Lunar dark mantle sediment identification method and system based on YOLOv8 model

Abstract

The present invention discloses a method and system for identifying lunar dark mantle sediments based on the YOLOv8 model. The lunar dark mantle sediment identification model introduces a multi-scale feature extraction module in the process of outputting feature maps by the skeleton network module, so that the model uses a multi-scale convolutional neural network to extract image features, thereby enhancing the scale adaptability and nonlinearity of the model and improving the performance of the model in detecting multi-scale targets. The present invention combines dark mantle sediment data with a deep learning algorithm to enhance the ability to identify complex terrain and sediments of different scales. At the same time, the present invention has higher detection accuracy and speed, can quickly respond to needs under different observation conditions, and provide more reliable data support for scientific research. The detection of dark mantle sediments by the present invention can provide an important scientific basis for astronomical exploration, mineral resource assessment, and revealing the history of volcanic activity and tectonic evolution of the early moon.

Description

Technical Field

[0001] The present invention belongs to the field of deep space exploration technology, and specifically relates to a remote sensing image processing technology based on the Yolov8 model, which is used to identify dark mantle deposits (DMD) on the lunar surface. Background Art

[0002] Lunar dark mantle deposits (DMDs) are a key geological feature formed by volcanic eruptions on the lunar surface. DMD regions are typically rich in elements such as iron and titanium. Previous researchers have categorized these deposits into two groups: "regional" dark mantle deposits (RDMDs), characterized by thin, flat overburden deposits covering areas of several thousand square kilometers, and "localized" dark mantle deposits (LDMDs), which are smaller (typically less than 100 square kilometers) and have recognizable positive topographic features. These deposits are typically dark in color and composed primarily of glassy volcanic eruptions. They primarily formed during early lunar volcanism, when ejected material cooled and deposited on the lunar surface, covering portions of younger basaltic terrain. The study of DMDs has helped shed light on the early lunar volcanic history and tectonic evolution. Traditionally, the identification of DMDs has relied on manual analysis of high-resolution remote sensing imagery. Researchers typically visually identify dark areas in images and combine them with spectral and topographic data to confirm whether they are DMDs. However, this method is inefficient, time-consuming, and reliant on the researcher's expertise, making it susceptible to subjective factors. Furthermore, manual analysis, when faced with massive amounts of lunar data, struggles to ensure comprehensive and accurate identification. With the advancement of artificial intelligence and deep learning technologies, computer vision algorithms are increasingly being used in remote sensing image analysis. Convolutional neural networks (CNNs) have made significant progress in tasks such as object detection and image classification, significantly improving the efficiency of image processing.

[0003] While deep learning models excel in ground object detection, their application to DMD recognition in lunar remote sensing imagery still faces numerous challenges. First, the complex lighting conditions and reflectivity characteristics of the lunar surface, coupled with the presence of significant noise in the imagery, can affect the model's detection accuracy. Second, DMD distribution patterns are complex and easily confused with other lunar landforms, making traditional automated recognition algorithms ineffective in this task.

[0004] The YOLO (You Only Look Once) model has become a representative algorithm in the field of object detection in recent years. Its unique architecture enables real-time detection while maintaining high detection accuracy. YOLOv8 (You Only Look Once Version 8) is a deep learning-based object detection model that excels in real-time object recognition and localization tasks. This model utilizes a fully convolutional neural network (CNN) architecture for efficient feature extraction and fast inference. Trained on multiple datasets, YOLOv8 can accurately detect and classify a wide range of objects. However, applying this model to DMD recognition in lunar remote sensing imagery still faces numerous challenges. Summary of the Invention

[0005] To address the issues of efficiency and reliability in lunar dark mantle deposit (DMD) identification, the present invention provides a lunar dark mantle deposit identification method and system based on the YOLOv8 model. Deep learning technology is used to achieve efficient DMD identification, and the model prediction results are verified in combination with lunar-derived FeO and TiO2 percentage content data to improve the reliability and scientificity of the identification results.

