Mining area ground object target detection method, device, medium and product

Abstract

The invention discloses a mining area ground object target detection method and device, a medium and a product, and relates to the field of target detection, and the method comprises the steps: obtaining a mining area image; constructing a ground object target detection model by taking YOLOv8s as a reference model; and inputting the mining area image into the ground object target detection model to obtain a ground object target detection result. The spatial distribution of the ground feature data of the mining area can be quickly and accurately obtained, and the detection precision of the ground feature target of the mining area is improved.

Description

Technical Field

[0001] The present application relates to the field of target detection, and in particular to a method, equipment, medium and product for detecting ground objects in mining areas. Background Art

[0002] With the development of economy and technology, the number of large open-pit mines is also increasing day by day. As the shallow mineral resources are exhausted, more and more mining companies choose to gradually shift their mining focus to deeper underground resources. However, uninterrupted open-pit mining activities have brought great impacts on sustainable mining, soil structure, climate and environment. In this context, timely and comprehensive understanding and investigation of the distribution of land features in open-pit mining areas, comprehensive remediation of mining areas in accordance with the law and strengthening ecological restoration of mining areas can minimize the impact of mining activities on the environment.

[0003] The monitoring and management of open-pit mines is not only closely related to national energy security, but also closely related to environmental governance. In the past, the monitoring of mining areas could only be carried out manually. This method requires a lot of manpower and material resources, and takes a lot of time to complete. Due to the impact of rockfall and secondary disasters, this method is dangerous. With the development of remote sensing technology, high-resolution images are widely used in practice. High-resolution remote sensing images have rich representation and edge detail information. Using the rich information of high-resolution images, we can understand and monitor the changes and characteristics of the earth's surface more comprehensively and accurately. Regular observation and analysis of mining areas using remote sensing technology can obtain key information and monitor changes in the mining environment. Remote sensing technology provides high-resolution and wide-coverage information for mining area monitoring by acquiring data provided by drones, satellites, aviation or ground sensors.

[0004] Remote sensing technology can obtain the distribution of ground objects in open-pit mines. By analyzing the spectral data of different bands, the characteristics of ore storage areas, mining areas and waste storage areas in open-pit mines can be accurately identified. Secondly, remote sensing technology can detect the surface deformation of the mining area. By comparing remote sensing images at different times, surface subsidence, uplift or landslides can be observed. In the early days, the Landsat series of satellites were mainly used to monitor open-pit mines. However, in recent years, by obtaining images and data collected by platforms such as remote-controlled aircraft or drones, the mining environment can be surveyed and monitored more quickly in multiple temporal phases. With the development of remote sensing technology, the integrated monitoring network of space, land and air combines remote sensing platforms and technologies to form a multi-level, multi-angle and multi-scale remote sensing monitoring network. This integrated monitoring network can realize all-round and all-temporal monitoring of the mining environment, providing more comprehensive and diversified data sources. At the same time, the application of deep learning and remote sensing technology has also brought new breakthroughs in remote sensing monitoring of open-pit mines. Remote sensing technology uses the computing and storage capabilities of cloud platforms to process and analyze large-scale remote sensing data and accelerate the extraction and interpretation of mining area data. Deep learning can train and learn from a large amount of remote sensing image data to achieve automated mining area information extraction and change monitoring, thereby improving monitoring efficiency and accuracy.

[0005] Deep learning image algorithms can be used for tasks such as image classification, image segmentation, and target detection. Among them, target detection is a task that achieves object classification and positioning at the same time, which can be regarded as a combination of classification and regression problems. At present, researchers have a small target selection range in remote sensing object recognition for open-pit mining scenes, mainly focusing on mining sites, open-pit mining roads, tailings ponds, and surface vegetation in open-pit mining areas, which makes it impossible to accurately detect mining area objects. Summary of the invention

[0006] In order to understand the distribution of ground objects, resource exploitation conditions and potential dangerous situations in open-pit mines in real time, the present application provides a method, equipment, medium and product for detecting ground objects in mines, which can quickly and accurately obtain the spatial distribution of ground object data in mines and improve the accuracy of ground object detection in mines.

[0007] To achieve the above objectives, this application provides the following solutions:

[0008] In a first aspect, the present application provides a method for detecting ground objects in a mining area, comprising:

[0009] Acquire images of the mining area;

[0010] Build a ground object detection model using YOLOv8s as the benchmark model;

[0011] The mining area image is input into the ground object target detection model to obtain a ground object target detection result.

