Remote sensing image ship segmentation method, system, equipment and medium

Abstract

The invention provides a remote sensing image ship segmentation method, system and device and a medium, and the method comprises the following steps: obtaining a to-be-processed remote sensing image, and carrying out the preprocessing of the to-be-processed remote sensing image, and obtaining a preprocessed remote sensing image; processing the preprocessed remote sensing image by using the trained remote sensing image ship segmentation model to obtain a corresponding remote sensing image ship segmentation result; the processing of the preprocessed remote sensing image comprises the following steps: obtaining a feature map by using a feature map extraction network, obtaining a deep enhanced feature map by using a feature map enhancement network, and obtaining a remote sensing image ship segmentation result by using a ship segmentation network. According to the method, the generalization ability of a remote sensing image ship segmentation model is improved by combining a composite loss function and an Adam optimizer with a cosine annealing strategy, and by combining operations such as depth separable convolution and multi-scale feature fusion, optimization processing can be performed on different levels of feature maps, and the capture and segmentation accuracy of ship features under a complex background is enhanced.

Description

Technical Field

[0001] The present invention relates to the field of remote sensing image processing, and more particularly to a remote sensing image ship segmentation method, system, equipment and medium. Background Art

[0002] With the development of remote sensing technology, it is becoming increasingly important to automatically identify and segment ships from high-resolution satellite images. This capability has broad application prospects in maritime monitoring, maritime traffic management, environmental protection, and military applications. Remote sensing image ship segmentation aims to accurately extract the contour information of the ship from the complex ocean background, providing key data support for the above application scenarios.

[0003] Existing ship segmentation methods usually rely on a single loss function and a fixed optimization strategy during the training process, which makes it difficult for the model to find the best balance between classification accuracy and robustness. Traditional loss functions may not be able to fully capture the difference between foreground and background pixels, especially when dealing with small target objects such as ships, the detection accuracy is often insufficient. In addition, the fixed learning rate adjustment mechanism makes the model converge slowly in the early stage of training, and it is easy to fall into local optimal solutions or overfitting problems in the later stage, thus limiting the generalization ability of the model and the actual application effect.

[0004] In terms of feature extraction and enhancement, most traditional methods use simple convolution operations, which lack effective fusion of multi-scale features and enhanced processing of targets under complex backgrounds. This method not only makes it difficult to retain the original image information, but may also make the ship features not prominent, affecting the accuracy of the final segmentation results. Especially when faced with complex backgrounds and multi-directional and multi-scale ship features, the limitations of traditional feature extraction methods are more obvious, and it is difficult to meet the needs of high-precision remote sensing image ship segmentation.

[0005] Therefore, how to design a remote sensing image ship segmentation method that can not only accurately extract ship contours under complex backgrounds, but also optimize the training process, improve detection accuracy and generalization performance, and ensure high-precision and robust segmentation results through effective multi-scale feature enhancement technology is an urgent problem that technicians in this field need to solve. Summary of the invention

[0006] In view of this, the present invention provides a remote sensing image ship segmentation method, which not only enhances the small target detection accuracy and the generalization ability of the model by adopting a composite loss function and an Adam optimizer combined with a cosine annealing strategy. In terms of feature extraction and enhancement, the deep separable convolution, layer normalization, and multi-scale feature fusion techniques are used to significantly improve the capture ability and segmentation accuracy of ship features under complex backgrounds, thereby achieving high-quality ship segmentation results.

[0007] In order to achieve the above object, the present invention adopts the following technical solution:

[0008] In a first aspect, the present invention provides a remote sensing image ship segmentation method, comprising the following steps:

[0009] S1, obtaining a remote sensing image to be processed and preprocessing it to obtain a preprocessed remote sensing image;

[0010] S2, training a remote sensing image ship segmentation model, and using the trained remote sensing image ship segmentation model to process the pre-processed remote sensing image to obtain a corresponding remote sensing image ship segmentation result;

[0011] The processing of the pre-processed remote sensing image includes: obtaining a feature map using a feature map extraction network, obtaining a depth enhanced feature map using a feature map enhancement network, and obtaining a remote sensing image ship segmentation result using a ship segmentation network.

[0012] Furthermore, in S2, a composite loss function L is used to train the remote sensing image ship segmentation model; the composite loss function L is expressed as:

[0013] L=λ1L a +λ2L b

[0014]

[0015] Among them, λ1 and λ2 are weight coefficients, L a is the cross entropy loss function, L b is the Dice coefficient loss function; P true (i) is the image The true label of the i-th pixel, Predict probabilities for the model.

