A ship detection method, device and electronic equipment based on deep learning

Abstract

The application provides a ship detection method, device and electronic equipment based on deep learning. The ship detection method comprises the following steps: acquiring a remote sensing image of a target ship and multi-channel magnetic field data caused by the target ship; performing target detection on the remote sensing image by using an improved YOLOv11n model to obtain a remote sensing detection result; performing anomaly detection on the multi-channel magnetic field data by using a trained convolutional neural network model to obtain a magnetic anomaly detection result, the magnetic anomaly detection result being used to indicate whether the ship target exists; taking the remote sensing detection result and the magnetic anomaly detection result as two evidence sources, respectively constructing a basic trust distribution function, and fusing the basic trust distribution functions of the two evidence sources by using a DS evidence theory synthesis algorithm to obtain a final detection result. The technical scheme of the application has the comprehensive advantages of high recognition accuracy, fast detection speed, strong robustness and wide adaptability.

Description

Technical Field

[0001] This invention relates to the field of computer graphics technology, specifically to a ship detection method, apparatus, and electronic device based on deep learning. Background Technology

[0002] With the increasing demands for safeguarding maritime rights and ensuring maritime security, ship target detection has significant application value in fields such as remote sensing monitoring, maritime search and rescue, and military reconnaissance. Traditional ship detection methods mainly rely on Synthetic Aperture Radar (SAR) or optical remote sensing images, using manually designed features combined with classifiers to achieve target identification. However, these methods exhibit poor robustness under complex sea conditions, varying lighting, and cloud cover, and struggle to effectively detect small targets, limiting their application in real-world scenarios.

[0003] In recent years, the rapid development of deep learning technology has brought new breakthroughs to ship detection. Target detection algorithms, represented by the YOLO series, have been widely used in remote sensing image target recognition tasks due to their high detection speed and accuracy. However, the YOLO series models still have problems in the feature extraction process, such as insufficient perception of small targets and inadequate modeling of channel information, leading to a significant drop in detection performance under low signal-to-noise ratio conditions. In addition, single-modal detection methods often struggle to guarantee stable detection results in complex marine environments.

[0004] On the other hand, magnetic anomaly detection, as a passive detection method, can effectively capture magnetic field disturbances caused by the movement of ship targets, and has advantages such as strong concealment and immunity to weather conditions. However, existing magnetic anomaly detection methods mostly rely on manual feature extraction and threshold judgment, which makes it difficult to adapt to dynamically changing ocean magnetic background noise, and the detection accuracy and generalization ability need to be improved.

[0005] Multimodal information fusion technology is considered an effective way to improve detection robustness. Dempster-Shafer (DS) evidence theory, as a commonly used decision-level fusion method, can handle uncertain information. However, traditional DS synthesis algorithms are prone to counterintuitive results when there is a high degree of conflict in the evidence, which limits their application in complex environments. Summary of the Invention

[0006] In view of this, this application proposes a ship detection method, device, and electronic equipment based on deep learning. Specifically, this application is implemented through the following technical solution: According to a first aspect of the embodiments of this specification, a ship detection method based on deep learning is provided, the method comprising the following steps: Step S1: Acquire remote sensing images of the target ship and multi-channel magnetic field data caused by the target ship; Step S2: Target detection is performed on the remote sensing image using the improved YOLOv11n model to obtain remote sensing detection results; In the improved YOLOv11n model, the C3k sub-blocks of each C3k2 block in the backbone network are replaced with multiple C3k-MSCB blocks connected in series. Each C3k-MSCB block contains a main path and parallel branches. The main path contains a convolutional layer and two MSCB blocks connected in series after the convolutional layer. Each parallel branch contains a convolutional layer. The output of the main path and the output of the parallel branch are concatenated and then input into a convolutional layer for feature fusion and channel adjustment. In the improved YOLOv11n model, the Bottleneck sub-block in the last C3k2 block of the neck network is replaced with the MSCB block; Furthermore, the PSA sub-block of the C2PSA block of the improved YOLOv11n model includes a main branch and two residual branches. The main branch includes a SENetV2 block and two convolutional layers connected in series after the SENetV2 block. The output of the SENetV2 block is concatenated with the input of the PSA sub-block through the first residual branch. The concatenated feature is processed by the two convolutional layers and then concatenated with the concatenated feature again through the second residual branch before being output. Step S3: Use the trained convolutional neural network model to perform anomaly detection on the multi-channel magnetic field data to obtain magnetic anomaly detection results. The magnetic anomaly detection results are used to indicate whether the ship target exists. Step S4: The remote sensing detection result and the magnetic anomaly detection result are used as two evidence sources. Basic trust assignment functions are constructed for each source. The DS evidence theory synthesis algorithm is used to fuse the basic trust assignment functions of the two evidence sources to obtain the final detection result. The DS evidence theory synthesis algorithm determines whether there is a conflict between the evidence sources by calculating an improved conflict coefficient. When a conflict is determined to exist, the credibility weight of each evidence source is calculated. The basic trust allocation function of the corresponding evidence source is weighted and averaged using the credibility weight. Then, the improved conflict coefficient is used to synthesize the processed basic trust allocation function. The expression for the improved conflict coefficient is: ; in, For the improved conflict coefficient; The original conflict coefficient, This is the basic trust assignment function when the remote sensing detection results are used as a source of evidence. This is the basic trust assignment function when the magnetic anomaly detection result is used as a source of evidence. The propositions supported by the basic trust assignment function corresponding to the remote sensing detection results. The propositions supported by the basic trust assignment function corresponding to the magnetic anomaly detection results, and , This indicates that the proposition of the objective exists. The proposition indicating that the objective does not exist; This is the preset balance coefficient.