[0012] Optionally, a ground object detection model is constructed using YOLOv8s as the benchmark model, including:

[0013] Obtain image data and public data in real scenes of mining areas; the image data are remote sensing images collected by drones and Gaofen-2 satellites; the public data are obtained by screening historical data of Google Earth images;

[0014] Annotating and preprocessing the image data and the public data to obtain an annotated data set;

[0015] Build the initial model using YOLOv8s as the benchmark model;

[0016] The initial model is trained using the labeled data set until the trained initial model meets a preset condition, thereby obtaining the trained initial model;

[0017] The trained initial model is used as the ground object detection model.

[0018] Optionally, build an initial model using YOLOv8s as the baseline model, including:

[0019] The spatial pyramid pooling structure of the benchmark model is improved by using the serialized ASPP structure and the ASPP-SP structure;

[0020] The DC2f structure is adopted and deformable convolution is introduced to improve the cross-stage local network structure of the baseline model to obtain the initial model.

[0021] Optionally, the serialized ASPP structure is suitable for multi-scale target features in mining area object detection by serializing the dilation rate of the void convolution.

[0022] Optionally, the ASPP-SP structure improves the global pooling operation by introducing a coordinate attention mechanism; the ASPP-SP structure uses an adaptive average pooling layer to perform pooling operations on the input feature map in the horizontal direction and the vertical direction respectively; the features are divided into two branches along the horizontal direction and the vertical direction; each branch performs convolution, batch normalization and activation function processing on the pooled input feature map to generate attention weights; the input feature map is multiplied by the attention weights of the two branches to obtain an output result.

[0023] Optionally, the preprocessing includes one or more of resizing, denoising, contrast enhancement, color space conversion, image cropping and rotation, normalization and data enhancement.

[0024] Optionally, the image data and the public data are annotated with a bounding box; the bounding box only contains a unique target area, and the bounding boxes of the same target area are unique and do not overlap.

[0025] In a second aspect, the present application provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of any one of the above-described methods for detecting mining objects.

[0026] In a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the above-mentioned methods for detecting ground objects in mining areas.

[0027] In a fourth aspect, the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of any of the above-mentioned methods for detecting mining objects.

[0028] According to the specific embodiments provided in this application, this application has the following technical effects:

[0029] The present application provides a method, device, medium and product for detecting ground objects in mining areas. By constructing a ground object detection model based on YOLOv8s as a benchmark model, the spatial distribution of ground object data in mining areas can be obtained quickly and accurately, thereby improving the accuracy of ground object detection in mining areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0031] Figure 1 This is an application environment diagram of a method for detecting ground objects in a mining area in an embodiment of the present application;

[0032] Figure 2 A schematic diagram of a process flow of a method for detecting ground objects in a mining area provided in one embodiment of the present application;

[0033] Figure 3 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0034] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0035] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0036] The mining area ground object target detection method provided in the embodiment of the present application can be applied to Figure 1 In 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 set up separately, or integrated on the server 104, or placed on the cloud or other servers. The terminal 102 can send the mining area image to the server 104. After the server 104 receives the mining area image, for the mining area image, the server 104 inputs the mining area image into the ground object detection model to obtain the ground object target detection result. Among them, the ground object target detection model is constructed with YOLOv8s as the benchmark model. The server 104 can feedback the obtained ground object target detection result to the terminal 102. In addition, in some embodiments, the mining area ground object target detection method can also be implemented by the server 104 or the terminal 102 alone, such as the terminal 102 can directly perform mining area ground object target detection on the mining area image, or the server 104 can obtain the mining area image from the data storage system and perform mining area ground object target detection on the mining area image.

[0037] The terminal 102 may be, but is not limited to, various desktop computers, laptop computers, smart phones, tablet computers, IoT devices, and portable wearable devices. The IoT devices may be smart speakers, smart TVs, smart air conditioners, smart vehicle-mounted devices, etc. The portable wearable devices may be smart watches, smart bracelets, head-mounted devices, etc. The server 104 may be implemented as an independent server or a server cluster consisting of multiple servers, or may be a cloud server.

[0038] In an exemplary embodiment, Figure 2 As shown, a method for detecting ground objects in a mining area is provided. The method is executed by a computer device, and can be executed by a computer device such as a terminal or a server alone, or by a terminal and a server together. In the embodiment of the present application, the method is applied to Figure 1 The server 104 in the example is used as an example to illustrate, including the following steps 200 to 202. Among them:

[0039] Step 200: Acquire a mining area image.

