vegetation index. This represents the near-infrared mean value of the mining area patch. Indicates the mean value of the red band; The normalized water index is calculated using the following formula: , in, Indicates the normalized water quality index. Indicates the mean value of the green band. This represents the near-infrared average value of a patch in the mining area. The exponential relationship between the blue and red bands in the mining area map is calculated using the formula shown below: , in, This indicates the exponential relationship between the blue and red bands in a mining area map. Indicates the blue band mean. Indicates the mean value of the red band; The exponential relationship between the green and red bands in the mining area map is calculated using the formula shown below: , in, This indicates the exponential relationship between the green and red bands in a mining area map. Indicates the mean value of the green band. Indicates the mean value of the red band; The exponential relationship between the blue and green bands in the mining area map is calculated using the formula shown below: , in, This indicates the exponential relationship between the blue and green bands in a mining area map. Indicates the blue band mean. Indicates the mean value of the green band; The area of ​​a mining zone patch is calculated using the following formula: , in, Indicates the area of ​​the mining area. Indicates the first The actual area of ​​a pixel n Indicates the number of pixels contained in a mining area patch; The aspect ratio of a mining area patch is calculated using the following formula: , in, Indicates the aspect ratio of the mining area patch. Indicates length, Indicates width, Indicates the area of ​​the mining area. This indicates the length of the bounding rectangle. This indicates the width of the rectangle surrounding the border.

[0060] Specifically, the object characteristics are calculated using the following formula.

[0061] 1) Normalized Difference Vegetation Index (NDVI) In high-resolution remote sensing images, vegetation can usually be extracted using various vegetation indices composed of red and near-infrared bands, with the Normalized Difference Vegetation Index (NDVI) being a commonly used one. , Represents the normalized vegetation index. This represents the near-infrared mean value of the mining area patch. This represents the mean value of the red band.

[0062] 2) Normalized Difference Water Index Normalized Difference Water Index (NDWI) is a measure of water quality using the green band of remote sensing. G ) and near-infrared band ( NIR After interpolation processing, water body information is extracted from remote sensing images with good results. The specific calculation formula is as follows: , Indicates the normalized water quality index. Indicates the mean value of the green band. This represents the near-infrared average value of the mining area patch.

[0063] 3) BR Index The BR index calculates the relationship between the blue (B) and red (R) bands in a mining area map. The specific calculation formula is as follows: , This indicates the exponential relationship between the blue and red bands in a mining area map. Indicates the blue band mean. This represents the mean value of the red band.

[0064] 4) GR Index Similar to the BR index, the GR index is used to calculate the exponential relationship between the green band (G) and the red band (R). The specific calculation formula is as follows: , This indicates the exponential relationship between the green and red bands in a mining area map. Indicates the mean value of the green band. This represents the mean value of the red band.

[0065] 5) BG Index Similar to the BR and GR indices, the BG index is used to calculate the exponential relationship between the green band (G) and the blue band (B). The specific calculation formula is as follows: , This indicates the exponential relationship between the blue and green bands in a mining area map. Indicates the blue band mean. This represents the mean value of the green band.

[0066] 6) Area (A) The area A of an image object refers to the sum of the areas of the pixels that make up the image object. If the image does not have geographic coordinates, the area of ​​the object is the number of pixels that make up the image object. The specific calculation formula is as follows: , Indicates the area of ​​the mining area. Indicates the first The actual area of ​​a pixel n This indicates the number of pixels contained in a mining area patch. 7) Aspect Ratio (Length / Width, LW) LW, or the length-to-width ratio of an image object, is typically used to distinguish linear features. The aspect ratio of an image object can usually be approximated by its bounding box. , Indicates the aspect ratio of the mining area patch. Indicates length, Indicates width, Indicates the area of ​​the mining area. l 0 , w 0 The length and width of the bounding rectangle.

[0067] Step S105: Based on the segmented image of the mining area module, retain the mining area patches whose land cover category is mineral, and obtain the remote sensing mining area annotation results.