[0007] According to a second aspect of the embodiments of this specification, a ship detection device based on deep learning is provided, the device comprising: The data acquisition unit is used to acquire remote sensing images of the target ship and multi-channel magnetic field data caused by the target ship. An image detection unit is used to perform target detection on the remote sensing image using an improved YOLOv11n model to obtain remote sensing detection results; In the improved YOLOv11n model, the C3k sub-blocks of each C3k2 block in the backbone network are replaced with multiple C3k-MSCB blocks connected in series. Each C3k-MSCB block contains a main path and parallel branches. The main path contains a convolutional layer and two MSCB blocks connected in series after the convolutional layer. Each parallel branch contains a convolutional layer. The output of the main path and the output of the parallel branch are concatenated and then input into a convolutional layer for feature fusion and channel adjustment. In the improved YOLOv11n model, the Bottleneck sub-block in the last C3k2 block of the neck network is replaced with the MSCB block; Furthermore, the PSA sub-block of the C2PSA block of the improved YOLOv11n model includes a main branch and two residual branches. The main branch includes a SENetV2 block and two convolutional layers connected in series after the SENetV2 block. The output of the SENetV2 block is concatenated with the input of the PSA sub-block through the first residual branch. The concatenated feature is processed by the two convolutional layers and then concatenated with the concatenated feature again through the second residual branch before being output. The magnetic anomaly detection unit is used to perform anomaly detection on the multi-channel magnetic field data using a trained convolutional neural network model, and obtain a magnetic anomaly detection result, which is used to indicate whether the ship target exists. The fusion decision unit is used to take the remote sensing detection results and the magnetic anomaly detection results as two evidence sources, construct basic trust assignment functions for each, and use the DS evidence theory synthesis algorithm to fuse the basic trust assignment functions of the two evidence sources to obtain the final detection result. The DS evidence theory synthesis algorithm determines whether there is a conflict between the evidence sources by calculating an improved conflict coefficient. When a conflict is determined to exist, the credibility weight of each evidence source is calculated. The basic trust allocation function of the corresponding evidence source is weighted and averaged using the credibility weight. Then, the improved conflict coefficient is used to synthesize the processed basic trust allocation function. The expression for the improved conflict coefficient is: ; in, For the improved conflict coefficient; The original conflict coefficient, Here is the basic trust assignment function when the target detection result is used as the source of evidence. This is the basic trust assignment function when the magnetic anomaly detection result is used as a source of evidence. The propositions supported by the basic trust assignment function corresponding to the remote sensing detection results. The propositions supported by the basic trust assignment function corresponding to the magnetic anomaly detection results, and , This indicates that the proposition of the objective exists. The proposition indicating that the objective does not exist; This is the preset balance coefficient.

[0008] According to a third aspect of the embodiments of this specification, an electronic device is provided, comprising: a processor; and a computer-readable storage medium storing computer program instructions that, when executed by the processor, cause the processor to perform the method as described in the first aspect.

[0009] The embodiments of this application have at least the following technical effects: This application embodiment achieves multimodal collaborative detection by fusing an improved YOLOv11n model with a convolutional neural network model and employing an improved DS evidence theory synthesis algorithm in the decision-making stage, which significantly improves the recognition accuracy and robustness of ship targets in complex environments. Specifically, the C3k-MSCB block was designed and the SENetV2 block was introduced into the YOLOv11n model to balance lightweight and high accuracy. The improved DS evidence theory synthesis algorithm solves the fusion failure problem of traditional methods when evidence conflicts by introducing optimized conflict coefficients and weighted preprocessing based on information entropy. This reduces the uncertainty of multi-source information fusion and gives the system comprehensive advantages such as high recognition accuracy, fast detection speed, strong robustness and wide adaptability. Attached Figure Description

[0010] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Some specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings in an exemplary and non-limiting manner. The same reference numerals in the drawings indicate the same or similar parts or components. Those skilled in the art should understand that these drawings are not necessarily drawn to scale. In the drawings: Figure 1 This is a schematic flowchart illustrating an exemplary embodiment of a deep learning-based ship detection method. Figure 2 This is a schematic diagram of the network structure of an MSCB block shown in an exemplary embodiment of this application; Figure 3 This is a schematic diagram of a network structure of a C3k-MSCB block shown in an exemplary embodiment of this application; Figure 4 This is a schematic diagram of the network structure of a SENetV2 block shown in an exemplary embodiment of this application; Figure 5 This is a schematic diagram of a network structure of a C2PSA block shown in an exemplary embodiment of this application; Figure 6 This is a schematic diagram of the network structure of a convolutional neural network model shown in an exemplary embodiment of this application; Figure 7 This is a block diagram illustrating an electronic device according to an exemplary embodiment of this application; Figure 8 This is a block diagram illustrating a deep learning-based ship detection device in an exemplary embodiment of this application. Detailed Implementation

[0011] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0012] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used in this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0013] To address the shortcomings of existing ship detection technologies, such as poor robustness under complex sea conditions, varying lighting, and cloud cover, and the difficulty in effectively detecting small targets, this application proposes a system architecture based on multimodal parallel processing and decision-level fusion. The system input includes remote sensing images of the ship and multi-channel magnetic field data collected by magnetic anomaly detection sensors; the output is the fused ship target detection result. First, the remote sensing images are processed using an improved YOLOv11n model to extract ship target identification results, while the magnetic field data is processed using a convolutional neural network model to output the magnetic anomaly detection probability. Subsequently, the two types of detection results are fused at the decision level using an improved DS evidence theory synthesis algorithm. After conflict resolution and trust allocation, the final detection decision is generated.

[0014] The following detailed explanation of each step is provided in conjunction with the accompanying drawings.

[0015] Figure 1 This is a schematic flowchart illustrating an exemplary embodiment of a deep learning-based ship detection method, as shown in this application. Figure 1 As shown, the ship detection method includes the following steps: Step S1: Acquire remote sensing images of the target ship and multi-channel magnetic field data caused by the target ship.