[0006] According to one aspect of the present invention, a method for identifying lunar dark mantle sediments based on the Yolov8 model is provided, comprising:

[0007] Obtain slice images of lunar remote sensing image data;

[0008] Inputting the slice image into a trained lunar dark mantle deposit recognition model to output a predicted potential DMD area;

[0009] Use FeO and TiO2 data to verify potential DMD areas, identify potential DMD areas that meet preset requirements, and mark them in lunar remote sensing images;

[0010] The training of the lunar dark mantle sediment identification model includes:

[0011] Construct a lunar remote sensing image dataset with DMD regions marked;

[0012] Build a yolov8 model and embed a multi-scale feature extraction module into the model's skeleton network to accurately focus on areas with high DMD correlation;

[0013] Based on the constructed DMD dataset, the yolov8 model is trained using an adaptive optimizer and a dynamic batch size strategy, and the trained model is output as a lunar dark mantle sediment recognition model.

[0014] As a further technical solution, a lunar remote sensing image dataset with DMD region annotated was constructed, including:

[0015] Obtain remote sensing image data of the lunar surface and obtain DMD position coordinates in DMD-related papers;

[0016] The obtained DMD position coordinates are marked on the global map of the moon to establish a DMD dataset.

[0017] As a further technical solution, based on the constructed DMD dataset, the yolov8 model is trained using an adaptive optimizer and a dynamic batch size strategy, including:

[0018] The skeleton network is used to extract features from the input DMD training samples and output feature maps; the multi-scale feature extraction module is used to extract multi-scale features from the output feature maps; the feature fusion network is used to sample and fuse the extracted multi-scale features; and the head output module is used to output the prediction box.

[0019] As a further technical solution, the multi-scale feature extraction module includes two branches, one branch uses residual connection to speed up model training; the other branch uses channel attention mechanism, average pooling and spatially separable convolution to extract DMD feature information.

[0020] As a further technical solution, the slice image is input into a trained lunar dark mantle sediment identification model to output a predicted potential DMD area, further comprising:

[0021] Multiple detection boxes are generated based on feature maps of different scales, and the potential DMD area with the highest confidence is output based on the detection threshold.

[0022] As a further technical solution, FeO and TiO2 data were used to verify the potential DMD areas, including:

[0023] The lunar FeO and TiO2 data are compared with the predicted potential DMD areas to generate the geographical location of the DMD areas that meet the preset requirements, as well as the matching information between the DMD areas and the FeO and TiO2 percentage content data.

[0024] According to one aspect of the present invention, a lunar dark mantle sediment identification system based on the Yolov8 model is provided, comprising:

[0025] A data acquisition module is used to obtain slice images of lunar remote sensing image data;

[0026] a potential DMD region prediction module, configured to input the slice image into a trained lunar dark mantle sediment recognition model and output a predicted potential DMD region;

[0027] The DMD region verification and identification module is used to verify potential DMD regions using FeO and TiO2 data, identify potential DMD regions that meet preset requirements, and mark them in lunar remote sensing images;

[0028] The training of the lunar dark mantle sediment identification model includes:

[0029] Construct a lunar remote sensing image dataset with DMD regions marked;

[0030] Build a yolov8 model and embed a multi-scale feature extraction module into the model's skeleton network to accurately focus on areas with high DMD correlation;

[0031] Based on the constructed DMD dataset, the yolov8 model is trained using an adaptive optimizer and a dynamic batch size strategy, and the trained model is output as a lunar dark mantle sediment recognition model.

[0032] According to one aspect of the present invention, a device for identifying lunar dark mantle sediments based on the YOLOv8 model is provided, comprising a memory and a processor, wherein the memory stores program instructions executed by the processor, and the processor calls the program instructions to execute the steps of the method for identifying lunar dark mantle sediments based on the YOLOv8 model.

[0033] According to one aspect of the present invention, a non-transitory computer-readable storage medium is provided, wherein the non-transitory computer-readable storage medium stores computer instructions, and the computer instructions enable the computer to execute the steps of the lunar dark mantle sediment identification method based on the YOLOv8 model.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] 1. This paper utilizes the YOLOv8 model to achieve efficient DMD recognition: The YOLOv8 model maintains efficient target detection performance when processing large-scale remote sensing imagery. Its rapid feature extraction and detection capabilities are particularly well-suited for DMD recognition tasks. Even with the small dataset required for lunar dark mantle sediment identification, the introduction of a multi-scale feature extraction module into the YOLOv8 model enhances the model's ability to extract DMD detail features at different scales.