[0068] Specifically, the mining area module segmentation image is combined with the land cover category. Based on the mining area module segmentation image, the mining area patches with the land cover category of mineral are retained to obtain the remote sensing mining area annotation results.

[0069] To verify the effectiveness of the benchmark mining area module segmentation model, visualization and quantitative comparison of evaluation metrics were conducted. The images used for visualization were all from the test instrument, and the most comprehensive evaluation metric values ​​were derived from the average of all remote sensing sample images in the test set.

[0070] like Figure 11 and Figure 12 As shown in the figure, the first row of images represents the remote sensing sample image data used as input to the benchmark mining area segmentation model in the test set. The second row (P) represents the prediction results after processing by the benchmark mining area segmentation model, and the third row (GT) represents the manually labeled real labels of the mining areas in the dataset. From the above images, it can be seen that the prediction results processed by the benchmark mining area segmentation model are quite close to the manually labeled labels, proving the effectiveness of the benchmark mining area segmentation model.

[0071] To verify the effectiveness of the benchmark mining area segmentation model, a quantitative evaluation was conducted by comparing its accuracy with commonly used deep learning image segmentation models such as UNet, SegNet, ST-Unet, and Easy-Net on a mining data test set. To reflect the overall recognition performance of various models on mining area patches, the mining area patches were treated as a whole, without distinguishing between mineral types; that is, recognition and evaluation were performed only based on "mining area patches" and "non-mining area patches." The calculation results of various model indicators are shown in Table 3.

[0072] Table 3 Comparison of overall recognition accuracy of several segmentation methods OP 0.8224 0.9423 0.9371 0.9431 0.9733 0.9788 0.9834 Precision 0.6509 0.8015 0.7718 0.8014 0.8532 0.9023 0.9033 Recall 0.5475 0.6048 0.5931 0.6182 0.8825 0.8564 0.9063 F1 0.5947 0.6573 0.6336 0.6662 0.8712 0.8234 0.9038 As shown in Table 3, the method based on shallow feature knowledge base performs the worst in all metrics, lagging significantly behind other deep learning methods in almost all metrics. However, the differences in metrics among these deep learning methods are relatively small. Methods using only shallow feature statistics combined with traditional segmentation and classification are far less accurate than deep learning methods. The OMSegNet method proposed in this application significantly outperforms other models in multiple metrics, particularly methods using only shallow feature selection and models like U-Net, SegNet, and DUSegNet, especially in terms of accuracy. It also shows significant advantages compared to other recent semantic segmentation models. The benchmark mining area module segmentation model achieves an F1 score of 0.9038, a 3.26% improvement over the second-best method's F1 score (0.8712), and far exceeds the other deep learning methods listed in the table.

[0073] Five remote sensing sample images were randomly selected from the test set for comparative experiments and visualization to comprehensively evaluate the performance of various methods when processing different remote sensing sample images, such as... Figure 13 As shown in the figure, TP, TN, FP, and FN are highlighted in white (mining areas detected as mining areas), black (non-mining areas detected as non-mining areas), dark gray (non-mining areas detected as mining areas), and light gray (mining areas detected as non-mining areas), respectively. The specific color representation can be set arbitrarily according to needs. It can be clearly seen from the figure that the benchmark mining area module segmentation model has fewer dark gray and light gray areas in its segmentation map compared to the other models. Especially when dealing with some complex terrain and smaller mining areas, the benchmark mining area module segmentation model shows relatively superior performance. From the overall visualization, it can be seen that the benchmark mining area module segmentation model performs better than the other models in remote sensing mining area segmentation tasks.

[0074] Due to GPU limitations, the baseline mining area segmentation model can only process relatively small image data. Therefore, when a large remote sensing image is input, it is first segmented into smaller images (256x256) of the same size as the dataset image. This operation can lead to some non-mining areas being falsely identified as mining areas due to significant color differences from their surroundings. Figure 14 As shown, many bare areas in the upper right corner were mistakenly identified as mining areas due to their significant color difference from the surrounding environment, as well as some roads below.

