An object-oriented visible light remote sensing image mine repair identification and classification method and system
By improving the YOLOv8n depth algorithm and dataset partitioning strategy, the problem of low identification efficiency of mine ecological restoration in visible light remote sensing images was solved, and efficient and accurate identification and classification of mine ecological restoration was achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- JIANGSU GEOLOGICAL SURVEY INST
- Filing Date
- 2025-04-01
- Publication Date
- 2026-05-19
AI Technical Summary
Existing technologies are inefficient and lack real-time monitoring in the identification of mine ecological restoration in visible light remote sensing images, while deep learning algorithms require a lot of computing power and are difficult to classify efficiently.
An improved YOLOv8n deep learning algorithm is adopted, replacing the original convolution with GhostConv and combining it with the C2f_RepGhost module for feature fusion to extract spectral and spatial features, constructing a mine ecological restoration identification model. The X-anglabeling tool and manual interpretation are used to divide the dataset, achieving fast and accurate mine ecological restoration identification.
It improves the efficiency and accuracy of mine ecological restoration identification, reduces computing power requirements, and achieves efficient mine ecological restoration identification and classification.
Smart Images
Figure CN120298789B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of remote sensing technology, and in particular to an object-oriented method and system for identifying and classifying mine restoration using visible light remote sensing images. Background Technology
[0002] With the rapid development of global industrialization and urbanization, mineral resources have become a crucial foundation for the operation of modern society. However, mineral resource extraction activities are often accompanied by the destruction of natural landscapes and the degradation of the ecological environment. In particular, the abandoned sites created during mining operations not only affect the aesthetics of the region but may also trigger geological disasters, threatening human life and property. Therefore, the ecological restoration and management of abandoned mining sites has become an urgent issue to be addressed in mine environmental governance and sustainable development.
[0003] Early mine ecological restoration identification relied primarily on manual interpretation. With the rapid development of remote sensing satellites and drones, satellite and drone imagery became the main detection methods, using metrics such as Normalized Difference Vegetation Index (NDVI), Normalized Differential Vegetation Index (EVI), Greenness Index (GI), and Free Fiber Capacity (FVC) for monitoring mine ecological restoration identification. However, as the monitoring scope expanded, the emergence of machine learning significantly reduced manual work time and improved monitoring efficiency, with methods such as random forests, support vector machines, and maximum likelihood estimation. However, with the increasing volume of data available today, deep learning has begun to replace machine learning and is widely used in other fields. But high-efficiency and real-time monitoring of mine ecological restoration identification using visible light remote sensing imagery is still rare; most methods utilize simple convolutional networks and require substantial computational power. Therefore, to improve the speed of mine ecological restoration identification using visible light remote sensing imagery, it is essential to continuously develop deep learning classification techniques for visible light remote sensing images. Summary of the Invention
[0004] The purpose of this application is to provide an object-oriented visible light remote sensing image mine restoration identification and classification method and system, which can improve the efficiency and accuracy of mine ecological restoration identification.
[0005] To achieve the above objectives, this application provides the following solution:
[0006] Firstly, this application provides an object-oriented method for identifying and classifying mine restoration using visible light remote sensing images, including:
[0007] Acquire visible light remote sensing images of the target mine from remote sensing satellites;
[0008] The visible light remote sensing image is input into the trained mine ecological restoration identification model to obtain the classification result of mine ecological restoration;
[0009] The training process of the mine ecological restoration identification model is as follows:
[0010] Acquire several sample visible light remote sensing images from the visible light remote sensing image database;
[0011] The sample visible light remote sensing images are divided into study areas, and target labels for each type of land cover are made for each study area to obtain labeled visible light remote sensing images of the study areas.
