Tailing pond dry beach length identification method and device, computer equipment and storage medium

Abstract

The invention relates to a tailing pond dry beach length identification method and device, computer equipment and a storage medium. The method comprises the following steps: acquiring a tailing pond remote sensing image; inputting the tailing pond remote sensing image into a pre-trained target detection model, and outputting a tailing pond target detection result through the target detection model; classifying multiple types of tailings pond elements contained in the tailings pond target detection result by adopting a superpixel segmentation algorithm and a random forest algorithm to obtain multiple types of tailings pond elements; and merging the same tailings pond elements, determining reference lines corresponding to the tailings pond elements according to a merging result, and obtaining the length of the dry beach based on the reference lines corresponding to the tailings pond elements. By adopting the method, the dry beach length can be automatically and accurately identified.

Description

Technical Field

[0001] The present disclosure relates to the field of image processing technology, and in particular to a data transmission method, device, computer equipment, and storage medium. Background Art

[0002] With the development of image processing technology, the dry beach length of the tailings pond is the horizontal distance from the water edge line in the reservoir to the top of the beach, which is one of the flood control elements of the tailings pond. The dry beach length closest to the top of the beach on the water edge line in the reservoir is extracted from the high-resolution remote sensing image as a dynamic monitoring indicator. The dry beach length is the horizontal distance from the closest point to the top of the beach on the water edge line in the reservoir to the top of the beach. At present, the identification of the dry beach length of the tailings pond is mostly carried out by setting up optical cameras and distance monitoring sensors on site. The deployment cost of such equipment is high, and most of them can only monitor the dry beach length of a single tailings pond, making it difficult to achieve large-scale automated monitoring.

[0003] Therefore, the existing detection technology has problems such as low tailings pond detection accuracy and insufficient calculation accuracy of dry beach length, which seriously restricts the ability to automatically identify the length of the dry beach of the tailings pond. Summary of the invention

[0004] Based on this, it is necessary to provide a method, device, computer equipment, and storage medium for identifying the dry beach length of a tailings pond in response to the above-mentioned technical problems.

[0005] In a first aspect, the present disclosure provides a method for identifying the dry beach length of a tailings pond. The method comprises: Obtain remote sensing images of tailings ponds; Inputting the tailings pond remote sensing image into a pre-trained target detection model, and outputting the tailings pond target detection result via the target detection model; A superpixel segmentation algorithm and a random forest algorithm are used to classify multiple types of tailings pond elements contained in the tailings pond target detection result to obtain multiple types of tailings pond elements; The same tailings pond elements of each type are merged, and the reference line corresponding to the tailings pond element is determined according to the merged result. Based on the reference line corresponding to the tailings pond element, the dry beach length is obtained.

[0006] In one embodiment, the superpixel segmentation algorithm and the random forest algorithm are used to classify the multiple types of tailings pond elements contained in the tailings pond target detection result to obtain multiple types of tailings pond elements, including: Assigning the tailings pond target detection result to a coordinate reference system to obtain a target frame vector; Based on a preset buffer distance, the target frame vector is expanded to obtain an expanded target frame vector; Determine a seed center point in the expanded target frame vector, and merge similar pixels based on the seed point center through a clustering method to form a superpixel; Detecting the edge of the superpixel, and obtaining a superpixel segmentation surface by extracting the center line and converting the line to the surface; The random forest algorithm is used to classify the superpixel segmentation surface to obtain multiple types of tailings pond elements.

[0007] In one embodiment, the random forest algorithm is used to classify the superpixel segmentation surface to obtain multiple types of tailings pond elements, including: Determine the element characteristics of each type of tailings pond element based on the predetermined classification samples; Using the random forest algorithm to learn the feature features and form a superpixel classification model; The superpixel classification model is used to classify the superpixel segmentation surface to obtain multiple types of tailings pond elements.

[0008] In one embodiment, the tailings pond elements include: a dam element, a dry beach element, and a water body element; merging the same tailings pond elements of each type, determining a reference line corresponding to the tailings pond element according to the merging result, and obtaining the dry beach length based on the reference line corresponding to the tailings pond element, including: Merging the same tailings pond elements among the multiple types of tailings pond elements to obtain a merged result of each tailings pond element; The combined result of the accumulation dam elements is processed by using a curve simplification algorithm to obtain an initial dam baseline; The outer boundary of the water body element is determined, and the length of the dry beach is determined based on the distance between the initial dam baseline and the outer boundary of the water body element.

[0009] In one embodiment, the method further comprises: Acquire a training remote sensing image, wherein the training remote sensing image contains a plurality of tailings pond areas; Determine the tailings pond area contained in the training remote sensing image, draw a vector frame and assign values ​​to the tailings pond area, and obtain a vector boundary of the tailings pond target detection frame; A target detection model is trained based on the tailings pond target detection box vector boundary and the training remote sensing image.

[0010] In one embodiment, the target detection model includes a backbone network, a neck network and a head network, and the backbone network includes a rectangular feature extraction module; the target detection model is trained based on the tailings pond target detection box vector boundary and the training remote sensing image, including: Based on the backbone network, extracting the tailings pond feature information contained in the tailings pond target detection frame vector boundary in the training remote sensing image; Performing secondary feature extraction on the tailings pond feature information based on the rectangular feature extraction module to obtain high-order feature information of the tailings pond; The high-order feature information of the tailings pond is fused with the feature information of the tailings pond after dimension upgrading, and the target detection model is trained using the fused feature information and the vector boundary of the tailings pond target detection box.

[0011] In a second aspect, the present disclosure further provides a tailings pond dry beach length device. The device comprises: Image acquisition module, used to obtain remote sensing images of tailings ponds; A model processing module, used for inputting the tailings pond remote sensing image into a pre-trained target detection model, and outputting the tailings pond target detection result via the target detection model; A classification module is used to classify multiple types of tailings pond elements contained in the tailings pond target detection result by using a superpixel segmentation algorithm and a random forest algorithm to obtain multiple types of tailings pond elements; The length calculation module is used to merge the same tailings pond elements of each type, determine the reference line corresponding to the tailings pond element according to the merging result, and obtain the dry beach length based on the reference line corresponding to the tailings pond element.

[0012] In a third aspect, the present disclosure further provides a computer device, wherein the computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps in any of the above method embodiments are implemented.

[0013] In a fourth aspect, the present disclosure further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps in any of the above method embodiments are implemented.

[0014] In a fifth aspect, the present disclosure further provides a computer program product, wherein the computer program product comprises a computer program, and when the computer program is executed by a processor, the steps in any of the above method embodiments are implemented.