[0016] Furthermore, in S2, the remote sensing image ship segmentation model training is performed using the Adam optimizer to update parameters; the parameter update rule of the Adam optimizer is:

[0017]

[0018] Among them, θ t+1 represents the updated parameters, θ t represents the current parameter, η represents the learning rate, m t and v t denote the estimates of the first-order moment and the second-order moment, respectively, and ε denotes a constant to prevent division by zero.

[0019] Furthermore, the Adam optimizer also combines the cosine annealing strategy to adjust the learning rate;

[0020]

[0021] Among them, η t Represents the learning rate when the training step is t, T max represents the total number of training steps, and η0 represents the initial learning rate.

[0022] Furthermore, in S2, obtaining a feature map using a feature map extraction network includes:

[0023] The preprocessed remote sensing image I is processed by 3*3 convolution and activation function to obtain the feature map X; wherein the size of the feature map X is H×W×C, H represents height, W represents width, and C represents the number of channels.

[0024] Furthermore, in S2, obtaining a depth enhanced feature map using a feature map enhancement network includes:

[0025] Perform layer normalization on the feature map X to obtain the layer normalized feature map X norm ;

[0026] The feature map X norm Perform 3*3 depth-separable convolution processing to obtain feature map X1;

[0027] For feature map X norm Add it to the feature map X1 to get the feature map X ’ 1;

[0028] The feature map X ’ 1 Perform 1×1 convolution to obtain feature map X o ;

[0029] For feature map X o Perform 7×7 convolution to obtain feature map X a ; and the feature map X o Perform 3*3 depth separable convolution and upsampling to obtain feature maps

[0030] The feature map Perform subtraction and multiplication respectively to obtain the feature map and

[0031] For feature maps and Perform 3*3 depth separable convolution processing respectively, and convert the feature map and The 3*3 depth-separable convolution processing results are spliced ​​to obtain the feature map

[0032] The feature map Perform 1×1 convolution to obtain the feature map

[0033] For feature maps and feature map X a Perform splicing to obtain the preliminary enhanced feature map X en ;

[0034] The initial enhanced feature map X en Perform 1×1 convolution processing and add the result of 1×1 convolution processing to feature map X to obtain feature map X ’ en ;

[0035] For feature map X ’ en Perform layer normalization to obtain the layer normalized feature map X ’ norm ;

[0036] The feature map X ’ norm Perform 1×1 convolution to obtain the feature map

[0037] For feature maps Perform 3*3 depth separable convolution processing to obtain feature maps

[0038] The feature map and feature map Perform addition processing to obtain the feature map

[0039] For feature maps Perform 1×1 convolution and combine the 1×1 convolution result with the feature map X ’ norm Perform addition processing to obtain a deep enhanced feature map

[0040] Furthermore, in S2, obtaining a remote sensing image ship segmentation result using a ship segmentation network includes:

[0041] Deep Enhanced Feature Map Perform 1×1 convolution layer, upsampling layer and activation function processing to obtain the segmentation result image

[0042] In a second aspect, this embodiment provides a remote sensing image ship segmentation system, comprising:

[0043] Preprocessing module: used to obtain the remote sensing image to be processed and perform preprocessing to obtain the preprocessed remote sensing image;

[0044] Ship segmentation module: Train the remote sensing image ship segmentation model, and use the trained remote sensing image ship segmentation model to process the pre-processed remote sensing image to obtain the corresponding remote sensing image ship segmentation result;

[0045] The processing of the pre-processed remote sensing image includes: obtaining a feature map using a feature map extraction network, obtaining a depth enhanced feature map using a feature map enhancement network, and obtaining a remote sensing image ship segmentation result using a ship segmentation network.

[0046] In a third aspect, the present invention provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned remote sensing image ship segmentation method when executing the computer program.

[0047] In a fourth aspect, the present invention provides a computer-readable storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the remote sensing image ship segmentation method described above is implemented.

[0048] The descriptions of the second to fourth aspects of the present invention can refer to the detailed description of the first aspect; and the beneficial effects of the descriptions of the second to fourth aspects can refer to the analysis of the beneficial effects of the first aspect, which will not be repeated here.

[0049] It can be seen from the above technical solution that compared with the prior art, the present invention has the following beneficial effects:

[0050] 1. During the training of the remote sensing image ship segmentation model, by introducing the linear combination of the cross entropy loss function and the Dice coefficient loss function as a composite loss function, not only the ability to classify foreground and background pixels is enhanced, but also the detection accuracy of small target objects (such as ships) is improved. In addition, the Adam optimizer and its combined cosine annealing learning rate adjustment strategy are used to enable the model to converge quickly in the early stage of training, and fine-tune the parameters in the later stage to avoid overfitting, thereby obtaining better generalization performance.