[0016] Step S2: Target detection is performed on the remote sensing image using the improved YOLOv11n model to obtain remote sensing detection results; In the improved YOLOv11n model, the C3k sub-blocks of each C3k2 block in the backbone network are replaced with multiple C3k-MSCB blocks connected in series. Each C3k-MSCB block contains a main path and parallel branches. The main path contains a convolutional layer and two MSCB blocks connected in series after the convolutional layer. Each parallel branch contains a convolutional layer. The output of the main path and the output of the parallel branch are concatenated and then input into a convolutional layer for feature fusion and channel adjustment. In the improved YOLOv11n model, the Bottleneck sub-block in the last C3k2 block of the neck network is replaced with the MSCB block; Furthermore, the PSA sub-block of the C2PSA block in the improved YOLOv11n model includes a main branch and two residual branches. The main branch includes a SENetV2 block and two convolutional layers connected in series after the SENetV2 block. The output of the SENetV2 block is concatenated with the input of the PSA sub-block through the first residual branch. The concatenated feature is processed by the two convolutional layers and then concatenated with the concatenated feature again through the second residual branch before being output.

[0017] Step S3: The trained convolutional neural network model is used to perform anomaly detection on the multi-channel magnetic field data to obtain magnetic anomaly detection results. The magnetic anomaly detection results are used to indicate whether the ship target exists.

[0018] Step S4: Using the remote sensing detection result and the magnetic anomaly detection result as two evidence sources, construct basic trust assignment functions for each, and use the DS evidence theory synthesis algorithm to fuse the basic trust assignment functions of the two evidence sources to obtain the final detection result.

[0019] The DS evidence theory synthesis algorithm determines whether there is a conflict between the evidence sources by calculating an improved conflict coefficient. When a conflict is determined to exist, the credibility weight of each evidence source is calculated. The basic trust allocation function of the corresponding evidence source is then weighted and averaged using the credibility weight. Finally, the improved conflict coefficient is used to synthesize the processed basic trust allocation function.

[0020] Figure 1 The ship detection method shown replaces the original C3k sub-blocks in each C3k2 block of the backbone network with multiple interconnected C3k-MSCB blocks. Each C3k-MSCB block connects two MSCB blocks along the main path. The MSCB blocks extract multi-scale features through parallel deep separable convolutions with different kernel sizes, effectively enhancing the network's ability to perceive multi-scale ship targets, especially small targets. Residual connections maintain gradient flow stability while extracting rich semantic information. In the neck network, the Bottleneck sub-block in the last C3k2 module is replaced with an MSCB block, ensuring a certain level of multi-scale feature response while effectively controlling the overall parameter and computational load of the model, achieving a balance between detection accuracy and model lightweighting. Furthermore, the SENetV2 attention mechanism is integrated into the PSA sub-block of the C2PSA block. This mechanism enhances the nonlinear interaction between channels through multi-branch fully connected layers, accurately focusing on key regions of ship targets. Simultaneously, the PSA sub-block's structure of a main branch and two residual branches effectively preserves more detailed information.

[0021] Furthermore, a trained convolutional neural network model is used to detect magnetic anomalies in multi-channel magnetic field data. This detection method passively detects magnetic field disturbances caused by the movement of ship targets, unaffected by environmental factors such as lighting conditions and cloud cover, offering advantages such as strong concealment and all-weather operation. The multi-channel magnetic field data includes three components of the magnetic field and their composite total field. After filtering, denoising, and feature extraction, the data is input into the convolutional neural network, effectively capturing weak magnetic anomaly signals under low signal-to-noise ratio conditions. This detection branch complements the remote sensing image detection branch. When the detection performance of remote sensing images deteriorates due to severe weather or sea conditions, magnetic anomaly detection can still provide a stable probability of target presence, offering a reliable source of evidence for decision-level fusion.

[0022] Furthermore, this embodiment employs an improved DS evidence theory synthesis algorithm to perform decision-level fusion of remote sensing identification results and magnetic anomaly detection results. This DS evidence theory synthesis algorithm treats the two detection results as independent evidence sources, constructs basic trust assignment functions for each, and calculates an improved conflict coefficient to determine whether conflicts exist between the evidence. The improved conflict coefficient introduces a trust assignment difference term between evidence sources on top of the traditional conflict coefficient, enabling a more comprehensive measurement of local consistency and global conflict distribution among evidence sources, overcoming the limitation of the traditional conflict coefficient which only considers focal element intersections. When a conflict is determined to exist, the credibility weight of each evidence source is calculated separately. This weight is then used to perform a weighted average of the basic trust assignment functions of the corresponding evidence sources, and the improved conflict coefficient is used to synthesize the processed basic trust assignment functions, thereby effectively reducing the negative impact of unreliable evidence on the fusion result. This mechanism addresses the technical deficiency of traditional DS evidence theory, which is prone to producing counterintuitive fusion results when evidence is highly conflicting, significantly improving the reliability and robustness of multi-source information fusion.

[0023] Next, combined Figures 2 to 6 This application provides a detailed description of the ship inspection plan.

[0024] In some embodiments, the YOLOv11n model is improved in three aspects: the C3k2-MSCB block is designed to enhance the perception of small targets; the SENetV2 attention mechanism is introduced into the C2PSA block to optimize feature channel modeling; and the Inner-EIoU loss function is used to replace the original CIoU loss function to improve the bounding box regression accuracy.

[0025] (1) Regarding the C3k2-MSCB block.

[0026] Multi-Scale Convolution Block (MSCB) is an information enhancement block that fuses multiple receptive fields, and its structure is as follows: Figure 2As shown, the MSCB block performs multi-scale feature extraction by paralleling three sets of depthwise separable convolutions (DWConv) with different kernel sizes (k=p,q,s) to enhance the receptive field of the feature map and achieve the collaborative expression of local details and global semantics.

[0027] This embodiment improves each C3k2 block in the backbone network to achieve an optimal balance between computational resources and detection performance. The network structure of the C3k2-MSCB block is as follows: Figure 3 As shown, the C3k-MSCB block connects two sets of MSCB blocks in series on the main path, performs two multi-scale feature extractions to obtain a wider receptive field and richer semantic information, and introduces local residual connections to maintain gradient flow stability.

[0028] In this embodiment, the Bottleneck sub-block in the last C3k2 block of the neck network is replaced with an MSCB block. This effectively controls the overall parameter count and computational cost of the model while maintaining a certain multi-scale feature response capability to extract multi-granularity information. This design allows the network to adaptively select the feature extraction granularity according to the target scale. Compared with the traditional Bottleneck, the MSCB block can effectively compensate for the blind spots in modeling some details of single-scale convolution without significantly increasing the computational burden, especially improving the detection accuracy of small targets on ships.