[0036] 2. This paper validates predictions using FeO and TiO2 data. Lunar-derived FeO and TiO2 percentage data are important chemical composition indicators in lunar geology. Since DMD regions are typically rich in elements such as iron and titanium, they can help distinguish them. In DMD identification, using derived FeO and TiO2 data to validate the model's predicted DMD regions effectively improves the model's reliability. This verification ensures that the model-identified DMD regions are consistent with actual geological characteristics. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, a brief introduction will be given below to the drawings used in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0038] Figure 1 A schematic flow chart of a method for identifying lunar dark mantle sediments based on the YOLOv8 model provided in an embodiment of the present invention.

[0039] Figure 2 A schematic diagram of the training process of the lunar dark mantle sediment identification model provided in an embodiment of the present invention.

[0040] Figure 3 A schematic diagram of the multi-scale feature extraction module flow provided in an embodiment of the present invention.

[0041] Figure 4 A schematic diagram of the process of predicting potential DMD areas of dark mantle deposits using the model provided in an embodiment of the present invention.

[0042] Figure 5 Schematic diagram of dark mantle deposits provided in an embodiment of the present invention. The picture on the left is a regional dark mantle deposit, and the picture on the right is a local dark mantle deposit.

[0043] Figure 6 A schematic diagram of part of the effects of the lunar dark mantle sediment identification process provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0044] The terms "including" and "having" and any variations thereof in the description and claims of the present invention and the above-mentioned drawings are intended to cover non-exclusive inclusions, for example, a process, method, system, product or apparatus that includes a series of steps or units is not necessarily limited to the steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products or apparatuses.

[0045] It should be noted that:

[0046] While the YOLOv8 model excels in ground object detection, its application to DMD recognition in lunar remote sensing imagery still faces numerous challenges. First, the complex lighting conditions and reflectivity characteristics of the lunar surface, coupled with the presence of significant noise in the imagery, can affect the model's detection accuracy. Second, DMD distribution patterns are complex and easily confused with other lunar landforms, making traditional automated recognition algorithms poorly suited for this task. Therefore, a new approach combining deep learning models with geological validation data is needed to improve recognition efficiency while ensuring the scientific validity of detection results.

[0047] To address these issues, the present paper proposes a DMD identification method based on the Yolov8 model and the Multi-Scale Feature Extraction (MSFE) module. This method verifies the results using derived FeO and TiO2 content data to ensure accurate and scientific identification. This method not only enables efficient DMD identification but also enhances the reliability of the results by comparing physical and chemical compositions.

[0048] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. In addition, the technical features in the various embodiments or single embodiments provided by the present invention are arbitrarily combined with each other to form a new technical solution. This combination is not restricted by the sequence of steps and / or structural composition mode, but must be based on the ability of ordinary technicians in this field to implement it. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that this combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0049] See also Figure 1 The lunar dark mantle sediment identification method based on the YOLOv8 model provided in an embodiment of the present invention includes: obtaining slice images of lunar remote sensing image data; inputting the slice images into a trained lunar dark mantle sediment identification model to output predicted potential DMD areas; verifying the potential DMD areas using FeO and TiO2 data, identifying potential DMD areas that meet preset requirements, and marking them in the lunar remote sensing image.

[0050] The method described in this embodiment of the present invention incorporates a multiscale feature extraction module based on the YOLOv8 model to identify lunar dark mantle deposits. This method aims to efficiently identify DMD using deep learning techniques. The model's predictions are validated using lunar-derived FeO and TiO2 percentage data, improving the reliability and scientific nature of the identification results. This method not only addresses the limitations of small datasets but also ensures the reliability of the identification results through comparison of chemical composition data.

[0051] See also Figures 2 to 4 The construction, training, and identification process of the lunar dark mantle sediment identification model described in the embodiment of the present invention includes the following steps:

[0052] Step 1: Download the data and obtain the DMD position coordinates in the DMD-related papers.