[0051] 2. In the process of processing the pre-processed remote sensing images, based on the careful design of the remote sensing image ship segmentation model for different levels of feature maps, including but not limited to multiple convolution operations, addition and splicing operations, and upsampling processing, this solution can better retain the original image information while highlighting the ship features and improve the accuracy of the segmentation results. In particular, the alternating use of 1×1 convolution and 3×3 depth-separable convolution achieves efficient feature extraction and enhancement, ensuring the high-quality output of the final ship segmentation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0053] Figure 1 A flow chart of a remote sensing image ship segmentation method provided by an embodiment of the present invention;

[0054] Figure 2 A schematic diagram of a process for processing a pre-processed remote sensing image provided by an embodiment of the present invention;

[0055] Figure 3 A schematic diagram of a process for obtaining a depth enhanced feature map using a feature map enhancement network provided by an embodiment of the present invention;

[0056] Figure 4 A schematic diagram of the structure of a remote sensing image ship segmentation system provided by an embodiment of the present invention;

[0057] Figure 5 A schematic diagram of the structure of an electronic device is provided for an embodiment of the present invention. DETAILED DESCRIPTION

[0058] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only 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 creative work are within the scope of protection of the present invention.

[0059] The detection method provided in the embodiment of the present application can be applied to a remote sensing image ship segmentation server. The above-mentioned remote sensing image ship segmentation server can be hardware or software. When the remote sensing image ship segmentation server is hardware, it can be implemented as a distributed server cluster providing detection services, or it can be implemented as a single server. When the remote sensing image ship segmentation server is software, it can be installed in the servers listed above. It can be implemented as multiple software or software modules, or it can be implemented as a single software or software module, which is not specifically limited here.

[0060] Embodiment 1;

[0061] like Figure 1 As shown, this embodiment provides a remote sensing image ship segmentation method, comprising the following steps:

[0062] S1, obtaining a remote sensing image to be processed and preprocessing it to obtain a preprocessed remote sensing image;

[0063] S2, training a remote sensing image ship segmentation model, and using the trained remote sensing image ship segmentation model to process the pre-processed remote sensing image to obtain a corresponding remote sensing image ship segmentation result;

[0064] like Figure 2 As shown, the processing of the preprocessed remote sensing image includes: using a feature map extraction network to obtain a feature map, using a feature map enhancement network to obtain a depth enhanced feature map, and using a ship segmentation network to obtain a remote sensing image ship segmentation result.

[0065] This method improves the generalization ability of the remote sensing image ship segmentation model through a composite loss function and Adam optimizer combined with a cosine annealing strategy. It also combines operations such as deep separable convolution and multi-scale feature fusion to optimize the feature maps at different levels, thereby enhancing the capture and segmentation accuracy of ship features under complex backgrounds.

[0066] The above steps and related technical features are further described in detail below:

[0067] In this embodiment S1, a remote sensing image to be processed is acquired and preprocessed to obtain a preprocessed remote sensing image;

[0068] In the remote sensing image ship segmentation method, the S1 step is the key to ensure the quality of subsequent analysis. First, multispectral or hyperspectral remote sensing images covering the target sea area are obtained through high-resolution remote sensing platforms such as satellites, aerial photography or drones. These images may contain multiple bands such as visible light and near-infrared, providing a rich information basis for ship identification. After acquisition, the original image needs to undergo a series of preprocessing operations to improve data quality and consistency. This includes but is not limited to radiation correction to standardize image brightness values, geometric correction to make pixel coordinates accurately correspond to geographic coordinates, and atmospheric correction to reduce the effects of atmospheric scattering and absorption. In addition, in order to remove noise interference in the image, applying spatial domain filters such as mean filtering or adaptive median filtering can effectively reduce the level of random noise and improve image clarity. At the same time, in the case of cloud occlusion, cloud detection algorithms are used to remove clouds to ensure the purity of the image background.