[0012] Based on the labeled visible light remote sensing images of the study area, an image dataset is constructed according to a partitioning strategy; the image dataset includes a training set and a test set partitioned according to a set ratio;
[0013] The improved YOLOv8n deep learning algorithm extracts spectral and spatial features from the image dataset, and trains the model through feature fusion to obtain a trained mine ecological restoration identification model. The improved YOLOv8n deep learning algorithm is based on the original YOLOv8n deep learning algorithm. In the information extraction part of the backbone extraction network, GhostConv replaces the original convolution, and in the feature fusion part, the C2f_RepGhost module is used for refinement.
[0014] Optionally, based on the labeled visible light remote sensing images of the study area, an image dataset is constructed according to a partitioning strategy; the image dataset includes a training set and a test set partitioned according to a set ratio, specifically including:
[0015] Based on the labeled visible light remote sensing images of the study area, the X-ang labeling tool was used in conjunction with a classification strategy combining deep learning and manual interpretation to divide the labeled visible light remote sensing images of the study area into training and test sets.
[0016] Optionally, the labeled visible light remote sensing images of the study area are divided into a training set and a test set, specifically including:
[0017] The ratio λ of the pixels participating in training to all pixels in the original image is obtained, as well as the number of labeled pixels in each training block; the original image is a labeled visible light remote sensing image of the study area.
[0018] Based on the ratio λ, training images and verification-test images to be segmented are obtained from the original image;
[0019] Based on the number T of labeled pixels in each training block, a portion of labeled pixels are randomly selected from the training images to be divided and retained to obtain the training images.
[0020] Based on the number T of labeled pixels in each training block, the images to be divided into verification images and test images according to the proposed verification / test sample ratio.
[0021] The training image, validation image, and test image are each divided into W×H×B blocks; where W and H represent the width and height of the block, and B represents the number of spectral bands.
[0022] Remove blocks that do not contain labeled pixels to obtain training blocks and test blocks;
[0023] A sliding window strategy is used to expand the number of training and test blocks to obtain training and test sets.
[0024] Optionally, spectral and spatial features are extracted from the image dataset based on the improved YOLOv8n depth algorithm, specifically including:
[0025] By utilizing the improved YOLOv8n deep learning algorithm, including yolov8n, Ghost convolution, shared convolutional layer conv_shared, and reconstruction segment module, spectral and spatial features of the visible light remote sensing images of the study area in the image dataset are extracted, resulting in spectral and spatial features.
[0026] Optionally, the size of the study area is 640 pixels × 640 pixels.
[0027] Optionally, the land feature categories include mining areas, vegetation, restored mining areas, farmland, water bodies, buildings, roads, and background.
[0028] Secondly, this application provides an object-oriented visible light remote sensing image mine restoration identification and classification system, including:
[0029] The image acquisition module is used to acquire visible light remote sensing images of the target mine from remote sensing satellites.
[0030] The identification module is used to input the visible light remote sensing image into the trained mine ecological restoration identification model to obtain the classification result of mine ecological restoration.
[0031] The training module includes:
[0032] The image acquisition submodule is used to acquire several sample visible light remote sensing images from the visible light remote sensing image database.
[0033] The segmentation and labeling submodule is used to divide the sample visible light remote sensing image into study areas and to label the target categories of each land cover in each study area to obtain the labeled visible light remote sensing image of the study area.
[0034] The dataset construction submodule is used to construct an image dataset based on the labeled visible light remote sensing images of the study area and a partitioning strategy; the image dataset includes a training set and a test set partitioned according to a set ratio;
[0035] The training submodule is used to extract spectral and spatial features from the image dataset based on the improved YOLOv8n deep learning algorithm, and to train the model through feature fusion to obtain a trained mine ecological restoration identification model. The improved YOLOv8n deep learning algorithm is based on the original YOLOv8n deep learning algorithm. In the information extraction part of the backbone extraction network, GhostConv replaces the original convolution, and in the feature fusion part, it is refined by the C2f_RepGhost module.
[0036] Optionally, the dataset construction submodule specifically includes:
[0037] The dataset construction unit is used to divide the labeled visible light remote sensing images of the study area into training and test sets based on the X-ang labeling tool combined with a partitioning strategy of deep learning and manual interpretation.