[0015] In the above embodiments, by obtaining the remote sensing image of the tailings pond and processing it with the pre-trained target detection model, the tailings pond related targets can be accurately located from the macroscopic remote sensing image level, avoiding the problems of limited vision and low efficiency that may exist in manual field surveys. For example, for a remote and large tailings pond area, the specific location and approximate range of the tailings pond in the image can be quickly and accurately determined by combining the remote sensing image with the target detection model, thereby improving the comprehensiveness and accuracy of the tailings pond target detection. This model-based automated detection method can process a large amount of remote sensing image data at the same time, is suitable for monitoring multiple tailings ponds or large areas containing tailings ponds, and is helpful to realize the large-scale and batch tailings pond information extraction work. The superpixel segmentation algorithm and the random forest algorithm are used to classify the multi-class tailings pond elements contained in the detected tailings pond target detection results, which can further refine the cognition of different elements inside the tailings pond, and can more deeply understand the details of the tailings pond, thereby improving the accuracy of dry beach identification. The same tailings pond elements of each type are merged, and then the reference line corresponding to the tailings pond elements is determined according to the merged results to obtain the dry beach length, which is a key indicator for measuring the safety status of the tailings pond. Accurately grasping the dry beach length can help relevant personnel to promptly determine whether the tailings pond is in a safe operating state. For example, if the dry beach length is too short, it may indicate that the tailings pond has safety risks such as overtopping. This method can achieve accurate dry beach length measurement, providing reliable data support for safety warning and risk prevention and control of tailings ponds. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the specific embodiments of the present disclosure or the technical solutions in the prior art, the drawings required for use in the specific embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0017] Figure 1 A schematic diagram of an application environment of a method for identifying the dry beach length of a tailings pond in one embodiment; Figure 2 A schematic diagram of a process for identifying the dry beach length of a tailings pond in one embodiment; Figure 3 This is a schematic diagram of the process of step S206 in one embodiment; Figure 4 Schematic diagram of the process of step S310 in one embodiment; Figure 5 This is a schematic diagram of the process of step S208 in one embodiment; Figure 6A schematic diagram of a flow chart of a model training step in an embodiment; Figure 7 is a schematic diagram of the process of step S606 in one embodiment; Figure 8 It is a schematic block diagram of the structure of a device for identifying the dry beach length of a tailings pond in one embodiment; Fig. 9 is a schematic diagram of the internal structure of a computer device in one embodiment; Fig.10 Schematic diagram of the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solution and advantages of the present disclosure more clear, the present disclosure is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present disclosure and are not used to limit the present disclosure.

[0019] It should be noted that the terms "first", "second", etc. in the specification and claims of this article and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of this article described here can be implemented in an order other than those illustrated or described here. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, device, product or equipment that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or equipment.

[0020] In this article, the term "and / or" is only a description of the association relationship between related objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the related objects before and after are in an "or" relationship.

[0021] The present disclosure provides a method for identifying the dry beach length of a tailings pond, which can be applied to Figure 1In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated on the server 104, or it can be placed on the cloud or other network servers. The terminal 102 can obtain the remote sensing image of the tailings pond stored by the server 104. The terminal 102 can input the remote sensing image of the tailings pond into the pre-trained target detection model, and output the target detection result of the tailings pond through the target detection model. The terminal 102 uses the superpixel segmentation algorithm and the random forest algorithm to classify the multiple types of tailings pond elements contained in the target detection result of the tailings pond to obtain multiple types of tailings pond elements. The terminal 102 merges the same tailings pond elements of each type, determines the baseline corresponding to the tailings pond element according to the merged result, and obtains the dry beach length based on the reference line corresponding to the tailings pond element. Among them, the terminal 102 can be, but is not limited to, various personal computers, laptops, smart phones, tablet computers, Internet of Things devices and portable wearable devices, and the Internet of Things devices can be smart speakers, smart TVs, smart air conditioners, smart car-mounted devices, etc. The portable wearable device may be a smart watch, a smart bracelet, a head-mounted device, etc. The server 104 may be implemented as an independent server or a server cluster consisting of multiple servers.

[0022] In one embodiment, Figure 2 As shown in the figure, a method for identifying the dry beach length of a tailings pond is provided. Figure 1 The terminal 102 in the example is used as an example to illustrate, and the following steps are included: S202, obtaining remote sensing images of the tailings pond.

[0023] The tailings pond remote sensing image may be image data of the tailings pond obtained through aerial or satellite remote sensing technology, and may be a high-resolution remote sensing image.

[0024] S204, inputting the tailings pond remote sensing image into a pre-trained target detection model, and outputting the tailings pond target detection result via the target detection model.

[0025] Among them, the tailings pond target detection result can be a target box. The target detection model is an artificial intelligence model based on deep learning. Common ones include architectures based on convolutional neural networks (CNN), such as Faster R-CNN, YOLO (YouOnly Look Once) series, SSD (Single Shot MultiBox Detector), etc. After being trained on a large number of annotated data sets, these models can automatically identify specific targets in images or images, and determine their location (usually expressed in bounding box coordinates) and category information. In some embodiments of the present disclosure, the target detection model can be a YOLOV8 model.

[0026] Specifically, remote sensing images may exist in a variety of formats (such as TIFF, GeoTIFF, etc.), and they need to be converted into a format that the target detection model can accept, generally a common image format, such as JPEG, PNG, etc. At the same time, ensure that the image size meets the model input requirements. If the size is too large or too small, scaling operations may be required, such as scaling high-resolution large-size images to the input size specified by the model. Common input sizes can be 512×512 pixels, 1024×1024 pixels, etc. Normalize the pixel values ​​of the image and map them to a specific numerical range, usually between 0 and 1 or -1 to 1. This helps the model converge faster and better, improves training results and the accuracy of subsequent detection. There are many ways to normalize, such as calculating the mean and standard deviation of the image pixel values, and then converting them according to the corresponding formula. If the remote sensing image carries geographic coordinate information, and the target detection model itself does not directly process geographic coordinates, it may be necessary to convert the geographic coordinates into image plane coordinates, or to establish a correspondence between the geographic coordinates and the image coordinates in subsequent processing to facilitate accurate marking and positioning of the tailings pond in the actual geographic space.

[0027] The detection process of the target detection model includes: 1. Feature extraction: When the pre-processed tailings pond remote sensing image is input into the target detection model, the model will first use its internal convolutional layer and other structures to extract features from the image. For example, the convolutional layer slides convolution on the image through a series of convolution kernels to extract textures, shapes and other features at different levels. These features will be continuously abstracted and integrated as the network goes deeper, forming high-level semantic feature maps that are valuable for subsequent target detection. Like in Faster R-CNN, features are extracted through backbone networks (such as ResNet as feature extraction networks); the YOLO series relies on its own unique network structure to quickly extract features and perform target detection. 2. Target positioning and classification: Based on the extracted features, the model further performs target positioning and classification operations. Different models adopt different strategies. For example, Faster R-CNN uses the Region Proposal Network (RPN) to first generate a series of candidate regions that may contain targets, and then classify these candidate regions (determine whether they are tailings ponds or other objects) and make more precise position adjustments (refine the bounding box coordinates). The YOLO series divides the image into multiple grids, each of which directly predicts whether the target exists, as well as its location, category and other information. The detection results of all targets can be obtained through one forward propagation, which is relatively faster.