[0029] (2) Regarding the C2PSA block.

[0030] The C2PSA block is a structure in YOLOv11n used to enhance spatial attention. This embodiment introduces the channel attention mechanism SENetV2 into its core PSA block. The network structure of the SENetV2 block is as follows: Figure 4 As shown, SENetV2 first compresses the spatial dimension of the input feature map X into a global description vector z for each channel through global average pooling to extract channel information. Then, four parallel fully connected (FC) branches are introduced to perform dimensionality reduction and nonlinear mapping on the global description vector z. Each branch is followed by ReLU after the FC layer to learn different channel relationships and fine-grained features. This multi-branch structure effectively improves the diversity and complexity of feature representation. Subsequently, the four dimensionality-reduced features are concatenated along the channel dimension to form a new feature vector, integrating the complementary information learned by the multi-branch FC layers and enhancing the diversity of feature representation. This concatenated vector is mapped back to the original number of channels through a fully connected layer and a sigmoid activation function, resulting in channel attention weights with values ​​ranging from [0,1], used to highlight key channels and suppress redundant features, thereby improving the accuracy of channel selection and overall representation ability.

[0031] The network structure of the C2PSA block in this embodiment is as follows: Figure 5As shown, the input features are first divided into a retention branch and a processing branch through a 1×1 convolution. The processing branch then enters a cascaded set of n PSA blocks. Finally, the processed features and the retention branch are concatenated along the channel dimension, and the original number of channels is restored through a 1×1 convolution. This mechanism enhances the nonlinear interaction between channels through multi-branch fully connected layers, effectively preserving more detailed information and enabling the model to more accurately focus on key regions of ship targets.

[0032] (3) Regarding model training.

[0033] To simplify the optimization objective and improve regression accuracy, this embodiment uses EIoU loss, decoupling the aspect ratio penalty term into independent width and height penalties. The EIoU loss function expression is as follows: (1) in, This is the intersection-union ratio (IUU) of the sample detection bounding boxes and the sample annotation bounding boxes. Loss due to center distance; For width and height loss; , These represent the width and height of the smallest bounding rectangle formed by merging the sample detection box and the sample annotation box, respectively. , These are the center points of the sample detection box and the sample annotation box, respectively. The distance between the two center points. and These represent the width and height of the sample detection box, respectively. and These represent the width and height of the sample annotation box, respectively.

[0034] Compared to the traditional CIoU loss, the EIoU loss function in this application decomposes the complex aspect ratio term of CIoU loss into independent width and height penalty terms, directly optimizing the side length difference.

[0035] To further improve the performance of small target detection, this embodiment introduces the Inner-IoU mechanism. Inner-IoU calculates the loss value using the intersection-union ratio of the auxiliary control boxes. Inner-EIoU applies the geometric penalty term of EIoU to the auxiliary boxes. The loss function expression of Inner-EIoU is: (2) in, For geometric penalty terms; ; ; ; ; ; ; With a preset scaling factor, Inner-EIoU controls the scaling ratio of the auxiliary box relative to the original box by adjusting the ratio. It dynamically adjusts the scaling ratio according to the objects to be detected, which can adapt to targets of different sizes, balance the detection performance of large, medium and small scale targets, and improve the robustness to multi-scale targets. These are the coordinates of the left, right, top, and bottom boundaries of the sample detection box, respectively. These are the coordinates of the left, right, top, and bottom boundaries of the sample annotation box, respectively.

[0036] In practical applications, remote sensing images of ships in sea areas and ports can be collected to create a ship detection dataset; the original images can be preprocessed by size normalization and other methods; after expanding the samples by applying data augmentation techniques, the dataset can be divided into training set, validation set and test set to train the improved YOLOv11n model.

[0037] In this embodiment, all algorithms were written in Python 3.10, run on Windows 11, and utilized an NVIDIA GeForce RTX 4060 Ti GPU with 16GB of VRAM and 32GB of RAM. CUDA v11.8 was used to accelerate the GPU, and training was performed using the PyTorch 2.0.0 deep learning framework. The input image size was uniformly 640×640 pixels. The training hyperparameters were set as follows: SGD optimizer, initial learning rate 0.01, final learning rate factor 0.01; batch size 16; total number of iterations 200. During training, after each epoch, the mAP (mean accuracy) was evaluated on the validation set, and the model weight with the highest mAP on the validation set was saved as the best model.

[0038] To test the performance improvement of the method in ship detection according to the embodiments of this application, precision, recall, mean average precision (mAP), number of parameters, computational cost (GFLops), and frame rate (FPS) were used as performance evaluation indicators, and comparative experiments were conducted with benchmark models such as YOLOv12, YOLOv11, and RT-DETR. The comparison results of relevant indicators are shown in Table 1.

[0039] Table 1 Comparative Experiment

[0040] As can be seen from the comparative experimental results in Table 1, the improved model in this embodiment exhibits significant advantages in all key detection indicators. Compared with the baseline model YOLOv11n, the accuracy of the improved model increased from 87.4% to 89.3%, the mean precision (mAP@0.5) increased significantly from 85.0% to 88.5%, and the mAP@0.5:0.95 also improved from 64.0% to 66.0%.

[0041] Furthermore, while ensuring higher detection accuracy, the number of parameters in this model has been reduced to 2.509M, and the computational cost has been reduced to 6.2, achieving further weight reduction.

[0042] Compared with other mainstream lightweight target detection models such as the YOLO series and RT-DETR-n, the method proposed in this application has achieved the best results in terms of accuracy and average precision, which fully demonstrates that the improved model not only has high recognition accuracy but also low computational cost in ship detection tasks under complex conditions, and has strong comprehensive competitiveness.

[0043] Furthermore, to verify the effectiveness of each improved module, an ablation experiment was designed: C3k2-MSCB, SENetV2, or Inner-EIoU were removed respectively, and their performance metrics (including mAP, recall, and precision) on the test set were compared. The experimental results are shown in Table 2.