[0040] Step 201: Build a ground object detection model using YOLOv8s as the benchmark model.

[0041] Step 202: Input the mining area image into the ground object detection model to obtain the ground object detection result.

[0042] By implementing the above steps 200 to 202, the spatial distribution of mining area ground object data can be quickly and accurately acquired, thereby improving the detection accuracy of mining area ground object targets.

[0043] The YOLOv8 used above is a new single-stage target detection version released by Ultralytics on January 10, 2023. It includes a new backbone network, anchor-free detection head, and optimized loss function. YOLOv8 includes different versions: YOLOv8-n, YOLOv8-s, YOLOv8-m, YOLOv8-l, etc.

[0044] The Backbone module of YOLOv8 is the basic component of the network, which consists of five CBS convolution modules, four C2f modules and one SPPF module. The core module in the YOLOv8 network is the CBS module, which consists of a standard convolution layer (CNN), a normalization layer (BN) and an activation function (SiLU). In order to improve performance and reduce the size of the network, Backbone also uses residual connections and bottleneck structures. Compared with the C3 (CSPBottleneckWith 3Convolutions) module of YOLOv5, the Backbone of YOLOv8 uses a lighter C2f (CSPBottleneckWith2 Convolutions) module with fewer parameters and more powerful feature extraction capabilities.

[0045] The Neck module of YOLOv8 adopts a comprehensive multi-scale feature fusion technology, combining the SPPF module, the FPN module and two PAN modules to better understand the different scale information of the image. Specifically, FPN is used to extract features from images of different scales, while PAN uses a cross-layer network to aggregate these features at different levels, thereby enhancing the network's ability to detect objects of different sizes and resolutions. Overall, the Neck module provides the YOLOv8 network with a more comprehensive visual understanding capability by comprehensively utilizing feature information at multiple scales, thereby improving the performance of target detection.

[0046] YOLOv8 improves the original detection head structure by introducing the concept of decoupled-head. In the traditional detection head, the three tasks of classification, confidence and target position regression are processed simultaneously in one module. However, the localization task and the classification task have different preferences in network design, so it is not reasonable to put them in the same head for prediction. To solve this problem, YOLOv8 adopts an innovative method to split the original detection head into two parts, thereby achieving task decoupling. The classification task and regression task prediction use two CBS modules and one ordinary convolution module respectively, and the final prediction results will be spliced ​​together for output. After obtaining the output results, YOLOv8 classification loss uses VFLLoss, while regression loss uses DFLLoss and CIOU Loss at the same time. The combined use of these loss functions improves YOLOv8's classification accuracy and position regression accuracy for targets. YOLOv8 also adopts the Task-AlignedAssigner label matching strategy to dynamically correspond the true target box to the positive sample prediction box.

[0047] The anchor-free concept of YOLOv8 has great potential in the field of remote sensing object recognition. Compared with other YOLO series algorithms, the uniqueness of YOLOv8 is that it does not require the number and size of anchor boxes to be set in advance. Instead, it can directly predict the possible location and category of the target on the feature map. The advantage of this method is that it simplifies the design of hyperparameters, reduces the computational burden brought by anchor boxes, and can adapt to multi-scale targets, thereby improving the robustness and reasoning speed of the model.

[0048] In the public target detection COCO dataset, the YOLOv8 model has the highest detection accuracy compared to the mainstream YOLO series algorithms. At the same time, after experimental verification, the YOLOv8 model has excellent detection performance in remote sensing ground object target recognition tasks. Therefore, in the above step 201, this application selects YOLOv8 as the benchmark model, and improves on the basis of YOLOv8 to realize open-pit mine area ground object target detection.

[0049] In another exemplary embodiment of the present application, in order to further improve the accuracy of ground object detection, the implementation process of the above step 201 may include:

[0050] (1) Obtain image data and public data of real mining areas. Image data are remote sensing images collected by drones and Gaofen-2 satellites. Public data are obtained by screening historical data of Google Earth images.