[0028] 3. Output detection results: Finally, the model will output the tailings pond target detection results, which usually include location information: the location of the tailings pond in the remote sensing image is presented in the form of a bounding box. The bounding box is generally represented by the coordinates of the upper left corner and the lower right corner (such as the pixel coordinates in the image coordinate system). Through these coordinates, the specific scope of the tailings pond in the image can be intuitively seen.

[0029] S206, using a superpixel segmentation algorithm and a random forest algorithm to classify the multiple types of tailings pond elements contained in the tailings pond target detection result to obtain the multiple types of tailings pond elements.

[0030] Among them, the superpixel segmentation algorithm aims to segment the image into a series of small areas with similar features. These small areas are called superpixels. Compared with traditional pixel-based processing, superpixels can greatly reduce the amount of data while retaining the main contours and structural information of objects in the image. It aggregates adjacent and similar pixels together to form superpixels through clustering and other methods based on the color, texture, brightness and other features of the image. Common superpixel segmentation algorithms include the SLIC (Simple Linear Iterative Clustering) algorithm. Random forest is an ensemble learning algorithm based on decision trees, which consists of multiple decision trees. In the training phase, multiple decision trees are constructed by random sampling with replacement of the training data set. Each decision tree learns and grows based on the sampled data. When the node splits, a part of the features is randomly selected from many features to determine the best splitting method. In the prediction phase, for new data samples, each decision tree will give a prediction result, and then the opinions of multiple decision trees are combined by voting (classification task) or averaging (regression task) to obtain the final prediction result. Random forests have the advantages of strong resistance to overfitting, insensitivity to noisy data, and the ability to process high-dimensional data. They are suitable for dealing with complex classification problems, such as classifying multi-category situations such as tailings pond elements. Tailings pond elements can include: dams, dry beaches, water bodies, etc.

[0031] Specifically, the remote sensing image output by the target detection model containing the tailings pond (that is, the image corresponding to the tailings pond target detection result) is segmented according to the superpixel segmentation algorithm, and the image is divided into multiple superpixel regions. The pixels within each superpixel region have high similarity in color, texture, etc. Different superpixel regions can correspond to different tailings pond elements or different parts of the same element. For example, for a tailings pond image, different superpixel regions corresponding to the tailings accumulation area, dam body, vegetation around the reservoir area, etc. may be segmented, and these areas will serve as the basic units for subsequent classification.

[0032] Feature extraction: For each superpixel area, extract the corresponding features. These features can include color features (such as average color value, color histogram, etc.), texture features (such as grayscale co-occurrence matrix features, local binary pattern features, etc.), geometric features (such as area, perimeter, shape factor, etc.) and spatial position features (such as the coordinate position of the superpixel area in the image, the relative position relationship with other areas, etc.). Through these rich feature descriptions, different tailings pond elements can be better distinguished, providing a basis for subsequent classification using the random forest algorithm. Collect a large amount of labeled tailings pond element sample data (which can be manually labeled by professionals or labeled by other reliable methods). These sample data contain various features previously extracted from the superpixel area and the corresponding tailings pond element real categories (such as labeled as tailings accumulation area, dam body, drainage facilities, surrounding vegetation, etc.). Divide these sample data into training set and validation set (the commonly used division ratio can be 7:3 or 8:2, etc.). The training set is used to train the random forest model, and the validation set is used to evaluate the performance of the model and adjust the parameters of the model (such as the number of decision trees, the maximum depth of each tree, etc.). Model training: Input the sample feature data in the training set into the random forest algorithm for training, so that the model can learn the correspondence between the features of different tailings pond elements and categories. During the training process, the random forest will construct multiple decision trees according to the set parameters, and each decision tree will continuously adjust its own structure and node splitting rules to classify the training samples as accurately as possible. For example, a decision tree may judge that it is more likely to belong to the tailings accumulation area category based on the texture and color features of the superpixel area. As the training progresses, the classification ability of each decision tree continues to improve, and the comprehensive classification performance of the entire random forest model is gradually improved. The superpixel area feature data in the pre-processed (superpixel segmentation and feature extraction) tailings pond target detection result image is input into the trained and evaluated and optimized random forest model. The model will predict the tailings pond element category to which each superpixel area belongs, and finally obtain the classification results of multiple types of tailings pond elements, such as clearly indicating which areas in the image belong to the tailings accumulation area and which are the dam body parts.

[0033] S208, merging the same tailings pond elements of each type, determining the reference line corresponding to the tailings pond element according to the merging result, and obtaining the dry beach length based on the reference line corresponding to the tailings pond element.

[0034] The reference lines may include: baseline, outer edge, etc. Dry beach refers to the slope area on the downstream slope of the tailings accumulation dam in the tailings pond where the beach surface changes from wet to dry. The length of dry beach is an important indicator for measuring the safety status of the tailings pond. It reflects the safety-related characteristics of the tailings pond, such as its buffering capacity in special situations such as floods.

[0035] Specifically, after obtaining the classification results of multiple types of tailings pond elements through the previous superpixel segmentation algorithm and random forest algorithm, the tailings pond elements of the same category must first be identified. For example, for multiple superpixel areas classified as tailings accumulation areas (these areas have been determined to belong to the same type of elements, namely tailings accumulation areas, in previous classification operations), they need to be integrated together to form a relatively complete representation of the tailings accumulation area. This process can be achieved by analyzing and processing the coordinate range, boundaries and other information of each area. For example, the boundaries of adjacent superpixel areas belonging to the same category are fused to construct a continuous area so that the reference line can be determined based on these merged similar element areas. The reference is used to calculate the dry beach length. The calculation of the dry beach length is based on the baseline, and the distance from the baseline to the outer edge of the tailings accumulation area in a specific direction (usually along the downstream slope of the dam body) is measured.

[0036] In the above-mentioned tailings pond dry beach length identification method, by obtaining the remote sensing image of the tailings pond and processing it with the pre-trained target detection model, the tailings pond related targets can be accurately located from the macroscopic remote sensing image level, avoiding the problems of limited vision and low efficiency that may exist in manual field surveys. For example, for tailings pond areas located in remote areas with large areas, relying on remote sensing images combined with target detection models, the specific location and approximate range of the tailings pond in the image can be quickly and accurately determined, thereby improving the comprehensiveness and accuracy of tailings pond target detection. This model-based automated detection method can process a large amount of remote sensing image data at the same time, and is suitable for monitoring multiple tailings ponds or large areas containing tailings ponds, which helps to achieve large-scale and batch tailings pond information extraction. The use of superpixel segmentation algorithm and random forest algorithm to classify the multiple types of tailings pond elements contained in the detected tailings pond target detection results can further refine the cognition of different elements inside the tailings pond, and can more deeply understand the details of the tailings pond, thereby improving the accuracy of dry beach identification. Merge the same tailings pond elements of each type, and then determine the reference line corresponding to the tailings pond elements based on the merged results to obtain the dry beach length, which is a key indicator for measuring the safety status of the tailings pond. Accurately grasping the dry beach length can help relevant personnel to promptly determine whether the tailings pond is in a safe operating state. For example, a short dry beach length may indicate that the tailings pond has safety risks such as overtopping. This method can achieve accurate dry beach length measurement, providing reliable data support for safety warning and risk prevention and control of the tailings pond. The above process can be repeated based on regularly acquired remote sensing images to continuously monitor the changes in the dry beach length, realize dynamic tracking of the safety status of the tailings pond, and facilitate timely adoption of corresponding maintenance and governance measures to ensure the safety of the environment around the tailings pond and the lives and property of the people.