[0075] Shallow feature filtering significantly and effectively reduces the above-mentioned problems, such as Figure 15 As shown. In Figure 15 In the data, it is evident that many false positive areas outside mining areas were effectively filtered out using the exponential feature screening method. Figure 14Some very small false positives in the upper left corner were successfully eliminated through area filtering using geometric selection. Similarly, false positives for the road below were successfully eliminated through aspect ratio filtering using geometric selection. Shallow feature selection has a good corrective effect on deep learning models, significantly improving the segmentation effect of mining areas and increasing the overall segmentation accuracy of the mining area segmentation task.

[0076] To further improve the accuracy of automatic classification of mine remote sensing images and enhance the practicality of intelligent methods in mineral resource supervision, this application combines shallow features of mining activity patches in remote sensing images, deep learning theory, and GIS technology to construct a refined identification model for mining activity patches in multi-source remote sensing images. The advantages of this invention are as follows: 1) A multi-source image open-pit mine basic sample library was constructed, containing 3,979 pairs of samples. Each pair of samples contains a 256x256 remote sensing sample image and the binary mask image of the corresponding mining area, providing rich background data reference for automated identification of mine remote sensing.

[0077] 2) Shallow quantitative indicators for remote sensing images were established, and a knowledge base of shallow mine features was constructed.

[0078] 3) A benchmark mining area module segmentation model that integrates shallow features and the OMSegNet deep network is proposed, which can effectively delineate and classify mining area patches such as limestone, sandstone, and shale, thereby effectively improving the efficiency of actual interpretation work.

[0079] In the model building phase, a deep learning semantic segmentation model integrating shallow features and the OMSegNet network was proposed and trained using samples. After approximately 8 hours of training, the loss function remained around 0.15, and the training accuracy was approximately 98.34%. On the test data, it could effectively delineate mining area patches, and the extracted contours showed good conformity to the actual mine boundaries. After the initial extraction by the model, the mining patches were then manually refined. Compared with manual interpretation, this significantly improved work efficiency while maintaining accuracy, making it more practical.

[0080] 4) Experimental results of mining area identification show that the model in this project has high accuracy in multiple evaluation indicators, which is better than traditional segmentation methods and other semantic segmentation models that do not integrate shallow features.

[0081] In the accuracy evaluation stage, an evaluation index system was constructed to describe the model performance using accuracy (AP), precision (P), recall (R), and F-rating (F). The performance of statistical methods based on feature knowledge bases, U-Net, SegNet, DUSegNet, and the DUSegNet method incorporating shallow features was compared horizontally for classifying and recognizing mining area patches. The advantages and disadvantages of each method were qualitatively and quantitatively compared intuitively and through indicators. Experimental results show that the proposed OMSegNet method incorporating shallow features can effectively identify mining area patches, and the outlined mining area edges are close to the effect of manual interpretation. Its segmentation range has a good fit with the true boundary of the mining area. From the quantitative comparison of evaluation factors, this method has higher overall recognition accuracy, outperforming traditional segmentation and classification methods based on shallow feature statistics, U-Net, SegNet, and the DUSegNet method without incorporating shallow features. This model is applicable to the identification of mining area patches using seven types of remote sensing image data: Gaofen-1, Gaofen-2, Gaofen-6, Gaofen-7, Ziyuan-1, Ziyuan-3, and China-Brazil-4. It performs better in identifying images with an original data resolution of 2m.

[0082] It should be noted that the method of this embodiment can be executed by a single device, such as a computer or server. The method of this embodiment can also be applied to a distributed scenario, where multiple devices cooperate to complete the task. In such a distributed scenario, one of these devices may execute only one or more steps of the method of this embodiment, and the multiple devices will interact with each other to complete the method described.

[0083] It should be noted that the above description describes some embodiments of the present invention. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims may be performed in a different order than that shown in the above embodiments and still achieve the desired results. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0084] Based on the same inventive concept, and corresponding to any of the above embodiments, the present invention also provides an automatic labeling system for remote sensing mining areas based on deep learning and shallow feature statistics.