[0038] Optionally, the dataset construction unit specifically includes:
[0039] The information acquisition subunit is used to acquire the ratio λ of the pixels participating in training to all pixels in the original image, and the number of labeled pixels in each training block; the original image is a labeled visible light remote sensing image of the study area.
[0040] The sub-unit is used to divide the original image into a training image and a verification-test image according to the ratio λ.
[0041] The first sub-unit of image selection is used to randomly select a portion of labeled pixels in the training image to be divided based on the number T of labeled pixels in each training block, so as to obtain the training image.
[0042] The second sub-unit of image selection is used to divide the unselected verification-test images into verification images and test images according to the proposed verification / test sample ratio, based on the number T of labeled pixels in each training block.
[0043] The segmentation subunit is used to segment the training image, validation image, and test image into W×H×B blocks respectively; where W and H represent the width and height of the block, and B represents the number of spectral bands.
[0044] Remove sub-units to remove blocks that do not contain labeled pixels, resulting in training blocks and test blocks;
[0045] The expansion subunit is used to expand the number of training blocks and test blocks using a sliding window strategy to obtain training and test sets.
[0046] Optionally, the training submodule includes:
[0047] The feature extraction unit is used to extract spectral and spatial features from the visible light remote sensing images of the study area in the image dataset by utilizing the improved YOLOv8n deep learning algorithm, including yolov8n, Ghost convolution, shared convolutional layer conv_shared, and reconstruction segment module.
[0048] According to the specific embodiments provided in this application, the following technical effects are disclosed:
[0049] This application provides an object-oriented visible light remote sensing imagery-based method and system for identifying and classifying mine restoration. The method involves acquiring visible light remote sensing images of the target mine from remote sensing satellites; inputting these images into a trained mine ecological restoration identification model to obtain classification results; during the training of the model: acquiring several sample visible light remote sensing images from a database to cover more possible mine ecological restoration situations, thus enhancing the model's generalization ability; dividing the sample images into study areas and labeling each study area with local vegetation categories to provide clear training objectives and improve identification accuracy; constructing an image dataset based on the labeled study area images and the partitioning strategy to ensure the stability and effectiveness of the model training process; and improving the efficiency and accuracy of feature extraction through the introduction of GhostConv and C2f_RepGhost modules in the improved YOLOv8n deep learning algorithm, thereby enhancing the model's recognition performance. Attached Figure Description
[0050] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0051] Figure 1 A flowchart illustrating an object-oriented visible light remote sensing image mine restoration identification and classification method provided in an embodiment of this application;
[0052] Figure 2 A flowchart of a mine ecological restoration identification and classification method provided in an embodiment of this application;
[0053] Figure 3 This is a schematic diagram of the functional modules of an object-oriented visible light remote sensing image mine restoration identification and classification system provided in an embodiment of this application. Detailed Implementation
[0054] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0055] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0056] Example 1
[0057] like Figure 1 As shown in the figure, this embodiment provides an object-oriented visible light remote sensing image mine restoration identification and classification method, including:
[0058] Step 101: Acquire visible light remote sensing images of the target mine from remote sensing satellites;
[0059] Step 102: Input the visible light remote sensing image into the trained mine ecological restoration recognition model to obtain the classification result of mine ecological restoration;
[0060] The training process of the mine ecological restoration identification model is as follows:
[0061] Step 201: Obtain several sample visible light remote sensing images from the visible light remote sensing image database;
[0062] Step 202: Divide the sample visible light remote sensing image into study areas, and mark the target of each land cover category in each study area to obtain the marked visible light remote sensing image of the study area;
[0063] Step 203: Based on the labeled visible light remote sensing images of the study area, construct an image dataset according to the partitioning strategy; the image dataset includes a training set and a test set divided according to a set ratio;
[0064] Step 204: Extract spectral and spatial features from the image dataset using the improved YOLOv8n deep learning algorithm, and train the model through feature fusion to obtain a trained mine ecological restoration identification model. The improved YOLOv8n deep learning algorithm is based on the original YOLOv8n deep learning algorithm. In the information extraction part of the backbone extraction network, GhostConv replaces the original convolution, and in the feature fusion part, it is refined using the C2f_RepGhost module.