[0037] In one embodiment, Figure 3As shown, the superpixel segmentation algorithm and the random forest algorithm are used to classify the multiple types of tailings pond elements contained in the tailings pond target detection result to obtain multiple types of tailings pond elements, including: S302, assigning the tailings pond target detection result to a coordinate reference system to obtain a target frame vector.

[0038] Specifically, the tailings pond target detection result can be assigned to the coordinate reference system to obtain the target frame vector in shp format.

[0039] In some exemplary embodiments, it is first necessary to select an appropriate coordinate reference system according to the geographical location of the tailings pond and the specific requirements of the project. For example, the commonly used geographic coordinate system is WGS84 (World Geodetic System 1984, widely used in global positioning and other scenarios); in terms of projection coordinate system, Gauss-Kruger projection coordinate system is often used in large-scale map drawing in my country. If the relevant monitoring and management work of the tailings pond needs to be matched and integrated with the existing local geographic information data (such as regional maps, surrounding infrastructure geographic information, etc.), then the corresponding coordinate reference system is preferred to ensure the consistency and compatibility of the data. In some embodiments of the present disclosure, there is no restriction on how to determine the specific coordinate reference system. After determining the specific coordinate reference system, it is necessary to obtain the parameters corresponding to the coordinate system. For example, for the projection coordinate system, it is necessary to clarify its projection mode (such as equiangular projection, equal-area projection, etc.), central meridian, projection belt and other parameter information. These parameters will be used for the subsequent conversion and correction of the coordinates in the target detection results so that they fall accurately under the selected coordinate reference system. The tailings pond target detection results output by the target detection model are usually presented in a specific format, which may be text format (such as a CSV file containing target category, location coordinates and other information, each line records a detected target related information), JSON format (storing target related attributes and coordinates in the form of key-value pairs) or a custom format. It is necessary to use the corresponding parser (Python and other programming languages ​​can be used) to read and understand these data, extract the coordinate information (generally pixel coordinates in the image coordinate system) corresponding to each target box (for example, the upper left corner coordinates and lower right corner coordinates are usually used to represent the position of a rectangular target box, or the center point coordinates and width and height are used to represent different forms). The pixel coordinates in the image coordinate system extracted from the target detection results are converted to the actual geographic coordinates in the selected coordinate reference system. By establishing a mapping relationship between pixel coordinates and geographic coordinates (common methods include building a transformation matrix based on mathematical models such as affine transformation according to the known image corner coordinates and corresponding pixel coordinates, and then using the matrix to convert the pixel coordinates of the target box into geographic coordinates), the coordinate conversion is realized, so that the coordinates of the target box have actual geographic location significance. Use a programming library related to the geographic information system (GIS) (such as the Geopandas library in Python, which is a library based on Pandas and extends the geospatial data processing capabilities, or the GDAL / OGR library, which is a powerful geographic data processing library) to create a new vector dataset object and define its data type as polygon (because the target box can usually be regarded as a rectangular polygon) or other appropriate geometric types such as points and lines (if the target detection results have special geometric representation requirements).Set the corresponding fields in this vector dataset object (such as adding a "target category" field to record whether the detected tailings pond target belongs to the dam body, dry beach or other categories, and you can also add other descriptive fields such as "target number" to facilitate subsequent data management and query). Add the target frames with converted coordinates one by one to the created vector dataset, and accurately enter the relevant attributes and geometric shape information of each target into the vector dataset according to the set geometry type (for polygons, use the coordinates of the four corner points of the converted target frame to construct a polygon object) and field information, forming a complete vector data content containing the tailings pond target information. Use the function functions provided by the relevant GIS library to save the constructed vector dataset as a shp format file. For example, in Geopandas, you can use the simple "to_file" method to specify the save path and file name suffix as ".shp" to complete the output operation, and finally obtain a tailings pond target frame vector file with a coordinate reference system and stored in shp format, which is convenient for subsequent visualization, spatial analysis, and integration with other geographic information data in GIS software (such as ArcGIS, QGIS, etc.).

[0040] S304: Expand the target frame vector based on a preset buffer distance to obtain an expanded target frame vector.

[0041] The buffer distance is set based on the size of the tailings pond target detection results, such as 10,000 km. 2 A large tailings pond is set with a buffer zone of 10m; 50,000km 2 A 50m buffer zone is set for a tailings pond of a certain size; the purpose is to extend the border outward to a certain extent to ensure that all elements are included in order to avoid the border failing to accurately cover the tailings pond. According to the size of the tailings pond target detection result, those skilled in the art can flexibly adjust the buffer distance, and in some embodiments of the present disclosure, the size of the buffer distance is not limited.

[0042] Specifically, a buffer distance may be set for the target frame vector, and the expanded target frame vector is obtained after expansion using the buffer distance.

[0043] S306, determining a seed center point in the expanded target frame vector, and merging similar pixels based on the seed point center by a clustering method to form a superpixel.

[0044] Among them, seed center points refer to some representative points pre-selected or determined by a certain algorithm in a specific area. They are usually used as the starting reference point or core point for subsequent processing (such as clustering, regional growth and other operations). The surrounding pixels, objects or data elements are classified, merged, expanded and other operations are performed around these seed center points to form regions, groups, etc. with specific characteristics or meeting specific requirements. Superpixels refer to pixel blocks with certain semantic information formed by clustering and merging pixels with similar characteristics (such as color, texture, brightness, etc.) in an image. It can be regarded as a kind of regional division of the original image at the pixel level, replacing a large number of original pixels with relatively fewer "superpixel blocks" to simplify the image representation while retaining most of the key visual information and structural features in the image.

[0045] Specifically, several points can be randomly and evenly selected as seed center points in the area covered by the extended target frame vector according to certain rules. For example, the number of seed points to be selected can be set according to the size of the target frame range (if the target frame range is large and the superpixel division is expected to be finer, more seed points can be selected; otherwise, fewer can be selected), and then implemented by programming (using Python as an example), using a random number generation function combined with the coordinate range of the target frame, to generate a corresponding number of coordinate points within the range as seed center points. The area corresponding to the extended target frame vector can also be divided into regular grids, and then the center point of each grid or a point at a specific position can be selected as the seed center point. For example, the target frame area is divided into small square grids at a fixed spacing (such as one grid every 10 meters, and the spacing size can be adjusted according to actual needs and data accuracy), and the center coordinate point of the grid is taken as the seed point. Then you can choose a suitable clustering method, such as K-Means clustering or other clustering methods, and then perform clustering with the seed center point as the cluster center. In the scene formed by superpixels, the pixel points can be regarded as data points to be clustered. By calculating the distance between the pixel and the seed center point (usually using distance metrics such as Euclidean distance), the cluster to which the pixel belongs (i.e., superpixel) is continuously adjusted, and finally the same type of pixels are merged.