[0053] Acquire high-resolution remote sensing imagery of the lunar surface, including data from missions such as the Lunar Reconnaissance Orbiter (LRO) and Clementine UVVIS. Specifically, obtain the Lunar Reconnaissance Orbiter Camera (LROC) Wide Angle Camera (WAC) global morphology map, which will be used to construct the DMD dataset based on the acquired DMD coordinates; and the Clementine UVVIS-derived FeO abundance map and the Lunar Reconnaissance Orbiter Camera (LROC) Wide Angle Camera (WAC)-derived TiO2 abundance map for subsequent validation and ensuring the reliability of model predictions. Obtain DMD position coordinates from relevant DMD papers.

[0054] Step 2: Mark the obtained DMD position coordinates on the lunar global map and establish a DMD dataset.

[0055] In step 2, the lunar optical remote sensing image is cropped into training samples of appropriate size. The DMD dataset is manually annotated using the labeling tool (LabelImg). The core task of the annotation is to frame the DMD region with a rectangular box and add the "DMD" label to each box, ensuring that the annotation file format is correct and the annotation coordinates are accurate.

[0056] Specifically, the DMD location coordinates were marked on a downloaded Lunar Reconnaissance Orbiter Camera (LROC) Wide Angle Camera (WAC) global morphology map, and the DMD region was clipped using the ARCGIS raster clipping tool to generate the DMD dataset. It should be noted that the dataset should include accurate annotations of the DMD region to facilitate the model's learning of recognition features.

[0057] Step 3: Build a lunar dark mantle sediment identification network model and train it on the DMD dataset;

[0058] The yolov8 model was trained using the annotated DMD dataset. When the dataset was fed into the model, the model rescaled the images to adjust their resolution (typically 640x640 pixels) to meet the model's input specifications. This step helped improve the efficiency and effectiveness of model training. The MSFE module was then introduced to optimize model performance. During training, the model's loss function was monitored to ensure that the model maintained high recognition accuracy on the validation set.

[0059] The lunar dark mantle sediment recognition model in step 3 consists of five components: input, a backbone network (Backbone), a multi-scale feature extraction module (MSFE), a feature fusion network (Neck), and a detection head (Head). During model training, after the input image is processed, the backbone network first performs feature extraction. A powerful convolutional neural network, CSPNet (Cross Stage Partial Network), serves as the backbone network. The backbone network extracts multi-level features from the input image through a series of convolution operations. These include low-level features that capture basic information such as edges, texture, and color, and high-level features that capture complex geometric shapes and object structure. The feature maps output by the backbone network are then passed through the Multi-Scale Feature Extraction Module to extract features at multiple scales, ensuring the model can detect objects of various sizes. The Neck then performs feature fusion. The model combines a Feature Pyramid Network (FPN) and a Path Aggregation Network (PAN). By fusing multi-level feature maps, the model enhances its ability to detect objects of varying scales. This enables the model to simultaneously detect both small, detailed objects and larger, global structures within the image. Finally, the output of the head is used to predict the object. The model's detection head processes each feature map in the feature pyramid and outputs a prediction result. The detection head's main task is to regress the bounding box position of each candidate box and predict the category and confidence score of the object within each candidate box. After the candidate boxes are predicted, the model uses NMS to remove highly overlapping candidate boxes, retaining only the bounding boxes with the highest confidence.

[0060] This application innovatively introduces the Multi-Scale Feature Extraction (MSFE) module into the YOLOv8 model. Although similar implementations have been previously studied in the YOLOv5 model, the two models face different technical challenges due to their differing architectural characteristics. YOLOv5 focuses on fast detection of basic objects, and the introduction of MSFE emphasizes lightweight adaptation. YOLOv8, on the other hand, targets complex, high-resolution imagery (suitable for remote sensing imaging), and its integration presents several challenges. First, the interface and dimensionality rules between the Multi-Scale Feature Extraction module and the native network layers require fine-tuning to prevent feature mismatches and information flow obstruction. Second, hyperparameter settings require re-tuning to account for YOLOv8's computational and memory requirements. For example, the adaptive kernel parameters of the ECA component must be optimized based on the new channel dimensionality to ensure efficient utilization of computing resources. Third, post-training prediction validation is performed on FeO and TiO2 data, further improving the detection accuracy of the YOLOv8 model for remote sensing imaging tasks. By combining these data to verify the prediction results, the carefully designed MSFE module effectively improved the model's performance in high-precision detection, achieving a detection accuracy improvement that is significantly better than the basic framework.