[0065] In some embodiments, such as Figure 2 As shown, before performing step 201, the following steps are also included:
[0066] The system acquires RGB band remote sensing images from visible light remote sensing satellites, preprocesses the acquired images, and then converts the preprocessed images into JPG format. Preprocessing includes radiometric correction, atmospheric correction, and geometric correction.
[0067] In some embodiments, such as Figure 2 As shown, when performing step 202, the specific steps can be as follows:
[0068] Based on the obtained visible light remote sensing images, the study area is divided and corresponding labels are created, resulting in visible light remote sensing images and corresponding labels for the study area. A mine ecological restoration identification dataset is then created, which involves target labeling for each vegetation category within each study area. In practical applications, there are currently virtually no usable visible light mine ecological restoration identification datasets for scientific research; therefore, this method creates a visible light mine ecological restoration identification dataset for classification research.
[0069] When selecting study areas, the main focus is on mining areas with good imaging conditions, imaging time between June and October each year, and normal vegetation growth. The labeling mainly combines deep learning automatic recognition and manual interpretation to classify the ground into eight categories: mine, plant, repair mine, farmland, water, build, road, and background.
[0070] Specifically, the study area is characterized by normal vegetation growth and good imaging conditions from June to October. Previously, each label was manually drawn. As an improvement, this approach utilizes X-anglabeling and ArcGIS software to classify each land feature through deep learning-based automatic segmentation and manual verification, thus obtaining a mine ecological restoration label dataset.
[0071] The selected study area is 640 pixels × 640 pixels, characterized in that: the selected study area is 640 pixels × 640 pixels. Therefore, the marked study area visible light remote sensing image is a 640 × 640 × 3 visible spectrum remote sensing image, where the two 640s are the height and width of the image, respectively, in pixels, and 3 is the number of bands.
[0072] In some embodiments, such as Figure 2 As shown, when performing step 203, the specific steps can be as follows:
[0073] Based on the labeled visible light remote sensing images of the study area, the X-ang labeling tool was used in conjunction with a classification strategy combining deep learning and manual interpretation to divide the labeled visible light remote sensing images of the study area into training and test sets.
[0074] The labeled visible light remote sensing images of the study area were divided into a training set and a test set, specifically including:
[0075] The ratio λ of pixels participating in training to all pixels in the original image, and the number T of labeled pixels in each training block are obtained; the original image is a labeled visible light remote sensing image of the study area; thus, the number of training blocks in each category is:
[0076]
[0077] Where i represents the pixel category, N i This represents the number of training blocks (N) in the corresponding category. i ≥1), n i This represents the total number of pixels for each category in the original image.
[0078] Based on the ratio λ, training images and verification-test images to be segmented are obtained from the original image;
[0079] Based on the number T of labeled pixels in each training block, a portion of labeled pixels are randomly selected from the training images to be divided and retained to obtain the training images.
[0080] Based on the number T of labeled pixels in each training block, the images to be divided into verification images and test images according to the proposed verification / test sample ratio.
[0081] The training image, validation image, and test image are each divided into W×H×B blocks; where W and H represent the width and height of the block, and B represents the number of spectral bands.
[0082] Remove blocks that do not contain labeled pixels to obtain training blocks and test blocks;
[0083] A sliding window strategy is used to expand the number of training and test blocks to obtain training and test sets.