[0046] In addition, you can also use superpixel segmentation algorithms, such as the SLIC (Simple Linear Iterative Clustering) algorithm for processing. SLIC is an algorithm specifically used for image superpixel segmentation, which combines spatial proximity and color similarity to divide superpixels. At the beginning of the algorithm, similar to the determination of seed center points mentioned above (but the SLIC algorithm has its own specific seed point initialization method, which is usually selected at uniform intervals on the pixel grid of the image and fine-tuned according to the color and other characteristics of the pixels), then these seed points are used as the center, within a certain search range (usually a square area with a fixed side length, the side length is related to the desired superpixel size), while considering the color distance and spatial distance of the pixels, and iteratively optimizing the superpixel area corresponding to each seed point, so that the pixels inside the superpixel are as similar as possible in color and space, and finally superpixel segmentation is achieved to obtain superpixels.

[0047] S308, detecting the edge of the super-pixel, extracting the center line, and converting the line to a surface to obtain a super-pixel segmentation surface.

[0048] Among them, edge detection algorithms may include: Canny edge detection algorithm, Laplacian operator, Sobel operator and other algorithms. Superpixel segmentation surface refers to a way of expressing each superpixel area in a planar geometric form after dividing an image into multiple superpixels. It is a series of operations such as superpixel edge detection, centerline extraction, line-to-surface conversion, etc. after the superpixel segmentation operation on the image, so that the originally discrete superpixels are presented in the form of continuous surfaces in space. Each surface represents a set of pixels with similar features (such as color, texture, etc.), that is, a superpixel.

[0049] Specifically, edge detection algorithms can be used to detect superpixel edges, and the detected superpixel edges can be regarded as a region with a certain width. The center line can be obtained by using a thinning algorithm or a skeletonization algorithm. The center line is then converted to a surface using a polygon construction method or using the functions of GIS software or related libraries to obtain a superpixel segmentation surface.

[0050] In some exemplary embodiments, the thinning algorithm aims to gradually "thinnen" the edge by continuously removing redundant pixels in the edge area, and finally leaving a center line with a single pixel width to represent the original edge. For example, the Zhang-Suen thinning algorithm is a classic thinning algorithm, which is based on the neighborhood information of the pixel points in the image (such as the connectivity of the pixels in the 8 neighborhoods, the distribution of black and white pixels, etc.), and through multiple iterations, gradually deletes the removable pixels according to certain rules until they can no longer be deleted, thereby obtaining the thinned center line. The skeletonization algorithm is similar to the thinning algorithm. The goal of skeletonization is also to extract a center line with a single pixel width that can represent the shape of the object, but its idea is more from the perspective of maintaining the topological structure and shape characteristics of the object. For example, the skeletonization method based on distance transformation first calculates the distance from each pixel in the image to the background (such as the Euclidean distance, etc.), and then determines the pixel point at the center of the object based on the distance information, and gradually constructs the skeleton of the object, that is, the center line.

[0051] Method of using polygon construction: For the extracted center line, it can be regarded as the boundary line or skeleton line of the polygon, and the area enclosed by these lines is constructed into a polygonal surface through a certain algorithm. A common practice is to track along the center line, determine the vertices of the polygon at key positions such as the turning point and end point of the center line, and then connect these vertices in a certain order to form a closed polygon, thereby obtaining a superpixel segmentation surface. For example, you can start from one end point of the center line and move along the direction of the line. Whenever you encounter a point where the direction changes (determined by calculating the slope change of adjacent line segments, etc.), record it as a vertex, and so on, until you return to the starting end point or can be closed and connected with other constructed polygons, and finally form polygonal surfaces to represent the superpixel segmentation surface. In addition, if the data related to the superpixel is with geographic coordinate information (such as in scenarios such as geographic remote sensing image analysis), you can use the functions of geographic information system (GIS) software (such as ArcGIS, QGIS, etc.) or related geospatial data processing libraries (such as Geopandas, GDAL / OGR in Python, etc.) to realize the line-to-surface operation. For example, in ArcGIS, after importing the centerline vector data (usually in .shp format), use the "Line to Polygon" tool under the "Feature" module in the "Data Management Tools" and follow the software prompts to set the input centerline feature layer, output polygon feature layer path and other parameters. Click the "OK" button to complete the line-to-surface operation and generate the corresponding superpixel segmentation surface layer. You can then view, edit, and integrate and analyze it with other geographic information data in the software.

[0052] S310, using a random forest algorithm to classify the superpixel segmentation surface to obtain multiple types of tailings pond elements.

[0053] Specifically, features in superpixel segmentation surfaces can be extracted, and then the unclassified superpixel segmentation surfaces can be classified using a pre-trained random forest model.

[0054] In this embodiment, the tailings pond target detection result is assigned to the coordinate reference system to obtain the target frame vector, so that the detected tailings pond related targets have clear and accurate location information in the geographic space. This is very important for various subsequent geospatial analyses and integration with other geographic information data (such as surrounding terrain, infrastructure distribution, etc.). For example, when conducting regional planning and environmental impact assessment, the specific location of the tailings pond in the entire region can be accurately known, which is convenient for accurately measuring its relationship with surrounding elements. Based on the preset buffer distance, the target frame vector is expanded to obtain the expanded target frame vector, which can simulate the impact area within a certain range around the tailings pond. For example, when considering the potential impact of the tailings pond on the surrounding ecological environment, resident safety, etc., by reasonably setting the buffer distance (such as determined according to the pollutant diffusion model, safety protection standards, etc.), the scope that needs to be focused on and analyzed can be clearly defined, which is helpful to formulate corresponding prevention and control measures in advance, and provides an intuitive and quantitative spatial basis for risk management. The seed center point is determined in the expanded target frame vector, and then similar pixels are merged to form super pixels through a clustering method, which greatly simplifies the representation of image data. The originally huge number of pixels are aggregated into a relatively small number of superpixels, which reduces the complexity of subsequent data processing and analysis, while retaining the main visual features and structural information in the image. For example, when analyzing a large area of ​​tailings ponds, if a large amount of original pixel data is directly processed, the computational and storage costs will be very high. The formation of superpixels can improve processing efficiency and make analysis more feasible while ensuring key information. As a relatively stable area with certain semantic information, the internal pixels of superpixels are similar, which facilitates the extraction of more representative and robust features. For example, the color, texture, geometry and other features of superpixels are extracted to describe the characteristics of different areas of the tailings pond. Compared with the feature extraction based on a single pixel, these superpixel-level features can better reflect the essential differences between different elements in the tailings pond (such as dam bodies, dry beaches, waterlogged areas, etc.), providing a high-quality feature basis for subsequent classification and other operations.