[0044] Table 2 Ablation Experiment

[0045] As shown in Table 2, the ablation experiments demonstrate that modular improvements can significantly enhance the detection performance and efficiency of the YOLOv11n model. By progressively introducing the MSCB block, SENetV2 block, and Inner-EIoU loss function into the YOLOv11n model, the model's detection performance has been continuously improved. Introducing either the MSCB block or the SENetV2 block individually can effectively improve precision and mean average precision (mAP) without increasing computational burden. Combining the two improves precision to 89.3%, mAP@0.5 to 88.2%, and the number of parameters to 2.509M. Finally, by introducing the Inner-EIoU loss function, the model maintains precision while improving recall to 79.2%, mAP@0.5 to a maximum of 88.5%, and mAP@0.5:0.95 to 66.0%.

[0046] In this embodiment, the test image size is adjusted to 640×640, and the pixel values ​​are normalized before being input into the model. The model outputs a three-dimensional tensor, containing the predicted bounding box coordinates, target confidence, and class probability. Dynamic threshold non-maximum suppression is used for post-processing. Finally, the filtered detection boxes are overlaid on the original image to output the detection results.

[0047] In some embodiments, Figure 1 Step S2 includes the process of designing and training the convolutional neural network model.

[0048] Due to the high cost and difficulty in labeling ship magnetic anomaly data in real marine environments, this embodiment employs a simulation method based on a magnetic dipole model to generate a training dataset. The ship target is represented as a magnetic dipole, and the three-component magnetic field expression of its generated magnetic anomaly signal at the sensor location is as follows: (3) in, The magnetic moment vector of the target. Let be the position vector of the sensor relative to the target. is the vacuum permeability.

[0049] By setting different detection altitudes, target tonnage, and sailing speeds, the magnetic moment parameters and background noise intensity are adjusted. Gaussian white noise and colored noise are superimposed on the signal, and linear trends and instantaneous spikes are introduced to simulate the complex and variable magnetic background interference in the real marine environment.

[0050] Based on this, a continuous magnetic anomaly signal sequence containing positive samples (with targets passing through) and negative samples (without targets) is generated, totaling 150,000 samples. Simultaneously, continuous soft labels are constructed based on the target passage time as a basis for supervised training.

[0051] Raw three-component magnetic field data and its main field First, median filtering (with a window length of 5 sampling points) removes instantaneous spike interference. Then, a high-pass filter eliminates linear trends and low-frequency drift to enhance the magnetic anomaly signal caused by the target's passage. Continuous soft labels constructed based on the target's passage time are converted into binary labels through hard thresholding, providing a basis for supervised model training.

[0052] Based on the preprocessed signal, the trend-removed signal and its first-order difference are extracted as features. Specifically, for each channel... First-order differences were calculated separately, resulting in an 8-dimensional feature vector containing four original signals and four difference signals. A sliding window was then used to slice the feature sequence, with a window length of 128 sampling points. The label of the center point of the window was used as the label of the sample, forming an input sample set suitable for time-series models. Finally, the dataset was divided into training, validation, and test sets in an 8:1:1 ratio.

[0053] In some embodiments, step S3 includes: The feature extraction network, including cascaded residual convolutional layers, is used to extract features from the input feature sequence at multiple scales. An attention network is used to perform global average pooling on the feature sequence output by the feature extraction network, and then generate channel attention weights through fully connected layers and activation functions. The channel attention weights are then multiplied element-wise with the feature sequence output by the feature extraction network to obtain an enhanced feature sequence. A temporal feature encoding network models the temporal dependencies of the enhanced feature sequence and outputs global temporal features. The fully connected layer maps the global temporal features and outputs the probability of the target's existence through an activation function.

[0054] For example, this embodiment designs a hybrid neural network model that combines a residual network (ResNet) and a bidirectional long short-term memory network (BiLSTM), as shown in the structure diagram below. Figure 6As shown, the input features are first passed through a Gaussian noise layer with a standard deviation of 0.02 to improve the model's robustness and generalization ability during training. Then, the data enters three cascaded residual convolutional blocks for multi-scale feature extraction. Each residual block contains two one-dimensional dilated convolutional layers, batch normalization, a ReLU activation function, and a spatial dropout layer, and is added to the backbone output via a 1x1 convolutional skip connection. The kernel sizes of these three residual blocks are 7, 5, and 3, with dilation rates of 1, 2, and 4, and the number of output channels increases sequentially to 32, 64, and 128. After completing all residual feature extraction, the model introduces a squeeze-and-excitation (SE) attention module. This module first performs global average pooling on the 128-channel feature map, then sequentially passes it through two fully connected layers with 16 and 128 neurons, and combines this with a Sigmoid activation function to generate channel attention weights. These weights are multiplied by the original feature map to recalibrate the feature channels and enhance key features. Subsequently, the attention-weighted feature sequence is fed into a bidirectional LSTM (BiLSTM) layer with 64 hidden units. This layer is configured to output only the feature vector of the last time step, followed by a layer with a dropout rate of 0.5 to further prevent overfitting. Finally, the extracted global temporal features are sequentially fed into two fully connected layers with 32 and 1 neurons respectively, and the probability of the target's presence is output through a sigmoid activation function.

[0055] The convolutional neural network model uses the Adam optimizer with an initial learning rate of 0.0005, a batch size of 64, and 60 training epochs. Early stopping and learning rate decay are employed; the model stops if the validation set loss does not decrease for 10 consecutive epochs, and the learning rate is reduced to 0.5 when the validation loss stagnates. The best model is saved based on the F1 score of the validation set. During the testing phase, the probability sequence output by the model is smoothed using a moving average (window size 15), and post-processed with hysteresis threshold comparison (high threshold 0.75, low threshold 0.35) and morphological closing operations to obtain the final binary classification detection results.

[0056] After training the convolutional neural network model, the multi-channel magnetic field sequence caused by the target ship is input into the convolutional neural network model to obtain a binary classification detection result. The multi-channel magnetic field sequence is obtained through the following steps: Acquire multi-channel raw magnetic field data collected by the sensor, wherein the multi-channel raw magnetic field data includes three-component magnetic field data and their composite total field data; After preprocessing the multi-channel raw magnetic field data, the first-order difference of each channel data is calculated, and the raw magnetic field data of each channel is combined with the first-order difference data to form a multi-dimensional feature vector. The multi-channel magnetic field sequence is obtained by performing time-series slicing on the multi-dimensional feature vector using a sliding window.