[0051] In the actual application process, an open-pit mine is taken as an example to illustrate the dataset based on image data and public data. The dataset can be obtained by conducting a field survey of the open-pit mine, summarizing the types of objects in the mine and the characteristics of the distribution area, and producing a MAOD (Mining Area Object Detection) dataset. The dataset mainly focuses on the open-pit mine and the surrounding areas of the open-pit mine, and takes landslides, slag piles, farmland, buildings, and vehicles as detection targets. There are two main ways to obtain this dataset: one is to obtain the open-pit mine data of an open-pit mine in 2022 through drones and Gaofen-2 satellites. The imaging accuracy of this data is less than 1 meter, which belongs to high-resolution remote sensing images. A total of 8 remote sensing images are obtained according to different time periods and different regions. Using Python's OpenCV library, the images in the 8 remote sensing images are cropped using sliding windows to obtain 600 remote sensing images. After cleaning the 600 remote sensing images and removing the images without targets, we finally got 234 images that needed to be labeled. However, in actual production, there are fewer ground objects in open-pit mines and the environment is single, so the proportion of landslides, buildings, farmland, etc. in the data is very small. In addition, in actual use, the number of remote sensing images can be determined according to actual needs.

[0052] The second method is to automatically collect historical data of Google Earth images from 2012 to 2022, and collect a total of 2584 images. Invalid image data is screened out again, and 1849 images containing detection targets are obtained as public data. In order to improve the accuracy of the model and enrich the number of categories in the data set, this application merges data from two different data sources and obtains a total of 2083 valid images containing all categories.

[0053] (2) Label and preprocess image data and public data to obtain labeled datasets.

[0054] In the actual application process, after obtaining image data and public data in real scenes in open-pit mines, the images are annotated and made into a data set that meets the requirements of this solution. Bounding box annotation can be used for data annotation.

[0055] Bounding box annotation is currently the most commonly used annotation method in target detection. It uses a bounding box (usually a rectangular bounding box) to mark the position and size of the target in the image. Bounding box annotation is easier to implement and verify during the annotation process. It only requires determining the position and size of the box based on the target's circumscribed rectangular box. This annotation process is more efficient and faster. At the same time, there are a large number of targets in remote sensing datasets, such as houses and vehicles. Using bounding box annotation can collect and annotate a large amount of training data more quickly, thereby improving the speed and efficiency of training models.

[0056] Based on the above description, this embodiment uses the Roboflow online annotation tool to annotate the bounding boxes of the 2083 remote sensing data images obtained, and the annotation format is YOLOv8. During the annotation process, due to the large deformation of landslides, slag piles, and complex scenes, in order to enhance the robustness of the model, the bounding box only contains a unique target area, and the bounding boxes of the same target area are unique and do not overlap. Based on this, the bounding box annotation should follow the following rules:

[0057] 1) Select an appropriate bounding box for each object. The bounding box should tightly surround the target area while avoiding including other irrelevant areas.

[0058] 2) Each object has only one bounding box, and there should be no overlapping or repeated bounding boxes. If the landslide consists of multiple discontinuous parts, multiple bounding boxes can be used to annotate each part.

[0059] Further, the analysis of the labeled data set revealed that there were problems with different data sources, different image features, and a small number of data samples. Therefore, it is necessary to perform image preprocessing on the data set. Based on this, preprocessing includes operations such as resizing, denoising, contrast enhancement, color space conversion, image cropping and rotation, normalization, and data enhancement.

[0060] Mosaic enhancement, as a special data enhancement method, can be used to expand the training set to improve the generalization and robustness of the model. Mosaic enhancement can enrich the number of single image samples in the data set and increase the diversity of targets in the training data set. By combining multiple images into a mosaic image, the model can handle the complex situation of multiple targets and backgrounds at the same time, and improve the model's recognition and detection capabilities for multi-target scenes. Mosaic enhancement can also help the model learn the contextual information of the target. Since the mosaic image contains information from multiple images, the model can better understand the relationship between the target and its surrounding environment. This is particularly useful for target detection tasks, and is also very beneficial for establishing contextual information around the target for accurate positioning and classification. The training set of the original data set is expanded using Mosaic enhancement technology, and the total data volume is expanded from the original 2083 remote sensing image data to 4864 remote sensing image data, and a new data set is obtained after annotation.

[0061] (3) The initial model is constructed using YOLOv8s as the benchmark model. For example, the serialized ASPP structure and ASPP-SP structure are used to improve the spatial pyramid pooling structure of the benchmark model. The serialized ASPP structure is suitable for multi-scale target features in mining area object detection by serializing the dilation rate of the void convolution. The DC2f structure is adopted, and the deformable convolution is introduced to improve the cross-stage local network structure of the benchmark model to obtain the initial model. The ASPP-SP structure improves the global pooling operation by introducing the coordinate attention mechanism. The ASPP-SP structure uses an adaptive average pooling layer to perform pooling operations on the input feature map in the horizontal and vertical directions respectively. The features are divided into two branches in the horizontal and vertical directions. Each branch performs convolution, batch normalization and activation function processing on the pooled input feature map to generate attention weights. The input feature map is multiplied by the attention weights of the two branches to obtain the output result.