[0055] In one embodiment, Figure 4 As shown, the random forest algorithm is used to classify the superpixel segmentation surface to obtain multiple types of tailings pond elements, including: S402, determining the element characteristics of each type of tailings pond element based on the predetermined classification samples.

[0056] Among them, the classification samples mainly include sample point data (vector format) of tailings dams, dry beaches, water bodies and other elements. Classification samples can be pre-labeled. For example, there is an image, and superpixel segmentation surfaces are obtained through superpixel segmentation. Then, according to human visual observation, some sample points are selected on the surfaces of water, dams, dry beaches, etc., and the categories are marked to obtain classification samples for subsequent classification algorithms.

[0057] Specifically, the element characteristics of each type of tailings pond element can be determined according to the various characteristics in the predetermined classification samples.

[0058] S404: Using the random forest algorithm to learn the element features to form a superpixel classification model.

[0059] Specifically, the model can be trained using the random forest algorithm, the feature features, and each type of tailings pond corresponding to the feature features to obtain a superpixel classification model.

[0060] In some exemplary embodiments, Python's Scikit-learn library is used to build a random forest model. First, the corresponding module is imported and a random forest classifier object is created. At the same time, some key parameters of the model can be set, such as the number of decision trees (n_estimators), the maximum depth of each tree (max_depth), the minimum number of sample splits (min_samples_split), etc. The selection of parameters can first set the initial value based on experience, and then optimize it through cross-validation and other methods.

[0061] S406, using the superpixel classification model to classify the superpixel segmentation surface to obtain multiple types of tailings pond elements.

[0062] Among them, the tailings pond elements can be represented by element vector pixel blocks in some embodiments of the present disclosure.

[0063] Specifically, the superpixel classification model may be used to classify the superpixel segmentation surface to ultimately obtain element vector pixel blocks within the range of each tailings pond.

[0064] In this embodiment, the superpixel segmentation surface is classified using the formed superpixel classification model, and finally multiple types of tailings pond elements can be obtained. The tailings dam, dry beach, waterlogging area, vegetation around the reservoir and other parts can be clearly distinguished, and their area changes, spatial distribution changes and other changes can be grasped in real time to ensure the accuracy of dry beach identification.

[0065] In one embodiment, Figure 5As shown, the tailings pond elements include: dam elements, dry beach elements, and water body elements; the same tailings pond elements of each type are merged, and the reference line corresponding to the tailings pond element is determined according to the merged result. Based on the reference line corresponding to the tailings pond element, the dry beach length is obtained, including: S502, merging the same tailings pond elements among the multiple types of tailings pond elements to obtain a merging result of each tailings pond element.

[0066] S504, using a curve simplification algorithm to process the merged result of the accumulation dam elements to obtain an initial dam baseline.

[0067] S506, determining the outer boundary of the water body element, and determining the length of the dry beach based on the distance between the initial dam baseline and the outer boundary of the water body.

[0068] Among them, the initial dam baseline usually refers to a representative curve obtained by simplifying the geometric shape of the tailings dam. It can reflect the key geometric features of the main body of the dam, such as the direction and contour. It can be regarded as a line extracted from the complex actual boundary line of the dam to reflect its core form. To a certain extent, it is similar to the basic contour line of the bottom or key position of the dam. The outer edge of the water body is essentially a definition of the scope of the water body. It outlines the boundary between the water body and the surrounding tailings dam, dry beach, land around the reservoir and other different elements in the form of a line. From a geometric point of view, if the water body is regarded as a plane area (usually represented by a polygon), then the outer edge of the water body is the boundary contour line of this polygon. It clarifies the distribution range of the water body on the two-dimensional plane, just like drawing a "boundary line" for the water body area, so that it can be clearly distinguished in space.

[0069] Specifically, based on the category field information of multiple types of tailings pond elements, the same elements are merged to obtain the merged results of each tailings pond element (the final vector surface of each element); the dam vector (dam element) is processed by the Douglas–Peucker algorithm to obtain the initial dam baseline vector; the distance between the initial dam baseline vector and the outer edge of the water body element in the corresponding vertical direction is traversed, and the shortest distance is taken and assigned to the corresponding dry beach vector attribute table, finally the dry beach length of each tailings pond is obtained.

[0070] In some exemplary embodiments, the data structure of multiple types of tailings pond elements can be determined. For the data structure representing the tailings pond elements, it is common to exist in the form of polygons (such as areas representing tailings dams, dry beaches, waterlogging areas, etc.), points (such as certain specific monitoring points, etc.) or lines (such as the boundary line of the dam body, etc.). Merge the elements of the same type. Then select the Douglas–Peucker algorithm or the Visvalingam-Whyatt algorithm to process the merged results of the accumulation dam elements to obtain a simplified curve, i.e., the initial dam baseline. If it is in GIS software (taking ArcGIS as an example): load the vector data layer containing water body elements (such as waterlogging areas, etc.) into ArcGIS. Water body elements generally exist in the form of polygons. Through the "Boundary" tool under the "Feature" module in the "Data Management Tools", select the water body feature layer as the input feature, specify the path and name of the output boundary feature layer (usually saved in .shp format), and click the "OK" button. ArcGIS will extract the outer boundary line of each water body polygon, that is, the outer boundary line of the water body. These boundaries are also stored in the form of vector lines, which can be easily used for subsequent spatial analysis and distance calculation operations. Use the "Neighbor Analysis" tool under the "Distance Analysis" module in the "Analysis Tools" of ArcGIS, use the initial dam baseline as the input feature and the outer boundary line of the water body as the neighboring feature, set the path and name of the output distance result table and other parameters, click the "OK" button, ArcGIS will calculate the shortest distance from the dam baseline to each outer boundary line of the water body, and record the results in the output table. Then, by checking the table data or further visualizing it in the software, find the distance value that meets the requirements (such as taking the minimum value, because the dry beach length usually refers to the shortest distance from the dam body to the water body; or determine the value method based on actual business rules and analysis needs), which is the dry beach length.

[0071] If it is done programmatically (using Python combined with the Geopandas library as an example): Assuming that the vector data of the water body features has been loaded into a GeoDataFrame object (such as water_gdf), extract its boundary lines by traversing each water body polygon. First, ensure that the initial dam baseline and the outer edge of the water body are in the same coordinate reference system. Then, by traversing each line object of the outer edge of the water body, calculate its minimum distance from the initial dam baseline.