[0084] In some embodiments, step 204 may specifically include:
[0085] By utilizing the improved YOLOv8n deep learning algorithm, including yolov8n, Ghost convolution, shared convolutional layers (conv_shared), and the reconstruction segment module, spectral and spatial features of the visible light remote sensing images of the study area in the image dataset were extracted. Among these, the Ghost convolution, as a crucial feature extraction module, reduces redundant feature information when extracting important features. During feature information transmission, the shared convolutional layer further filters the feature information, eliminating similar features extracted from different regions, further reducing identical feature information. The improved segment module divides the remote sensing images according to the categories used in training. Compared with the original algorithm, the improved algorithm shows significant advantages in accuracy and efficiency in identifying the state of mine ecological restoration.
[0086] Specifically, in the information extraction part of the backbone network, GhostConv replaces the original convolutions. In remote sensing images, many features are redundant; GhostConv can extract key information through lightweight operations. GhostConv can magnify the field of view, enhance the detection capability for large targets, provide efficient feature extraction capabilities, and capture basic edge and texture information at shallow layers. Furthermore, a new C2f module is reconstructed using the RepGhostBottleneck module, which consists of reparameterized Ghost modules, enhancing low-level computational efficiency while maintaining accuracy.
[0087] In the feature fusion part, the C2f_RepGhost module is used for refinement. After upsampling, the features are further refined to extract more accurate edge and semantic information.
[0088] Lightweight processing introduces a shared convolutional layer (conv_shared) for feature extraction, and then completes the regression and classification tasks through two branches respectively. This reduces redundant computation in the convolutional layers. Shared feature extraction reduces the number of parameters and memory usage, and the Segment module completes the segmentation task by adding prototype features (Proto) and mask coefficients. Therefore, this application provides an object-oriented visible light remote sensing image classification method for mine ecological restoration identification based on deep learning algorithms, which is accurate in classification and requires only a small amount of computing power.
[0089] Example 2
[0090] like Figure 3 As shown, this embodiment provides an object-oriented visible light remote sensing image mine restoration identification and classification system, including:
[0091] Image acquisition module 301 is used to acquire visible light remote sensing images of the target mine from remote sensing satellites;
[0092] The identification module 302 is used to input the visible light remote sensing image into the trained mine ecological restoration identification model to obtain the classification result of mine ecological restoration.
[0093] Training module 303 includes:
[0094] The image acquisition submodule 311 is used to acquire several sample visible light remote sensing images from the visible light remote sensing image database.
[0095] The division and labeling submodule 312 is used to divide the sample visible light remote sensing image into study areas and to label the target categories of each land cover in each study area to obtain the labeled visible light remote sensing image of the study area.
[0096] The dataset construction submodule 313 is used to construct an image dataset based on the labeled visible light remote sensing images of the study area and a partitioning strategy; the image dataset includes a training set and a test set partitioned according to a set ratio;
[0097] The training submodule 314 is used to extract spectral and spatial features from the image dataset based on the improved YOLOv8n deep algorithm, and to train the model through feature fusion to obtain a trained mine ecological restoration identification model. The improved YOLOv8n deep algorithm is based on the original YOLOv8n deep algorithm. In the information extraction part of the backbone extraction network, GhostConv replaces the original convolution, and in the feature fusion part, it is refined by the C2f_RepGhost module.
[0098] Specifically, the dataset construction submodule 313 includes:
[0099] The dataset construction unit is used to divide the labeled visible light remote sensing images of the study area into training and test sets based on the X-ang labeling tool combined with a partitioning strategy of deep learning and manual interpretation.
[0100] The dataset construction unit specifically includes:
[0101] The information acquisition subunit is used to acquire the ratio λ of the pixels participating in training to all pixels in the original image, and the number of labeled pixels in each training block; the original image is a labeled visible light remote sensing image of the study area.
[0102] The sub-unit is used to divide the original image into a training image and a verification-test image according to the ratio λ.
[0103] The first sub-unit of image selection is used to randomly select a portion of labeled pixels in the training image to be divided based on the number T of labeled pixels in each training block, so as to obtain the training image.