[0057] After obtaining remote sensing detection results and magnetic anomaly detection results through the above embodiments, the DS evidence theory synthesis algorithm is used for fusion decision-making to obtain the final detection results.

[0058] First, a multimodal decision-level fusion framework is constructed.

[0059] Remote sensing and magnetic anomaly detection results are used as two independent sources of evidence. For remote sensing results, a confidence score is output for each detection box. , representing the probability of the target's existence, and also outputting the target category. For magnetic anomaly detection results, a target existence probability p is output at each time point.

[0060] Therefore, this embodiment defines the recognition framework. Where T represents the existence of the target. This indicates that the target does not exist.

[0061] For each source of evidence, construct a basic trust assignment function. Regarding remote sensing image evidence, let Regarding evidence of magnetic anomalies, let Without considering uncertainty, .

[0062] It is worth noting that the two sources of evidence need to be spatiotemporally aligned before evidence fusion.

[0063] Since remote sensing image detection provides image target information at a specific moment, while magnetic anomaly detection provides target passage information over a continuous time series, it is necessary to ensure that they correspond to targets at the same spatiotemporal location. In this embodiment, matching is performed using timestamps and geographic locations: for each remote sensing detection result, its image acquisition time and geographic location are recorded; for magnetic anomaly detection results, each time point corresponds to a detection location. A time matching window Δt and a spatial matching radius Δd are set. If the time difference between a magnetic anomaly detection result and the remote sensing detection time is less than Δt and the spatial distance is less than Δd, then they are considered to correspond to the same target. If multiple matches exist, the match result with the smallest weighted sum of time difference and spatial distance is selected.

[0064] For detection results that cannot be matched, their trust assignments are retained separately, but single-source evidence must be considered when merging them.

[0065] Then, the improved conflict coefficient is calculated. In the traditional DS evidence theory, the conflict coefficient... Used to measure the degree of conflict between pieces of evidence, defined as: (4) in, This is the basic trust assignment function when the remote sensing detection results are used as a source of evidence. Let be the basic trust assignment function when the magnetic anomaly detection result is used as a source of evidence. The propositions supported by the basic trust assignment function corresponding to the remote sensing detection results. The propositions supported by the basic trust assignment function corresponding to the magnetic anomaly detection results, and , This indicates that the proposition of the objective exists. This indicates that the objective does not exist.

[0066] To more accurately reflect the conflicting nature of the evidence, an improved conflict coefficient is introduced. Consider the local consistency and global conflict distribution among the evidence: (5) in, The preset balance coefficient is used to adjust the traditional conflict coefficient. The weight of the newly introduced evidence difference term in the improved conflict coefficient.

[0067] For example, setting a conflict threshold , .

[0068] when Exceeding the set threshold When conflicting evidence is identified, the evidence undergoes preprocessing. A weighted average method is used to adjust the basic trust allocation for any proposition. : , (6) Among them, weight Determined by the credibility of the evidence, it is defined as: (7) Information entropy is the information value of evidence, reflecting the uncertainty of the evidence. The lower the information entropy, the more credible the evidence, and the higher its weight.

[0069] For the binary framework, specifically: (8) Finally, based on the fused basic trust allocation, the proposition with the highest trust level is selected as the final detection result.

[0070] like If the threshold is ≥0.5, the target is identified as a ship; otherwise, it is a false alarm. This decision threshold of 0.5 can be adjusted according to actual needs.

[0071] For unmatched single-source evidence, its basic trust assignment can be used directly for decision-making, or it can be considered as fused with empty evidence.

[0072] The above fusion method can effectively integrate remote sensing images and magnetic anomaly information, reduce false alarms and missed detections of single modes, and improve the accuracy and robustness of ship detection in complex environments.

[0073] Figure 7 This is a schematic diagram of an electronic device illustrated in this specification according to an exemplary embodiment. Please refer to... Figure 7 At the hardware level, the device includes a processor 710, an internal bus 720, a network interface 730, memory 740, a hardware acceleration device 750, and non-volatile memory 760, and may also include other hardware required for its functions. One or more embodiments of this application can be implemented in software, for example, the processor 710 reads the corresponding computer program from the non-volatile memory 760 into the memory 740 and then runs it. Of course, in addition to software implementation, one or more embodiments of this application do not exclude other implementation methods, such as logic devices or a combination of hardware and software, etc. That is to say, the execution subject of the above processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0074] Figure 8 This is a structural block diagram of a deep learning-based ship detection device illustrated in an exemplary embodiment of this application. The ship detection device can be applied to, for example... Figure 7 The electronic device shown implements the technical solution of this application. The ship detection device includes: a data acquisition unit 810, an image detection unit 820, a magnetic anomaly detection unit 830, and a fusion decision unit 840, wherein: The data acquisition unit 810 is used to acquire remote sensing images of the target ship and multi-channel magnetic field data caused by the target ship. Image detection unit 820 is used to perform target detection on the remote sensing image using an improved YOLOv11n model to obtain remote sensing detection results; In the improved YOLOv11n model, the C3k sub-blocks of each C3k2 block in the backbone network are replaced with multiple C3k-MSCB blocks connected in series. Each C3k-MSCB block contains a main path and parallel branches. The main path contains a convolutional layer and two MSCB blocks connected in series after the convolutional layer. Each parallel branch contains a convolutional layer. The output of the main path and the output of the parallel branch are concatenated and then input into a convolutional layer for feature fusion and channel adjustment. In the improved YOLOv11n model, the Bottleneck sub-block in the last C3k2 block of the neck network is replaced with the MSCB block; Furthermore, the PSA sub-block of the C2PSA block of the improved YOLOv11n model includes a main branch and two residual branches. The main branch includes a SENetV2 block and two convolutional layers connected in series after the SENetV2 block. The output of the SENetV2 block is concatenated with the input of the PSA sub-block through the first residual branch. The concatenated feature is processed by the two convolutional layers and then concatenated with the concatenated feature again through the second residual branch before being output. The magnetic anomaly detection unit 830 is used to perform anomaly detection on the multi-channel magnetic field data using a trained convolutional neural network model to obtain a magnetic anomaly detection result, which is used to indicate whether the ship target exists. The fusion decision unit 840 is used to take the remote sensing detection result and the magnetic anomaly detection result as two evidence sources, construct basic trust assignment functions for each, and use the DS evidence theory synthesis algorithm to fuse the basic trust assignment functions of the two evidence sources to obtain the final detection result. The DS evidence theory synthesis algorithm determines whether there is a conflict between the evidence sources by calculating an improved conflict coefficient. When a conflict is determined to exist, the credibility weight of each evidence source is calculated. The basic trust allocation function of the corresponding evidence source is weighted and averaged using the credibility weight. Then, the improved conflict coefficient is used to synthesize the processed basic trust allocation function. The expression for the improved conflict coefficient is: ; in, For the improved conflict coefficient; The original conflict coefficient, This is the basic trust assignment function when the remote sensing detection results are used as a source of evidence. This is the basic trust assignment function when the magnetic anomaly detection result is used as a source of evidence. The propositions supported by the basic trust assignment function corresponding to the remote sensing detection results. The propositions supported by the basic trust assignment function corresponding to the magnetic anomaly detection results, and , This indicates that the proposition of the objective exists. The proposition indicating that the objective does not exist; This is the preset balance coefficient.