[0061] In this paper, we introduce the Multi-Scale Feature Extraction (MSFE) module into the YOLOv8 model. We analyze in detail the differences between its integration approach and that of the MSFE module in the YOLOv5 model and make corresponding adjustments to ensure efficient adaptation to YOLOv8's complex network architecture. Although both YOLOv5 and YOLOv8 utilize the MSFE module for feature extraction, differences in their network architectures lead to different integration methods and optimization strategies. YOLOv5 is primarily designed for rapid object detection and has a relatively simple architecture. Therefore, the MSFE module is embedded in the middle layer of the network as a feature fusion module to extract and fuse multi-scale features. Its focus is on rapidly extracting and fusing features at different scales to meet lightweight task requirements. YOLOv8, on the other hand, is a more complex model with a deeper and more detailed network structure, making it suitable for processing high-resolution and complex remote sensing imagery data. In YOLOv8, the MSFE module is embedded deep within the backbone network, enhancing the model's ability to extract and recognize fine-grained features. The introduction of the MSFE module requires that its output feature maps correctly interface with other layers in the YOLOv8 network. Therefore, during integration, its input and output dimensions must be precisely adjusted to ensure they match those of the YOLOv8 network layers. Specifically, the adaptation layer must be adjusted to ensure that the output dimensions of the MSFE module match the input dimensions of subsequent network layers. By precisely controlling parameters such as the number of channels, width, and height of the feature map, we can effectively avoid feature mismatches and information flow bottlenecks, ensuring smooth information flow between layers in the network.

[0062] The Multi-Scale Feature Extraction (MSFE) module flexibly configures submodules based on task characteristics and model architecture, aiming to accurately capture multi-scale image features. It features a dual-branch structure: the first branch utilizes residual connections, a classic architectural optimization technique in deep learning. This effectively mitigates vanishing gradients through skip connections. Adaptive tuning allows for strict data flow control, deeply aligning with the overall architecture and accelerating model training. The second branch integrates channel attention (ECA), average pooling, and spatial convolutional layers, working together to efficiently extract DMD feature information. This complements the first branch and fully empowers MSFE's exceptional performance.

[0063] The multi-scale feature extraction module consists of a channel attention mechanism (ECA), average pooling, and spatially separable convolution. The ECA module adaptively adjusts the convolution kernel size based on the feature vector calculated by global average pooling (GAP). This process needs to be adjusted in YOLOv8 based on the new channel dimension to ensure that the convolution kernel can achieve the maximum effect on larger or smaller channel dimensions. It is necessary to ensure that the kernel size matches the complexity of the network to avoid overly large or small convolution kernels affecting the feature extraction ability. The channel attention mechanism (ECA) focuses on improving feature selection capabilities without increasing excessive computational burden. ECA uses a global average pooling method without dimensionality reduction to aggregate features, adaptively determines the kernel size k, and then performs a one-dimensional 1D convolution of the kernel size k on the feature map. The channel attention vector is then obtained through the Sigmoid function. The ECA calculation method is as follows:

[0064] (1)

[0065] (2)

[0066] Where σ is the sigmoid function, Conv1D is a 1-D convolution with a kernel size of k, which is adaptively determined based on the channel dimension C. GAP(X) represents the global average pooling operation, which converts the result of global average pooling of each channel of the input feature map into a channel-level statistic vector. W represents the channel attention vector. X and Y represent the input feature map and output feature map, respectively.

[0067] (3)

[0068] where |x| odd represents the nearest odd number to x, γ and b are hyperparameters, represents a function mapping relationship that generates a specific output value according to the input channel dimension C. We assign the values ​​of formulas (2) and (1) to γ ​​and b.

[0069] Average pooling performs downsampling by averaging the pixels within each corresponding region of the input feature map according to a given kernel size. Because it has no parameters to optimize, it does not over-learn the details of the training data, thus avoiding overfitting. Average pooling is calculated as follows:

[0070] (4)

[0071] Where s represents the kernel size, Indicates the points in the calculation area The pixel value of .