[0085] refer to Figure 16 The remote sensing mining area automatic labeling system based on deep learning and shallow feature statistics includes: The sample acquisition module 161 is used to acquire a given remote sensing image sample set; wherein, the remote sensing image sample set includes: remote sensing sample images and benchmark mining area module segmented images; The model training module 162 is used to take each remote sensing sample image as the network source input and the corresponding benchmark mining area module segmentation image as the output. It combines a preset loss function and evaluation index, applies each remote sensing sample image, and trains the target deep learning image segmentation model to obtain the benchmark mining area module segmentation model. The patch segmentation module 163 is used to acquire a remote sensing image to be labeled, input the remote sensing image to be labeled into the benchmark mining area module segmentation model, and obtain a mining area module segmentation image; wherein, the mining area module segmentation image includes several mining area patches; The image patch classification module 164 is used to statistically analyze the object features of the mining area patches based on the segmented image from the mining area module, and to obtain the corresponding land cover category; wherein, the object features include: exponential features and geometric features; the land cover category includes: mineral category and non-mineral category; The mining area annotation module 165 is used to retain mining area patches whose land cover category is mineral based on the image segmented by the mining area module, and obtain remote sensing mining area annotation results.

[0086] For ease of description, the above system is described by dividing it into various modules based on their functions. Of course, in implementing this invention, the functions of each module can be implemented in one or more software and / or hardware components.

[0087] The system described in the above embodiments is used to implement the corresponding remote sensing mining area automatic labeling method based on deep learning and shallow feature statistics in any of the foregoing embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0088] Based on the same inventive concept, corresponding to the methods of any of the above embodiments, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the remote sensing mining area automatic labeling method based on deep learning and shallow feature statistics as described in any of the above embodiments.

[0089] Figure 17 This embodiment illustrates a more specific hardware structure of an electronic device. The device may include a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040, and a bus 1050. The processor 1010, memory 1020, input / output interface 1030, and communication interface 1040 are interconnected internally via the bus 1050.

[0090] The processor 1010 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this specification.

[0091] The memory 1020 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 1020 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented by software or firmware, the relevant program code is stored in the memory 1020 and is called and executed by the processor 1010.

[0092] The input / output interface 1030 is used to connect input / output modules to realize information input and output. Input / output modules can be configured as components within the device (not shown in the figure) or externally connected to the device to provide corresponding functions. Input devices may include keyboards, mice, touchscreens, microphones, various sensors, etc., while output devices may include displays, speakers, vibrators, indicator lights, etc.

[0093] The communication interface 1040 is used to connect a communication module (not shown in the figure) to enable communication between this device and other devices. The communication module can communicate via wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0094] Bus 1050 includes a pathway for transmitting information between various components of the device, such as processor 1010, memory 1020, input / output interface 1030, and communication interface 1040.

[0095] It should be noted that although the above-described device only shows the processor 1010, memory 1020, input / output interface 1030, communication interface 1040, and bus 1050, in specific implementations, the device may also include other components necessary for normal operation. Furthermore, those skilled in the art will understand that the above-described device may only include the components necessary for implementing the embodiments of this specification, and not necessarily all the components shown in the figures.

[0096] The electronic devices described above are used to implement the corresponding remote sensing mining area automatic labeling method based on deep learning and shallow feature statistics in any of the foregoing embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0097] Based on the same inventive concept, corresponding to the methods of any of the above embodiments, the present invention also provides a non-transitory computer-readable storage medium storing computer instructions, which are used to cause the computer to execute the remote sensing mining area automatic labeling method based on deep learning and shallow feature statistics as described in any of the above embodiments.

[0098] The computer-readable medium of this embodiment includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.

[0099] The computer instructions stored in the storage medium of the above embodiments are used to cause the computer to execute the remote sensing mining area automatic labeling method based on deep learning and shallow feature statistics as described in any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0100] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of the invention (including the claims) is limited to these examples; within the framework of the invention, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of the embodiments of the invention as described above, which are not provided in the details for the sake of brevity.