[0075] In some embodiments, the multi-channel magnetic field data is a multi-channel magnetic field sequence. The data acquisition unit 810 is used to acquire the multi-channel raw magnetic field data collected by the sensor. The multi-channel raw magnetic field data includes three-component magnetic field data and its composite total field data. After preprocessing the multi-channel raw magnetic field data, the first-order difference of each channel data is calculated. The raw magnetic field data of each channel is combined with the first-order difference data to form a multi-dimensional feature vector. The multi-dimensional feature vector is time-series sliced ​​using a sliding window to obtain the multi-channel magnetic field sequence.

[0076] In some embodiments, the fusion decision unit 840 is used to compare the improved conflict coefficient with a preset conflict threshold, and if the improved conflict coefficient is greater than the preset conflict threshold, it is determined that there is a conflict between the evidence sources.

[0077] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this application according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0078] Accordingly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the methods described in any of the above embodiments.

[0079] Accordingly, embodiments of this application also provide a computer program product configured to perform the methods described in any of the above embodiments.

[0080] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer, which can take the form of a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email sending and receiving device, game console, tablet computer, wearable device, or any combination of these devices.

[0081] In a typical configuration, a computer includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0082] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0083] Computer-readable media, including both permanent and non-permanent, removable and non-removable media, can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random-access memory (SRAM), dynamic random-access memory (DRAM), other types of random-access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, disk storage, quantum memory, graphene-based storage media or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0084] While this specification contains numerous specific implementation details, these should not be construed as limiting the scope of any invention or the scope of the claims, but rather are primarily intended to describe features of specific embodiments of a particular invention. Certain features described in the various embodiments herein may also be implemented in combination in a single embodiment. Conversely, various features described in a single embodiment may also be implemented separately in various embodiments or in any suitable sub-combination. Furthermore, while features may function in certain combinations as described above and even initially claimed in this way, one or more features from a claimed combination may be removed from that combination in some cases, and a claimed combination may refer to a sub-combination or a variation thereof.

[0085] Similarly, although the operations are depicted in a specific order in the accompanying drawings, this should not be construed as requiring these operations to be performed in the specific order shown or sequentially, or requiring all illustrated operations to be performed to achieve the desired result. In some cases, multitasking and parallel processing may be advantageous. Furthermore, the separation of various system modules and components in the above embodiments should not be construed as requiring such separation in all embodiments, and it should be understood that the described program components and systems can generally be integrated together in a single software product or packaged into multiple software products.

[0086] Thus, specific embodiments of the subject matter have been described. Other embodiments are within the scope of the appended claims. In some cases, the actions recited in the claims may be performed in a different order and still achieve the desired result. Furthermore, the processes depicted in the drawings are not necessarily shown in a specific order or sequence to achieve the desired result. In some implementations, multitasking and parallel processing may be advantageous.

[0087] It should be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0088] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A ship detection method based on deep learning, characterized in that, The method includes the following steps: Step S1: Acquire remote sensing images of the target ship and multi-channel magnetic field data caused by the target ship; Step S2: Target detection is performed on the remote sensing image using the improved YOLOv11n model to obtain remote sensing detection results; In the improved YOLOv11n model, the C3k sub-blocks of each C3k2 block in the backbone network are replaced with multiple C3k-MSCB blocks connected in series. Each C3k-MSCB block contains a main path and parallel branches. The main path contains a convolutional layer and two MSCB blocks connected in series after the convolutional layer. Each parallel branch contains a convolutional layer. The output of the main path and the output of the parallel branch are concatenated and then input into a convolutional layer for feature fusion and channel adjustment. In the improved YOLOv11n model, the Bottleneck sub-block in the last C3k2 block of the neck network is replaced with the MSCB block; Furthermore, the PSA sub-block of the C2PSA block of the improved YOLOv11n model includes a main branch and two residual branches. The main branch includes a SENetV2 block and two convolutional layers connected in series after the SENetV2 block. The output of the SENetV2 block is concatenated with the input of the PSA sub-block through the first residual branch. The concatenated feature is processed by the two convolutional layers and then concatenated with the concatenated feature again through the second residual branch before being output. Step S3: Use the trained convolutional neural network model to perform anomaly detection on the multi-channel magnetic field data to obtain magnetic anomaly detection results. The magnetic anomaly detection results are used to indicate whether the ship target exists. Step S4: The remote sensing detection result and the magnetic anomaly detection result are used as two evidence sources. Basic trust assignment functions are constructed for each source. The DS evidence theory synthesis algorithm is used to fuse the basic trust assignment functions of the two evidence sources to obtain the final detection result. The DS evidence theory synthesis algorithm determines whether there is a conflict between the evidence sources by calculating an improved conflict coefficient. When a conflict is determined to exist, the credibility weight of each evidence source is calculated. The basic trust allocation function of the corresponding evidence source is weighted and averaged using the credibility weight. Then, the improved conflict coefficient is used to synthesize the processed basic trust allocation function. The expression for the improved conflict coefficient is as follows: ； in, For the improved conflict coefficient; The original conflict coefficient, This is the basic trust assignment function when the remote sensing detection results are used as a source of evidence. This is the basic trust assignment function when the magnetic anomaly detection result is used as a source of evidence. The propositions supported by the basic trust assignment function corresponding to the remote sensing detection results. The propositions supported by the basic trust assignment function corresponding to the magnetic anomaly detection results; This is the preset balance coefficient.