[0072] Spatial separable convolution: Spatial separable convolution splits a standard convolution operation into multiple small kernel convolution operations in the spatial dimension.

[0073] (5)

[0074] (6)

[0075] in Indicates the midpoint of k×k convolution The value of Represents the value of the midpoint (i, 1) of the k-times × 1 convolution, Represents the value of the midpoint (1, j) in a 1×k convolution. The number of parameters for a k×1 convolution and a 1×k convolution is (k+1)+(1+k)=2k, which is smaller than the number of parameters for a k×k convolution when k>2. Spatially separable convolution reduces module parameters, speeds up model computation, and increases model depth.

[0076] When integrating the Multi-Scale Feature Extraction (MSFE) module into the YOLOv8 model, due to its complex network structure, hyperparameters during training require careful tuning, particularly the learning rate and batch size. First, optimizing the learning rate is crucial. A high learning rate can lead to exploding gradients, while a low learning rate can slow down training. To address this, a learning rate scheduling strategy, such as step decay, can be employed to gradually reduce the learning rate and help the model converge stably. Furthermore, the adaptive optimizer (Ranger) can automatically adjust the learning rate based on the gradient, improving training stability and addressing the computational complexity introduced by the MSFE module. Secondly, the batch size also needs to be optimized. A larger batch size improves computational efficiency, reduces gradient estimation fluctuations, and enhances training stability. However, this also increases graphics memory requirements, so a balance between memory and computing power is crucial. A dynamic batch size strategy can be employed, initially training with a smaller batch size and then gradually increasing it as training progresses, accelerating training. By fine-tuning these hyperparameters, the MSFE module can achieve stable training in YOLOv8 and improve model performance.

[0077] Step 4: Use the model trained in step 3 to predict and verify DMD for the lunar global slice image.

[0078] First, the Lunar Reconnaissance Orbiter Camera (LROC) Wide Angle Camera (WAC) global morphology map is sliced ​​and then loaded into the model using the saved model weights to maintain the trained network structure and parameters. The sliced ​​images are fed into the model, and the forward propagation network portion of the model generates predictions. After the predictions are generated, several post-processing steps are required, including non-maximum suppression (NMS) to filter out excessively overlapping detection boxes, thereby obtaining potential DMD detection results. Finally, the predicted potential DMD regions are identified and detected. The model output is presented as detection boxes, marking potential DMD regions. In this paper, for the identification of lunar dark mantle deposits (DMD), the FeO (wt.%) and TiO2 (wt.%) content ranges are specified as FeO: 10-17.5 wt.%, and TiO2: 2-4.5 wt.%. This range was selected based on an analysis of DMD characteristics in different lunar regions and relevant literature. In this study, the FeO and TiO2 content ranges are primarily used to screen for possible DMD areas, and combined with other features of remote sensing imagery for comprehensive assessment, this improves the model's accuracy in identifying DMD. Furthermore, FeO and TiO2 data are used to validate the model's predictions. By comparing the FeO and TiO2 content within the predicted area with the geological characteristics of the background area, the consistency of the test results with the actual geological characteristics is ensured.

[0079] Step 5: Find the latitude and longitude coordinates of the predicted DMD obtained in step 4, and mark the potential DMD that meets the requirements on the global remote sensing image of the moon. Figure 5 A schematic diagram of dark mantle deposits is given, where the left picture shows regional dark mantle deposits and the right picture shows local dark mantle deposits. Figure 6 A schematic diagram of some of the effects of the lunar dark mantle sediment identification process is given.

[0080] The implementation of each embodiment of the present invention is based on programmed processing performed by a device with processor functionality. Therefore, in practical engineering, the technical solutions and functions of each embodiment of the present invention are packaged into various modules. Based on this reality, and in addition to the above-mentioned embodiments, an embodiment of the present invention provides a lunar dark mantle sediment identification system based on the YOLOv8 model. This system is used to implement the lunar dark mantle sediment identification method based on the YOLOv8 model described in the above-mentioned method embodiment.