[0072] In this embodiment, identical tailings pond elements in multiple types of tailings pond elements are merged, and scattered similar elements are integrated into a unified geometric object, which greatly simplifies the data structure. By merging similar elements, the overall characteristics of each type of tailings pond element can be more conveniently extracted. The initial dam baseline is obtained by processing the merged results of the accumulation dam elements using a curve simplification algorithm, which can highlight the key geometric features of the accumulation dam. As a simplified representative curve, the initial dam baseline filters out some trivial details that may exist in the original dam body boundary data and have little effect on the overall shape, making the main outline and direction of the dam body clearer. By determining the outer edge of the water body element and based on the distance between the initial dam baseline and the outer edge of the water body, the dry beach length can be determined more accurately.

[0073] In one embodiment, Figure 6 As shown, the method also includes: S602, acquiring a training remote sensing image, wherein the training remote sensing image contains a plurality of tailings pond areas.

[0074] S604, determining the tailings pond area contained in the training remote sensing image, drawing a vector frame and assigning values ​​for the tailings pond area, and obtaining a vector boundary of a tailings pond target detection frame.

[0075] S606: Training a target detection model based on the tailings pond target detection box vector boundary and the training remote sensing image.

[0076] Specifically, cloud-free high-resolution training remote sensing images of each phase are selected, and the mining area is selected as the training remote sensing image. The resolution of the training remote sensing image is not less than 2m. Use ArcGIS software to create a new tailings pond vector frame surface vector; for the tailings pond range, draw a vector frame, and assign the attribute table to 1 to obtain the tailings pond target detection frame vector boundary. The sliding window algorithm can also be used to locate the center of the target frame and expand it to form a sample slice of 1024*1024 pixels. Convert the longitude and latitude coordinates of the vector to pixel coordinates in the slice, and save it as a txt file with the same name as the corresponding sample slice to obtain the tailings pond sample slice and target frame location information. Use the tailings pond sample slice and target frame location information to train the target detection model.

[0077] In one embodiment, the target detection model includes a backbone network, a neck network and a head network, and the backbone network includes a rectangular feature extraction module. The backbone network includes: CSPDarknet (rectangular feature extraction module) consists of a standard convolution layer (convolution with stride) and a pooling layer for preliminary feature extraction and downsampling. Each CSPBlock contains a residual network and a direct transmission path for combining the features of the two. By introducing multi-scale feature fusion, targets of different sizes can be better captured, and the feature map is finally output to the subsequent network layer for target detection or classification tasks. The neck network PAFPN is designed to enhance the detection capability of small and large targets through more efficient multi-scale feature fusion. It is mainly achieved through the following key parts: Feature Pyramid Network (FPN): used for top-down feature fusion to ensure that high-level feature information can be combined with low-level features; Path Aggregation Network (PANet): used for bottom-up feature fusion to further aggregate features of different scales and enhance the expressiveness of the model. At the same time, PAFPN introduces more convolution and connection operations to ensure that information can be efficiently transmitted at multiple scales. The head network is responsible for generating the final detection results. The detection head usually contains several parallel convolution layers, which are used to process the feature maps provided by the neck network and output the detection box and classification results on each feature map. In YOLOv8, the detection head adopts an anchor-free design to directly predict the coordinates and category probability of the target box, which simplifies the complexity of the model and improves the detection speed. Figure 7 As shown, the target detection model is trained based on the tailings pond target detection box vector boundary and the training remote sensing image, including: S702, based on the backbone network, extracting the tailings pond feature information contained in the vector boundary of the tailings pond target detection box in the training remote sensing image.

[0078] Specifically, the backbone network can be used to first extract the tailings pond feature information contained in the vector boundary of the tailings pond target detection box in the training remote sensing image, and the tailings pond feature information can be used for subsequent processing.

[0079] S704, performing secondary feature extraction on the tailings pond feature information based on the rectangular feature extraction module to obtain high-order feature information of the tailings pond.

[0080] Specifically, since there may be some feature information with a large difference in the length-to-width ratio in the tailings pond feature information, the feature information with a small difference in length and width can be well extracted using the feature module in the original backbone network. The rectangular feature extraction module is used to extract the feature information again on this basis to ensure that the features contained in the tailings pond feature information with a large difference in length and width are identified, thereby obtaining high-order feature information of the tailings pond.

[0081] S706, performing dimension upgrading on the high-order feature information of the tailings pond and fusing it with the feature information of the tailings pond, and using the fused feature information and the vector boundary of the tailings pond target detection box to train the target detection model.

[0082] Specifically, since the high-order feature information of the tailings pond is obtained by secondary feature extraction of the tailings pond feature information, the high-order feature information of the tailings pond and the tailings pond feature information have different dimensions. They can be upgraded so that they have the same dimension as the tailings pond feature information, and then the upgraded features are added and normalized to obtain the final feature extraction result (fused feature information). The feature extraction result is part of the model training data, which can be understood as extracting parameters such as the shape and texture of the image and calculating them together with the target label. Then, the fused feature information and the vector boundary of the tailings pond target detection box are used to train the neck network and the head network to obtain the final trained target detection model.

[0083] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0084] Based on the same inventive concept, the embodiment of the present disclosure also provides a tailings pond dry beach length identification device for implementing the tailings pond dry beach length identification method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more tailings pond dry beach length identification device embodiments provided below can refer to the limitations of the tailings pond dry beach length identification method above, and will not be repeated here.

[0085] In one embodiment, Figure 8As shown, a tailings pond dry beach length identification device 800 is provided, comprising: an image acquisition module 802, a model processing module 804, a classification module 806 and a length calculation module 808, wherein: An image acquisition module 802 is used to acquire remote sensing images of the tailings pond; The model processing module 804 is used to input the tailings pond remote sensing image into a pre-trained target detection model, and output the tailings pond target detection result through the target detection model; A classification module 806 is used to classify multiple types of tailings pond elements contained in the tailings pond target detection result by using a superpixel segmentation algorithm and a random forest algorithm to obtain multiple types of tailings pond elements; The length calculation module 808 is used to merge the same tailings pond elements of each type, determine the reference line corresponding to the tailings pond element according to the merging result, and obtain the dry beach length based on the reference line corresponding to the tailings pond element.

[0086] In one embodiment of the apparatus, the classification module 806 includes: A data processing module, used for assigning the tailings pond target detection result to a coordinate reference system to obtain a target frame vector; An expansion module, used to expand the target frame vector based on a preset buffer distance to obtain an expanded target frame vector; A merging module, used to determine a seed center point in the expanded target frame vector, and to merge similar pixels based on the seed point center through a clustering method to form a superpixel; A segmentation module is used to detect the edge of the superpixel and obtain a superpixel segmentation surface by extracting the center line and converting the line to the surface; The classification submodule is used to classify the superpixel segmentation surface using a random forest algorithm to obtain multiple types of tailings pond elements.

[0087] In one embodiment of the device, the classification submodule is further used to determine the element characteristics of each type of tailings pond element based on predetermined classification samples; use the random forest algorithm to learn the element characteristics to form a superpixel classification model; and use the superpixel classification model to classify the superpixel segmentation surface to obtain multiple types of tailings pond elements.