2. The method according to claim 1, characterized in that, The improved YOLOv11n model in step S2 is trained using a joint loss function. The expression is: ； in, For EIou loss term, For center distance loss, For the loss of width and height, , These represent the width and height of the smallest bounding rectangle formed by merging the sample detection box and the sample annotation box, respectively. , These are the center points of the sample detection box and the sample annotation box, respectively. The distance between the two center points. and These represent the width and height of the sample detection box, respectively. and These represent the width and height of the sample annotation box, respectively. This is the intersection-union ratio (IUU) of the sample detection bounding boxes and the sample annotation bounding boxes. For geometric penalty terms, , , , , , , The preset scaling factor. These are the coordinates of the left, right, top, and bottom boundaries of the sample detection box, respectively. These are the coordinates of the left, right, top, and bottom boundaries of the sample annotation box, respectively.

3. The method according to claim 1, characterized in that, The convolutional neural network model in step S3 includes: The feature extraction network, including cascaded residual convolutional layers, is used to extract features from the input feature sequence at multiple scales. An attention network is used to perform global average pooling on the feature sequence output by the feature extraction network, and then generate channel attention weights through fully connected layers and activation functions. The channel attention weights are then multiplied element-wise with the feature sequence output by the feature extraction network to obtain an enhanced feature sequence. A temporal feature encoding network models the temporal dependencies of the enhanced feature sequence and outputs global temporal features. The fully connected layer maps the global temporal features and outputs the probability of the target's existence through an activation function.

4. The method according to claim 1, characterized in that, The multi-channel magnetic field data in step S1 is a multi-channel magnetic field sequence, obtained through the following steps: Acquire multi-channel raw magnetic field data collected by the sensor, wherein the multi-channel raw magnetic field data includes three-component magnetic field data and their composite total field data; After preprocessing the multi-channel raw magnetic field data, the first-order difference of each channel data is calculated, and the raw magnetic field data of each channel is combined with the first-order difference data to form a multi-dimensional feature vector. The multi-channel magnetic field sequence is obtained by performing time-series slicing on the multi-dimensional feature vector using a sliding window.

5. The method according to claim 1, characterized in that, In step S4, the existence of a conflict is determined through the following steps: The improved conflict coefficient is compared with a preset conflict threshold. If the improved conflict coefficient is greater than the preset conflict threshold, it is determined that there is a conflict between the evidence sources.

6. A ship detection device based on deep learning, characterized in that, The device includes: The data acquisition unit is used to acquire remote sensing images of the target ship and multi-channel magnetic field data caused by the target ship. An image detection unit is used to perform target detection on the remote sensing image using an improved YOLOv11n model to obtain remote sensing detection results; In the improved YOLOv11n model, the C3k sub-blocks of each C3k2 block in the backbone network are replaced with multiple C3k-MSCB blocks connected in series. Each C3k-MSCB block contains a main path and parallel branches. The main path contains a convolutional layer and two MSCB blocks connected in series after the convolutional layer. Each parallel branch contains a convolutional layer. The output of the main path and the output of the parallel branch are concatenated and then input into a convolutional layer for feature fusion and channel adjustment. In the improved YOLOv11n model, the Bottleneck sub-block in the last C3k2 block of the neck network is replaced with the MSCB block; Furthermore, the PSA sub-block of the C2PSA block of the improved YOLOv11n model includes a main branch and two residual branches. The main branch includes a SENetV2 block and two convolutional layers connected in series after the SENetV2 block. The output of the SENetV2 block is concatenated with the input of the PSA sub-block through the first residual branch. The concatenated feature is processed by the two convolutional layers and then concatenated with the concatenated feature again through the second residual branch before being output. The magnetic anomaly detection unit is used to perform anomaly detection on the multi-channel magnetic field data using a trained convolutional neural network model, and obtain a magnetic anomaly detection result, which is used to indicate whether the ship target exists. The fusion decision unit is used to take the remote sensing detection results and the magnetic anomaly detection results as two evidence sources, construct basic trust assignment functions for each, and use the DS evidence theory synthesis algorithm to fuse the basic trust assignment functions of the two evidence sources to obtain the final detection result. The DS evidence theory synthesis algorithm determines whether there is a conflict between the evidence sources by calculating an improved conflict coefficient. When a conflict is determined to exist, the credibility weight of each evidence source is calculated. The basic trust allocation function of the corresponding evidence source is weighted and averaged using the credibility weight. Then, the improved conflict coefficient is used to synthesize the processed basic trust allocation function. The expression for the improved conflict coefficient is as follows: ； in, For the improved conflict coefficient; The original conflict coefficient, This is the basic trust assignment function when the remote sensing detection results are used as a source of evidence. This is the basic trust assignment function when the magnetic anomaly detection result is used as a source of evidence. The propositions supported by the basic trust assignment function corresponding to the remote sensing detection results. The propositions supported by the basic trust assignment function corresponding to the magnetic anomaly detection results, and , This indicates that the proposition of the objective exists. The proposition indicating that the objective does not exist; This is the preset balance coefficient.

7. An electronic device, characterized in that, include: processor; as well as A computer-readable storage medium storing computer program instructions that, when executed by the processor, cause the processor to perform the method as described in any one of claims 1 to 5.

Images