[0062] In the actual application process, the YOLO series models were compared and analyzed under the data set constructed in this application using the same environment and experimental parameters. YOLOv8s was finally selected as the benchmark model based on the experimental results. Secondly, the difficulty of ground object detection in open-pit mining scenarios was investigated, and the improvement plan in the above step (3) was designed to address the problem. Among them:

[0063] 1) Model improvement based on spatial pyramid pooling structure. Considering the problem of multi-scale target missed detection in open-pit mining scene. Spatial pyramid pooling (SPP) is a technology proposed under the idea of ​​FPN. By applying pooling operation in feature pyramid network, the speed of candidate box generation is improved and the computational cost is saved. SPP enables the network to accept input images of any size, which can overcome the limitation of fixed input size in known target detection methods. In YOLOv5, the SPPF (Spatial Pyramid Pooling Fast) structure is introduced as an improved version of SPP. The application of SPPF structure makes the network faster than the known SPP, and YOLOv8 also continues this structure. However, the maximum pooling operation in the SPPF structure will lose some feature information while increasing the receptive field of the model, and will cause the loss of some internal data structure and spatial level information. Unlike the convolution operation, the maximum pooling operation is deterministic and cannot be adjusted and optimized according to the characteristics of the data, such as the use of three fixed maximum pooling operations in YOLOv8. However, this deterministic maximum pooling operation will cause some problems: First, due to the irreversibility of the maximum pooling operation, once the maximum value is selected and used as the output, other feature information in the area will be lost. This will lead to the loss of some detail features, especially for small-sized objects or edge information. Secondly, as the pooling kernel increases, the increase in the receptive field will also lead to the loss of more feature information, which will have a negative impact on the positioning and boundary recognition of the object. Therefore, based on the benchmark model, this embodiment optimizes the structure of the model and proposes a serialized ASPP structure and an ASPP-SP structure. The serialized ASPP structure is suitable for multi-scale target features in open-pit mining area ground feature detection by serializing the dilation rate of the void convolution. Since the serialized ASPP structure still uses the pooling branch to obtain the global receptive field, the global pooling will perform a dimensionality reduction operation on the input image, resulting in the loss of some detail information. And the global pooling operation ignores the position information of the pixel, resulting in a reduced sensitivity of the model to the position of the target and difficulty in paying attention to the position information. In order to enhance the model's ability to focus on and distinguish target features, the ASPP-SP structure improves the global pooling operation by introducing the CoordinateAttention (CoordAtt) mechanism. The structure uses an adaptive average pooling layer to perform pooling operations on the input feature map in the horizontal and vertical directions respectively. This can capture the information of the feature map at different positions. Then, the pooled features are processed by convolution, batch normalization, and activation functions to further extract features and enhance representation capabilities. Next, the features are divided into two branches in the horizontal and vertical directions. Each branch uses an independent convolution layer and SiLU activation function to generate attention weights.Finally, the input feature map is multiplied by the attention weights in the horizontal and vertical directions, so that the model pays more attention to the positions with high attention weights, thereby improving the model's perception of the horizontal and vertical coordinate position information and obtaining the final output. Based on the above improved processing, the target positioning is more accurate, which can significantly improve the target detection accuracy.

[0064] 2) Model improvement based on cross-stage local network structure. For the detection of deformable targets in remote sensing objects in open-pit mining areas, although C2f provides rich gradient flow information, its detection effect is not good for the large number of deformable targets in the MAOD dataset. This is because the convolutional neural network (CNN) operation uses a fixed-shape receptive field, which makes it difficult to capture the local correlation of irregular input features. Secondly, CNN has greater limitations in position-sensitive tasks and cannot directly process information about position changes. At the same time, CNN uses the same convolution kernel at each position, which causes the number of parameters to increase as the size of the input feature map increases, increasing the complexity and computational cost of the model. In addition, CNN is spatially invariant when processing spatial transformations, and cannot directly capture the semantic information in spatial transformations. It has limited modeling capabilities for deformable targets, and the limitation of parameter sharing reduces the expressive power of the model. Therefore, this embodiment proposes a structure called DC2f (CSP Bottleneck With 2Deformable Convolutions) based on the idea of ​​cross-stage local networks and deformable convolutions. The DC2f structure enables the network to better capture gradient flow information while maintaining lightweight. In addition, deformable convolutions (DCN) and DCNv2 can be introduced to adapt to changes in target shape, thereby improving the model's ability to detect deformed targets.