[0101] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, and the scope of the preferred embodiments of the invention includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as will be understood by those skilled in the art to which embodiments of the invention pertain.

[0102] Additionally, to simplify the description and discussion, and to avoid obscuring the embodiments of the invention, the well-known power / ground connections to the integrated circuit (IC) chip and other components may or may not be shown in the provided drawings. Furthermore, systems may be illustrated in block diagram form to avoid obscuring the embodiments of the invention, and this also takes into account the fact that the details of implementation of these block diagram systems are highly dependent on the platform on which the embodiments of the invention will be implemented (i.e., these details should be fully understood by those skilled in the art). While specific details (e.g., circuits) have been set forth to describe exemplary embodiments of the invention, it will be apparent to those skilled in the art that the embodiments of the invention may be implemented without these specific details or with variations thereof. Therefore, these descriptions should be considered illustrative rather than restrictive.

[0103] Although the invention has been described in conjunction with specific embodiments thereof, many substitutions, modifications, and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may be used with the embodiments discussed.

[0104] The embodiments of this invention are intended to cover all such substitutions, modifications, and variations falling within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the embodiments of this invention should be included within the protection scope of this invention.

Claims

1. An automatic annotation method for remote sensing mining areas based on deep learning and shallow feature statistics, characterized in that, include: Obtain a given set of remote sensing image samples; wherein, the set of remote sensing image samples includes: remote sensing sample images and segmented images of a benchmark mining area module; Using each remote sensing sample image as the network source input and the corresponding benchmark mining area module segmentation image as the output, the target deep learning image segmentation model is trained using each remote sensing sample image, combined with a preset loss function and evaluation index, to obtain the benchmark mining area module segmentation model; wherein, the target deep learning image segmentation model includes: a feature extraction backbone network, a neighbor feature aggregation module, a feature enhancement module, and a decoding module. A remote sensing image to be labeled is acquired, and the image is input into the baseline mining area module segmentation model to obtain a mining area module segmentation image; wherein, the mining area module segmentation image includes several mining area patches; Based on the object features of the segmented images of the mining area module, the corresponding land cover categories are obtained; wherein, the object features include: exponential features and geometric features; the land cover categories include: mineral categories and non-mineral categories; Based on the segmented image of the mining area module, mining area patches with land cover category of mineral are retained to obtain remote sensing mining area annotation results; The step of acquiring the remote sensing image to be labeled, and inputting the remote sensing image to be labeled into the benchmark mining area module segmentation model to obtain the mining area module segmentation image includes: A remote sensing image to be labeled is acquired and input into the benchmark mining area segmentation model. The benchmark mining area segmentation model receives the input remote sensing image to be labeled and extracts preliminary image features through a feature extraction backbone network, obtaining features at five stages with different depths and sizes. , , , , ; The features are input into the neighbor feature aggregation module, which aggregates features from adjacent stages to obtain the first aggregated feature; wherein, the neighbor feature aggregation module includes four aggregation lines, and the first aggregation line inputs... and Features; Second aggregation line input 、 and Features; Third aggregation line input 、 and Features; Fourth aggregation line input and feature; The first aggregated feature is input into the feature enhancement module. The feature enhancement module obtains the superimposed features of different receptive fields through the superposition of dilated convolutions, and aggregates the superimposed features together through residual connections to obtain the second aggregated feature. The second aggregated feature is input into the decoding module, which restores the second aggregated feature to the size of the input remote sensing image to be labeled, thus obtaining the mining area module segmentation image.

2. The automatic annotation method for remote sensing mining areas based on deep learning and shallow feature statistics according to claim 1, characterized in that, The process of obtaining a given remote sensing image sample set also includes: Acquire remote sensing images; By receiving the annotation information, the baseline mining area module segmentation image corresponding to the remote sensing image is obtained; The benchmark mining area module segmentation image is segmented according to a preset segmentation step size to obtain an effectively labeled image; The valid labeled images are filtered and their formats are standardized to obtain filtered labeled images; The selected and labeled images are randomly flipped, randomly cropped, and randomly blurred in color to obtain remote sensing sample images.