[0104] The second sub-unit of image selection is used to divide the unselected verification-test images into verification images and test images according to the proposed verification / test sample ratio, based on the number T of labeled pixels in each training block.
[0105] The segmentation subunit is used to segment the training image, validation image, and test image into W×H×B blocks respectively; where W and H represent the width and height of the block, and B represents the number of spectral bands.
[0106] Remove sub-units to remove blocks that do not contain labeled pixels, resulting in training blocks and test blocks;
[0107] The expansion subunit is used to expand the number of training blocks and test blocks using a sliding window strategy to obtain training and test sets.
[0108] The training submodule 314 includes:
[0109] The feature extraction unit is used to extract spectral and spatial features from the visible light remote sensing images of the study area in the image dataset by utilizing the improved YOLOv8n deep learning algorithm, including yolov8n, Ghost convolution, shared convolutional layer conv_shared, and reconstruction segment module.
[0110] In summary, this application has the following technical effects:
[0111] This application uses visible light satellite remote sensing images to delineate the study area and create corresponding labels. A data segmentation strategy with no information leakage is employed to ensure that the training and test sets are free of information leakage. Finally, a deep learning algorithm is used to extract the spectral and spatial features of the visible light remote sensing data and then fuse them to achieve identification and classification for mine ecological restoration. Furthermore, the proposed improved YOLOv8n deep learning algorithm outperforms simple convolutional networks.
[0112] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0113] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.
Claims
1. An object-oriented method for identifying and classifying mine restoration from visible light remote sensing images, characterized in that, include: Acquire visible light remote sensing images of the target mine from remote sensing satellites; The visible light remote sensing image is input into the trained mine ecological restoration identification model to obtain the classification result of mine ecological restoration; The training process of the mine ecological restoration identification model is as follows: Acquire several sample visible light remote sensing images from the visible light remote sensing image database; The sample visible light remote sensing images are divided into study areas, and target labels for each study area are applied to the different land cover categories to obtain labeled visible light remote sensing images of the study areas. The labeling is done by combining deep learning automatic recognition and manual interpretation to classify the ground into eight categories: mine, plant, repair mine, farmland, water, build, road, and background. Based on the labeled visible light remote sensing images of the study area, an image dataset is constructed according to a partitioning strategy; the image dataset includes a training set and a test set partitioned according to a set ratio; The improved YOLOv8n deep learning algorithm extracts spectral and spatial features from the image dataset, and trains the model through feature fusion to obtain a trained mine ecological restoration identification model. The improved YOLOv8n deep learning algorithm is based on the original YOLOv8n deep learning algorithm. In the information extraction part of the backbone extraction network, GhostConv replaces the original convolution, and in the feature fusion part, the C2f_RepGhost module is used for refinement. Based on the labeled visible light remote sensing images of the study area, an image dataset is constructed according to a partitioning strategy. The image dataset includes a training set and a test set partitioned according to a predetermined ratio, specifically including: Based on the labeled visible light remote sensing images of the study area, the labeled visible light remote sensing images of the study area were divided into training set and test set using the X-anglabeling tool combined with deep learning and human interpretation. The labeled visible light remote sensing images of the study area were divided into training and test sets, specifically including: The ratio λ of the pixels participating in training to all pixels in the original image is obtained, as well as the number of labeled pixels in each training block; the original image is a labeled visible light remote sensing image of the study area. Based on the ratio λ, training images and verification-test images to be segmented are obtained from the original image; Based on the number T of labeled pixels in each training block, a portion of labeled pixels are randomly selected from the training images to be divided and retained to obtain the training images. Based on the number T of labeled pixels in each training block, the images to be divided into verification images and test images according to the proposed verification / test sample ratio. The training image, validation image, and test image are each divided into W×H×B blocks; where W and H represent the width and height of the block, and B represents the number of spectral bands. Remove blocks that do not contain labeled pixels to obtain training blocks and test blocks; A sliding window strategy is used to expand the number of training and test blocks to obtain training and test sets; Based on the improved YOLOv8n deep learning algorithm, spectral and spatial features are extracted from the image dataset, specifically including: By utilizing the improved YOLOv8n deep learning algorithm, including yolov8n, Ghost convolution, shared convolutional layers (conv_shared), and the reconstruction segment module, spectral and spatial features of the visible light remote sensing images of the study area in the image dataset are extracted. The Ghost convolution is used to reduce redundant feature information when extracting important feature information. During the feature information transmission process, the shared convolutional layer is used to filter feature information and remove similar feature information extracted from different regions. The segment module is used to divide the remote sensing images according to the categories in the training.