[0065] (4) Using the labeled data set to train the initial model until the trained initial model meets the preset conditions, a trained initial model is obtained.

[0066] (5) Use the trained initial model as the ground object detection model.

[0067] For the improvement of the above benchmark model, the model is trained in the same dataset, server and experimental parameters, and tested in the same CPU server. The experimental results of the model improvement based on the spatial pyramid pooling structure and the model improvement based on the cross-stage local network structure are compared and analyzed one by one. Finally, the integrated results are analyzed.

[0068] The experimental results show that the detection accuracy mAP@0.5 of the fused and improved model in real scenes increased from 88.0% to 89.4%, an increase of 1.4%. mAP@0.5:0.95 increased from 65.8% to 69.2%, an increase of 3.4%. It meets the high-precision requirements of the industry for remote sensing object recognition and detection. Therefore, it is used as the final open-pit mine area object detection model. Among them, mAP@* refers to the average accuracy when IoU=*.

[0069] In summary, this application is based on the actual scene of the open-pit mine area. It obtains remote sensing images through platforms such as drones and satellites to establish a set of open-pit mine area remote sensing ground object recognition data sets, and combines the advantages of YOLOv8 to build a ground object target detection model, and then it can establish a remote sensing ground object integrated intelligent extraction and recognition platform to achieve the classification and positioning of typical ground objects (landslides, slag piles, farmland, buildings, vehicles) in the open-pit mine area. Provide technical support for sustainable mining and environmental governance in open-pit mines.

[0070] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 3 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. 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 mining area ground object target detection data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication 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 mining area ground object target detection method is implemented.

[0071] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application 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. In an exemplary 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.

[0072] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0073] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0074] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0075] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods 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 in the present application 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), magnetic 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. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0076] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0077] 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.

[0078] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A method for detecting ground objects in a mining area, characterized in that: The mining area ground object target detection method comprises: Acquire images of the mining area; Build a ground object detection model using YOLOv8s as the benchmark model; The mining area image is input into the ground object target detection model to obtain a ground object target detection result.

2. The method for detecting ground objects in mining areas according to claim 1, characterized in that: The ground object detection model is constructed using YOLOv8s as the benchmark model, including: Obtain image data and public data in real scenes of mining areas; the image data are remote sensing images collected by drones and Gaofen-2 satellites; the public data are obtained by screening historical data of Google Earth images; Annotating and preprocessing the image data and the public data to obtain an annotated data set; Build the initial model using YOLOv8s as the benchmark model; The initial model is trained using the labeled data set until the trained initial model meets a preset condition, thereby obtaining the trained initial model; The trained initial model is used as the ground object detection model.

3. The method for detecting ground objects in mining areas according to claim 2, characterized in that: Build an initial model using YOLOv8s as the baseline model, including: The spatial pyramid pooling structure of the benchmark model is improved by using the serialized ASPP structure and the ASPP-SP structure; The DC2f structure is adopted and deformable convolution is introduced to improve the cross-stage local network structure of the baseline model to obtain the initial model.

4. The method for detecting ground objects in mining areas according to claim 3, characterized in that: The serialized ASPP structure is suitable for multi-scale target features in mining area ground feature detection by serializing the dilation rate of the void convolution.

5. The method for detecting ground objects in mining areas according to claim 3, characterized in that: The ASPP-SP structure improves the global pooling operation by introducing a coordinate attention mechanism; the ASPP-SP structure uses an adaptive average pooling layer to perform pooling operations on the input feature map in the horizontal direction and the vertical direction respectively; The features are divided into two branches along the horizontal and vertical directions; each branch performs convolution, batch normalization and activation function processing on the pooled input feature map to generate attention weights; Multiply the input feature map with the attention weights of the two branches to get the output result.

6. The method for detecting ground objects in mining areas according to claim 2, characterized in that: The preprocessing includes one or more of resizing, denoising, contrast enhancement, color space conversion, image cropping and rotation, normalization, and data enhancement.

7. The method for detecting ground objects in mining areas according to claim 2, characterized in that: The image data and the public data are annotated with a bounding box; the bounding box only contains a unique target area, and the bounding boxes of the same target area are unique and do not overlap.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method for detecting ground objects in mining areas as described in any one of claims 1 to 6.

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 method for detecting ground objects in mining areas according to any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for detecting ground objects in mining areas according to any one of claims 1 to 6 is implemented.

Images