[0088] In one embodiment of the device, the tailings pond elements include: an accumulation dam element, a dry beach element, and a water body element; the length calculation module 808 is further used to merge the same tailings pond elements among the multiple types of tailings pond elements to obtain a merged result of each tailings pond element; the merged result of the accumulation dam element is processed using a curve simplification algorithm to obtain an initial dam baseline; the outer edge of the water body element is determined, and the dry beach length is determined based on the distance between the initial dam baseline and the outer edge of the water body element.

[0089] In one embodiment of the device, the device also includes: a model training module, used to obtain a training remote sensing image, wherein the training remote sensing image contains multiple tailings pond areas; determine the tailings pond area contained in the training remote sensing image, draw a vector box and assign a value to the tailings pond area, and obtain a tailings pond target detection box vector boundary; based on the tailings pond target detection box vector boundary and the training remote sensing image, train a target detection model.

[0090] In one embodiment of the device, the target detection model contains a backbone network, a neck network and a head network, and the backbone network contains a rectangular feature extraction module; the model training module is also used to extract the tailings pond feature information contained in the tailings pond target detection box vector boundary in the training remote sensing image based on the backbone network; based on the rectangular feature extraction module, secondary feature extraction is performed on the tailings pond feature information to obtain high-order feature information of the tailings pond; the high-order feature information of the tailings pond is dimensionally upgraded and then fused with the tailings pond feature information, and the target detection model is trained using the fused feature information and the tailings pond target detection box vector boundary.

[0091] Each module in the tailings pond dry beach length identification device can be implemented in whole or in part by software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each module.

[0092] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Fig. 9As shown. The computer device includes a processor, a memory and a network interface connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store remote sensing image data of the tailings pond. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for identifying the dry beach length of a tailings pond is implemented.

[0093] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Fig.10 As shown. The computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a method for identifying the dry beach length of a tailings pond is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a key, trackball or touchpad set on the computer device shell, or an external keyboard, touchpad or mouse.

[0094] Those skilled in the art will understand that Fig. 9 The structure shown in or 10 is merely a block diagram of a partial structure related to the scheme of the present disclosure, and does not constitute a limitation on the computer device to which the scheme of the present disclosure is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0095] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0096] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0097] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0098] It should be noted that the tailings pond remote sensing images involved in the present disclosure (including but not limited to tailings pond remote sensing images used for training and tailings pond remote sensing images used for identification) are all information and data authorized by the user or fully authorized by all parties.

[0099] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided by the present disclosure can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided by the present disclosure may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided by the present disclosure may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic unit, a data processing logic unit based on quantum computing, etc., but are not limited to this.

[0100] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, 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.

[0101] The above-described embodiments only express several implementation methods of the present disclosure, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present invention. It should be pointed out that, for a person of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present disclosure, and these all belong to the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure shall be subject to the attached claims.

Claims

1. A method for identifying the dry beach length of a tailings pond, characterized in that: The method comprises: Obtain remote sensing images of tailings ponds; Inputting the tailings pond remote sensing image into a pre-trained target detection model, and outputting the tailings pond target detection result via the target detection model; A superpixel segmentation algorithm and a random forest algorithm are used to classify multiple types of tailings pond elements contained in the tailings pond target detection result to obtain multiple types of tailings pond elements; The same tailings pond elements of each type are merged, and the reference line corresponding to the tailings pond element is determined according to the merged result. Based on the reference line corresponding to the tailings pond element, the dry beach length is obtained.

2. The method according to claim 1, characterized in that The superpixel segmentation algorithm and the random forest algorithm are used to classify the multiple types of tailings pond elements contained in the tailings pond target detection result to obtain multiple types of tailings pond elements, including: Assigning the tailings pond target detection result to a coordinate reference system to obtain a target frame vector; Based on a preset buffer distance, the target frame vector is expanded to obtain an expanded target frame vector; Determine a seed center point in the expanded target frame vector, and merge similar pixels based on the seed center point through a clustering method to form a superpixel; Detecting the edge of the superpixel, and obtaining a superpixel segmentation surface by extracting the center line and converting the line to the surface; The random forest algorithm is used to classify the superpixel segmentation surface to obtain multiple types of tailings pond elements.

3. The method according to claim 2, characterized in that The random forest algorithm is used to classify the superpixel segmentation surface to obtain multiple types of tailings pond elements, including: Determine the element characteristics of each type of tailings pond element based on the predetermined classification samples; Using the random forest algorithm to learn the feature features and form a superpixel classification model; The superpixel classification model is used to classify the superpixel segmentation surface to obtain multiple types of tailings pond elements.

4. The method according to claim 1, characterized in that: The tailings pond elements include: a dam element, a dry beach element, and a water body element; the same tailings pond elements of each type are merged, and a reference line corresponding to the tailings pond element is determined according to the merged result. Based on the reference line corresponding to the tailings pond element, the dry beach length is obtained, including: Merging the same tailings pond elements among the multiple types of tailings pond elements to obtain a merged result of each tailings pond element; The combined result of the accumulation dam elements is processed by using a curve simplification algorithm to obtain an initial dam baseline; The outer boundary of the water body element is determined, and the length of the dry beach is determined based on the distance between the initial dam baseline and the outer boundary of the water body element.

5. The method according to claim 1, characterized in that The method further comprises: Acquire a training remote sensing image, wherein the training remote sensing image contains a plurality of tailings pond areas; Determine the tailings pond area contained in the training remote sensing image, draw a vector frame and assign values ​​to the tailings pond area, and obtain a vector boundary of the tailings pond target detection frame; A target detection model is trained based on the tailings pond target detection box vector boundary and the training remote sensing image.

6. The method according to claim 5, characterized in that The target detection model includes a backbone network, a neck network and a head network, and the backbone network includes a rectangular feature extraction module; The training of the target detection model based on the tailings pond target detection box vector boundary and the training remote sensing image includes: Based on the backbone network, extracting the tailings pond feature information contained in the tailings pond target detection frame vector boundary in the training remote sensing image; Performing secondary feature extraction on the tailings pond feature information based on the rectangular feature extraction module to obtain high-order feature information of the tailings pond; The high-order feature information of the tailings pond is fused with the feature information of the tailings pond after dimension upgrading, and the target detection model is trained using the fused feature information and the vector boundary of the tailings pond target detection box.

7. A tailings pond dry beach length identification device, characterized in that: The device comprises: Image acquisition module, used to obtain remote sensing images of tailings ponds; A model processing module, used for inputting the tailings pond remote sensing image into a pre-trained target detection model, and outputting the tailings pond target detection result via the target detection model; A classification module is used to classify multiple types of tailings pond elements contained in the tailings pond target detection result by using a superpixel segmentation algorithm and a random forest algorithm to obtain multiple types of tailings pond elements; The length calculation module is used to merge the same tailings pond elements of each type, determine the reference line corresponding to the tailings pond element according to the merging result, and obtain the dry beach length based on the reference line corresponding to the tailings pond element.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images