2. The object-oriented visible light remote sensing image mine restoration identification and classification method according to claim 1, characterized in that, The study area is 640 pixels × 640 pixels in size.
3. The object-oriented visible light remote sensing image mine restoration identification and classification method according to claim 1, characterized in that, The land feature categories include mining areas, vegetation, restored mining areas, farmland, water bodies, buildings, roads, and background.
4. An object-oriented visible light remote sensing image mine restoration identification and classification system, characterized in that, include: The image acquisition module is used to acquire visible light remote sensing images of the target mine from remote sensing satellites. The identification module is used to input the visible light remote sensing image into the trained mine ecological restoration identification model to obtain the classification result of mine ecological restoration. The training module includes: The image acquisition submodule is used to acquire several sample visible light remote sensing images from the visible light remote sensing image database. The segmentation and labeling submodule is used to divide the sample visible light remote sensing image into study areas and to label the target categories of each land cover in each study area to obtain the labeled visible light remote sensing image of the study area. The dataset construction submodule is used to construct an image dataset based on the labeled visible light remote sensing images of the study area and a partitioning strategy; the image dataset includes a training set and a test set partitioned according to a set ratio; The training submodule is used to extract spectral and spatial features from the image dataset based on the improved YOLOv8n deep learning algorithm, and to train the model through feature fusion to obtain a trained mine ecological restoration identification model. The improved YOLOv8n deep learning algorithm is based on the original YOLOv8n deep learning algorithm. In the information extraction part of the backbone extraction network, GhostConv replaces the original convolution, and in the feature fusion part, it is refined by the C2f_RepGhost module. The dataset construction submodule specifically includes: The dataset construction unit is used to divide the labeled visible light remote sensing images of the study area into training and test sets based on the X-ang labeling tool combined with a partitioning strategy of deep learning and manual interpretation. The dataset construction unit specifically includes: The information acquisition subunit is used to acquire the ratio λ of the pixels participating in training to all pixels in the original image, and the number of labeled pixels in each training block; the original image is a labeled visible light remote sensing image of the study area. The sub-unit is used to divide the original image into a training image and a verification-test image according to the ratio λ. The first sub-unit of image selection is used to randomly select a portion of labeled pixels in the training image to be divided based on the number T of labeled pixels in each training block, so as to obtain the training image. The second sub-unit of image selection is used to divide the unselected verification-test images into verification images and test images according to the proposed verification / test sample ratio, based on the number T of labeled pixels in each training block. The segmentation subunit is used to segment the training image, validation image, and test image into W×H×B blocks respectively; where W and H represent the width and height of the block, and B represents the number of spectral bands. Remove sub-units to remove blocks that do not contain labeled pixels, resulting in training blocks and test blocks; The expansion subunit is used to expand the number of training blocks and test blocks using a sliding window strategy to obtain training sets and test sets. The training submodule includes: The feature extraction unit utilizes the improved YOLOv8n deep learning algorithm, including yolov8n, Ghost convolution, shared convolutional layers (conv_shared), and the reconstruction segment module, to extract spectral and spatial features from the visible light remote sensing images of the study area in the image dataset. The Ghost convolution reduces redundant feature information when extracting important features. During feature information transfer, the shared convolutional layer filters features, removing similar features extracted from different regions. The segment module divides the remote sensing images according to the categories used in training.