[0069] Furthermore, enhancement processing and format adjustment are required. Enhancement processing involves contrast stretching, histogram equalization or adaptive histogram equalization to improve the visual effect of the image, highlight the characteristics of the ship, and especially to improve the contrast between the target and the background in a complex background. For specific task requirements, image fusion technology can also be implemented to combine data from different sensors to obtain a higher quality synthetic image. Subsequently, according to the input requirements of the ship segmentation model, the image size is uniformly adjusted and cropped into appropriate sub-image blocks as needed to facilitate batch processing and accelerate calculations. Finally, in order to expand the scale of the training set and increase the robustness of the model, a variety of data enhancement techniques are applied to the pre-processed images, such as rotation, flipping, scaling, and translation transformations to simulate different observation conditions, thereby ensuring that the model can adapt to the challenges of ship recognition in various actual scenarios.

[0070] In this embodiment S2, a remote sensing image ship segmentation model is trained, and the pre-processed remote sensing image is processed using the trained remote sensing image ship segmentation model to obtain a corresponding remote sensing image ship segmentation result;

[0071] Specifically, the remote sensing image ship segmentation model is trained in combination with a dataset that contains a large number of high-resolution remote sensing images covering a variety of marine environments, weather conditions, and lighting conditions to ensure the generalization ability of the model. Each image needs to be equipped with accurately labeled ship contour information to form a pixel-level true value label for supervised learning. In addition, the dataset should also include ship samples of different scales and directions, as well as examples in complex backgrounds, such as ports, busy waterways, etc., to enhance the model's adaptability to diverse and challenging situations.

[0072] At the same time, in order to increase the diversity of training samples and prevent overfitting, the original data set is expanded through data enhancement techniques (such as rotation, flipping, scaling, etc.), which not only helps to improve the detection accuracy of the model, but also ensures its reliability and robustness in practical applications.

[0073] The composite loss function L is used to train the remote sensing image ship segmentation model; the composite loss function L is expressed as:

[0074] L=λ1L a +λ2L b

[0075]

[0076] Among them, λ1 and λ2 are weight coefficients, L a is the cross entropy loss function, L b is the Dice coefficient loss function; P true (i) is the image The true label of the i-th pixel, Predict probabilities for the model.

[0077] By adjusting the weights λ1 and λ2, the best balance can be found between classification accuracy and region overlap. a It is used to measure the difference between the model prediction probability and the true label to ensure the classification accuracy of each pixel; the Dice coefficient loss function L b It focuses on improving the similarity between the predicted results and the actual target area, which is particularly suitable for enhancing the detection accuracy when dealing with small target objects such as ships.

[0078] Furthermore, the Adam optimizer is used to update parameters during the training of the pre-trained remote sensing image ship segmentation model; the parameter update rule of the Adam optimizer is:

[0079]

[0080] Among them, θ t+1 represents the updated parameters, θ t represents the current parameter, η represents the learning rate, m t and v t denote the estimates of the first-order moment and the second-order moment, respectively, and ε denotes a constant to prevent division by zero.

[0081] Specifically, the parameter update rule of the Adam optimizer comprehensively considers the first-order moment estimate m t and the second-order moment estimate v t The bias correction mechanism is used to ensure the accuracy of the initial update direction. In addition, a minimum value ε is introduced to prevent division by zero errors and ensure stable calculations. This optimization strategy not only helps to quickly converge to the global optimal solution, but also effectively avoids the problem of local optimal solutions, thus improving the efficiency and stability of model training.

[0082] Furthermore, in order to maintain efficient optimization capabilities of the remote sensing image ship segmentation model throughout the entire training cycle, the Adam optimizer also combines the cosine annealing strategy to adjust the learning rate;

[0083]

[0084] Among them, η t Represents the learning rate when the training step is t, T max represents the total number of training steps, and η0 represents the initial learning rate.

[0085] According to the number of training steps t and the total number of training steps T, the learning rate η tIt will gradually decrease according to the cosine curve, starting from the initial learning rate η0 and dropping to a lower level near the end of training. It allows the remote sensing image ship segmentation model to quickly explore the parameter space at a higher learning rate in the early stage of training, and fine-tune the parameters at a finer pace in the later stage to avoid overfitting and improve generalization performance. In this way, the remote sensing image ship segmentation model can obtain the most suitable learning rate at different stages, thereby achieving more robust and accurate ship segmentation results.

[0086] Furthermore, the pre-processed remote sensing image is processed based on the trained remote sensing image ship segmentation model, specifically including:

[0087] The feature map is obtained using the feature map extraction network, including:

[0088] The preprocessed remote sensing image I is processed by 3*3 convolution and activation function to obtain the feature map X; wherein the size of the feature map X is H×W×C, H represents height, W represents width, and C represents the number of channels.