3. The automatic annotation method for remote sensing mining areas based on deep learning and shallow feature statistics according to claim 2, characterized in that, The step of segmenting the benchmark mining area module image according to a preset segmentation step size to obtain a validly labeled image includes: The image segmentation of the benchmark mining area module is segmented using a preset cutting step size to obtain a segmented shp file; Based on the cut shapefile, coordinate transformation is performed using the following formula to obtain the effectively annotated image: ， ， in, To effectively label the x-coordinate of the image, This represents the x-coordinate of the shapefile being cut. This represents the x-coordinate of the top-left vertex of the image. Indicates the spatial resolution of pixels. To effectively label the ordinate of the image, The ordinate of the top left corner vertex of the image. This represents the vertical axis of the cut shapefile.

4. The automatic annotation method for remote sensing mining areas based on deep learning and shallow feature statistics according to claim 1, characterized in that, The method of obtaining the land cover category based on the statistical features of the segmented image from the mining area module includes: The mining area patches in the segmented image of the mining area module are obtained, and the exponential and geometric features of each mining area patch are calculated. The exponential features include: normalized vegetation index, normalized water index, the exponential relationship between green and red bands in the mining area patch, the exponential relationship between blue and green bands in the mining area patch, and the exponential relationship between blue and red bands in the mining area patch. The geometric features include: area and aspect ratio. Based on the index features and geometric features, and according to the statistical results of the preset features, the corresponding land cover category is obtained; The process of obtaining mining area patches from the segmented image of the mining area module and calculating the exponential features of each mining area patch includes: The normalized vegetation index is calculated using the following formula: ， in, Represents the normalized vegetation index. This represents the near-infrared mean value of the mining area patch. Indicates the mean value of the red band; The normalized water index is calculated using the following formula: ， in, Indicates the normalized water quality index. Indicates the mean value of the green band. This represents the near-infrared average value of a patch in the mining area. The exponential relationship between the blue and red bands in the mining area map is calculated using the formula shown below: ， in, This indicates the exponential relationship between the blue and red bands in a mining area map. Indicates the blue band mean. Indicates the mean value of the red band; The exponential relationship between the green and red bands in the mining area map is calculated using the formula shown below: ， in, This indicates the exponential relationship between the green and red bands in a mining area map. Indicates the mean value of the green band. Indicates the mean value of the red band; The exponential relationship between the blue and green bands in the mining area map is calculated using the formula shown below: ， in, This indicates the exponential relationship between the blue and green bands in a mining area map. Indicates the blue band mean. This represents the mean value of the green band.

5. An automatic annotation system for remote sensing mining areas based on deep learning and shallow feature statistics, characterized in that, for implementing the automatic annotation method for remote sensing mining areas based on deep learning and shallow feature statistics as described in any one of claims 1-4, it includes: The sample acquisition module is used to acquire a given remote sensing image sample set; wherein, the remote sensing image sample set includes: remote sensing sample images and benchmark mining area module segmented images; The model training module is used to take each remote sensing sample image as the network source input and the corresponding benchmark mining area module segmentation image as the output. It combines a preset loss function and evaluation index, applies each remote sensing sample image, and trains the target deep learning image segmentation model to obtain the benchmark mining area module segmentation model. The patch segmentation module is used to acquire the remote sensing image to be labeled, input the remote sensing image to be labeled into the benchmark mining area module segmentation model, and obtain the mining area module segmentation image; wherein, the mining area module segmentation image includes several mining area patches; The image patch classification module is used to statistically analyze the object features of the mining area patches based on the segmented image from the mining area module, and to obtain the corresponding land cover category; wherein, the object features include: exponential features and geometric features; the land cover category includes: mineral category and non-mineral category; The mining area labeling module is used to retain mining area patches whose land cover category is mineral, based on the image segmented by the mining area module, and obtain remote sensing mining area labeling results.

6. An apparatus comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 4.

7. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 4.

Images