[0081] The system includes: a data acquisition module for acquiring slice images of lunar remote sensing image data; a potential DMD area prediction module for inputting the slice images into a trained lunar dark mantle sediment recognition model and outputting predicted potential DMD areas; a DMD area verification and recognition module for verifying potential DMD areas using FeO and TiO2 data, identifying potential DMD areas that meet preset requirements and marking them in lunar remote sensing images; wherein, the training of the lunar dark mantle sediment recognition model includes: constructing a lunar remote sensing image dataset with DMD areas marked; constructing a YOLOv8 model and embedding a multi-scale feature extraction module in the model's skeleton network for accurately focusing on areas with high DMD correlation; based on the constructed DMD dataset, the YOLOv8 model is trained using an adaptive optimizer and a dynamic batch size strategy, and the trained model is output as a lunar dark mantle sediment recognition model.

[0082] The lunar dark mantle sediment identification system based on the yolov8 model provided by the embodiment of the present invention is aimed at the current situation that "the existing technology mainly relies on manual analysis of remote sensing images, which is inefficient and easily affected by subjective factors, and has difficulty in processing large amounts of image data; deep learning technology is prone to overfitting when training deep learning models on small data sets, and the model performs poorly when processing complex geological features." It adopts the above-mentioned several modules and uses deep learning technology to achieve efficient identification of DMD, and verifies the model prediction results in combination with lunar-derived FeO and TiO2 percentage content data, thereby improving the reliability and scientific nature of the identification results and solving the problems of efficiency and reliability in the identification of lunar dark mantle sediments.

[0083] It should be noted that the system embodiments provided by the present invention are not only used to implement the methods in the above-mentioned method embodiments, but also used to implement the methods in other method embodiments provided by the present invention. The only difference lies in the setting of corresponding functional modules, and the principles thereof are basically the same as the principles of the above-mentioned system embodiments provided by the present invention. As long as those skilled in the art refer to the specific technical solutions in other method embodiments on the basis of the above-mentioned system embodiments, obtain corresponding technical means and technical solutions composed of these technical means by combining technical features, and on the premise of ensuring the practicality of the technical solutions, improve the modules in the above-mentioned system embodiments to obtain corresponding system class embodiments for implementing the methods in other method class embodiments.

[0084] Based on the same inventive concept as the above embodiment, an embodiment of the present invention also provides a lunar dark mantle sediment identification device based on the YOLOv8 model, including a memory and a processor, the memory storing program instructions executed by the processor, and the processor calling the program instructions to execute the steps of the lunar dark mantle sediment identification method based on the YOLOv8 model.

[0085] The lunar dark mantle deposit identification device based on the YOLOv8 model in the above embodiment uses a multi-scale feature extraction module based on the YOLOv8 model for DMD identification. The detection results are verified by combining derived FeO and TiO2 percentage data to ensure accurate and scientific identification. This device not only enables efficient DMD identification but also enhances the reliability of the identification results through comparison of physical and chemical composition.

[0086] Based on the same inventive concept as the above embodiment, an embodiment of the present invention also provides a non-transitory computer-readable storage medium, which stores computer instructions, and the computer instructions enable the computer to execute the steps of the lunar dark mantle sediment identification method based on the YOLOv8 model.

[0087] In summary, the present invention discloses a method and system for identifying lunar dark mantle sediments based on multi-source remote sensing data and the YOLOv8 model. When training the lunar dark mantle sediment identification model based on the YOLOv8 model, a multi-scale feature extraction module is introduced in the process of outputting the feature map of the skeleton network module, so that the model uses a multi-scale convolutional neural network to extract the features of the image, effectively enhancing the scale adaptability and nonlinearity of the model and improving the performance of the model in detecting multi-scale targets. Compared with the prior art, the innovation of the present invention is to combine dark mantle sediment data with a deep learning algorithm to improve the recognition ability of complex terrain and sediments of different scales. At the same time, the method has higher detection accuracy and speed, can quickly respond to the needs under different observation conditions, and provide more reliable data support for scientific research. After the detection of dark mantle sediments, it can provide important scientific basis for astronomical exploration, mineral resource assessment, and revealing the history of volcanic activity and tectonic evolution of the early moon.

[0088] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the technical solutions of the embodiments of the present invention.

Claims

1. A lunar dark mantle sediment identification method based on the yolov8 model is characterized by: include: Obtain slice images of lunar remote sensing image data; Inputting the slice image into a trained lunar dark mantle deposit recognition model to output a predicted potential DMD area; Using FeO and TiO2 data to verify potential DMD areas, identify potential DMD areas that meet preset requirements, and mark them in lunar remote sensing images, including: using lunar FeO and TiO2 data to compare and analyze predicted potential DMD areas, generating geographic locations of DMD areas that meet preset requirements, and matching information between the DMD areas and FeO and TiO2 percentage content data; The training of the lunar dark mantle sediment identification model includes: Construct a lunar remote sensing image dataset with DMD regions marked; Build a yolov8 model and embed a multi-scale feature extraction module into the model's skeleton network to accurately focus on areas with high DMD correlation; Based on the constructed DMD dataset, the yolov8 model is trained using an adaptive optimizer and a dynamic batch size strategy, and the trained model is output as a lunar dark mantle sediment recognition model.

2. The lunar dark mantle sediment identification method based on the yolov8 model according to claim 1, characterized in that: Construct a lunar remote sensing image dataset with DMD region annotated, including: Acquire remote sensing image data of the lunar surface and obtain DMD position coordinates; The obtained DMD position coordinates are marked on the global map of the moon to establish a DMD dataset.

3. The lunar dark mantle sediment identification method based on the yolov8 model according to claim 1 is characterized in that: Based on the constructed DMD dataset, the yolov8 model is trained using an adaptive optimizer and a dynamic batch size strategy, including: The skeleton network is used to extract features from the input DMD training samples and output feature maps; the multi-scale feature extraction module is used to extract multi-scale features from the output feature maps; the feature fusion network is used to sample and fuse the extracted multi-scale features; and the head output module is used to output the prediction box.

4. The lunar dark mantle sediment identification method based on the YOLOv8 model according to claim 3 is characterized in that: The multi-scale feature extraction module includes two branches, one branch uses residual connection to speed up model training; the other branch uses channel attention mechanism, average pooling and spatially separable convolution to extract DMD feature information.

5. The lunar dark mantle sediment identification method based on the yolov8 model according to claim 1 is characterized in that: The slice image is input into the trained lunar dark mantle deposit identification model to output the predicted potential DMD area, further comprising: Multiple detection boxes are generated based on feature maps of different scales, and the potential DMD area with the highest confidence is output based on the detection threshold.

6. The lunar dark mantle sediment identification system based on the yolov8 model is characterized by: include: A data acquisition module is used to obtain slice images of lunar remote sensing image data; a potential DMD region prediction module, configured to input the slice image into a trained lunar dark mantle sediment recognition model and output a predicted potential DMD region; A DMD region verification and identification module is used to verify potential DMD regions using FeO and TiO2 data, identify potential DMD regions that meet preset requirements, and mark them in lunar remote sensing images. This module includes: comparing and analyzing lunar FeO and TiO2 data with predicted potential DMD regions, generating geographic locations of DMD regions that meet preset requirements, and matching information between the DMD regions and FeO and TiO2 percentage content data; The training of the lunar dark mantle sediment identification model includes: Construct a lunar remote sensing image dataset with DMD regions marked; Build a yolov8 model and embed a multi-scale feature extraction module into the model's skeleton network to accurately focus on areas with high DMD correlation; Based on the constructed DMD dataset, the yolov8 model is trained using an adaptive optimizer and a dynamic batch size strategy, and the trained model is output as a lunar dark mantle sediment recognition model.

7. A lunar dark mantle sediment identification device based on the yolov8 model, characterized in that: It includes a memory and a processor, the memory stores program instructions executed by the processor, and the processor calls the program instructions to execute the steps of the lunar dark mantle sediment identification method based on the YOLOv8 model as described in any one of claims 1 to 5.

8. A non-transitory computer-readable storage medium, characterized in that The non-transitory computer-readable storage medium stores computer instructions, which enable the computer to execute the steps of the lunar dark mantle sediment identification method based on the YOLOv8 model as described in any one of claims 1 to 5.

Images