[0089] like Figure 3 As shown, the feature map enhancement network is used to obtain a deep enhanced feature map, including:

[0090] Perform layer normalization on the feature map X to obtain the layer normalized feature map X norm ;

[0091] The feature map X norm Perform 3*3 depth-separable convolution processing to obtain feature map X1;

[0092] For feature map X norm Add it to the feature map X1 to get the feature map X ’ 1;

[0093] The feature map X ’ 1 Perform 1×1 convolution to obtain feature map X o ;

[0094] For feature map X o Perform 7×7 convolution to obtain feature map X a ; and the feature map X o Perform 3*3 depth separable convolution and upsampling to obtain feature maps

[0095] The feature map Perform subtraction and multiplication respectively to obtain the feature map and

[0096] For feature maps and Perform 3*3 depth separable convolution processing respectively, and convert the feature map and The 3*3 depth-separable convolution processing results are spliced ​​to obtain the feature map

[0097] The feature map Perform 1×1 convolution to obtain the feature map

[0098] For feature maps and feature map X a Perform splicing to obtain the preliminary enhanced feature map X en ;

[0099] The initial enhanced feature map X en Perform 1×1 convolution processing and add the result of 1×1 convolution processing to feature map X to obtain feature map X ’ en ;

[0100] For feature map X ’ en Perform layer normalization to obtain the layer normalized feature map X ’ norm ;

[0101] The feature map X ’ norm Perform 1×1 convolution to obtain the feature map

[0102] For feature maps Perform 3*3 depth separable convolution processing to obtain feature maps

[0103] The feature map and feature map Perform addition processing to obtain the feature map

[0104] For feature maps Perform 1×1 convolution and combine the 1×1 convolution result with the feature map X ’ norm Perform addition processing to obtain a deep enhanced feature map

[0105] The ship segmentation network is used to obtain the remote sensing image ship segmentation results, including:

[0106] Deep Enhanced Feature Map Perform 1×1 convolution layer, upsampling layer and activation function processing to obtain the segmentation result image

[0107] The preprocessed remote sensing image is processed based on the trained remote sensing image ship segmentation model, and a high-dimensional feature map X is generated through a feature map extraction network and a 3×3 convolution combined with an activation function. Subsequently, a series of complex operations are performed on X using a feature map enhancement network, including layer normalization, depthwise separable convolution, addition, 1×1 and 7×7 convolution, upsampling, and feature concatenation, to finally obtain a deep enhanced feature map. This process significantly improves the expressiveness and discrimination of features, especially the ability to capture small targets such as ships. Finally, the deep enhanced feature map is processed by a ship segmentation network with 1×1 convolution, upsampling, and activation functions to output a high-precision segmentation result image. This series of steps ensures efficient conversion from the original remote sensing image to the ship segmentation result, achieves accurate extraction of the ship contour under complex backgrounds, and greatly improves the accuracy and robustness of segmentation.

[0108] Embodiment 2;

[0109] like Figure 4 As shown, this embodiment provides a remote sensing image ship segmentation system, including:

[0110] Preprocessing module: used to obtain the remote sensing image to be processed and perform preprocessing to obtain the preprocessed remote sensing image;

[0111] Ship segmentation module: Train the remote sensing image ship segmentation model, and use the trained remote sensing image ship segmentation model to process the pre-processed remote sensing image to obtain the corresponding remote sensing image ship segmentation result;

[0112] The processing of the pre-processed remote sensing image includes: obtaining a feature map using a feature map extraction network, obtaining a depth enhanced feature map using a feature map enhancement network, and obtaining a remote sensing image ship segmentation result using a ship segmentation network.

[0113] This embodiment provides a remote sensing image ship segmentation system, which includes a preprocessing module and a ship segmentation module. The preprocessing module is responsible for acquiring and preprocessing remote sensing images, while the ship segmentation module processes the preprocessed images through a pre-trained model, using steps such as feature map extraction, feature map enhancement, and ship segmentation network, and finally obtains a high-precision remote sensing image ship segmentation result.

[0114] Embodiment 3;

[0115] like Figure 5 As shown, this embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the remote sensing image ship segmentation method in the above embodiment is implemented.

[0116] Embodiment 4;

[0117] This embodiment provides a computer-readable storage medium, wherein the storage medium stores a computer program. When the computer program is executed by a processor, the remote sensing image ship segmentation method in the above embodiment is implemented.

[0118] In the above embodiments provided in the present application, it should be understood that the disclosed methods, systems, devices and media can be implemented in other ways. The above described methods, systems, devices and media embodiments are only schematic. For example, the division of modules or units is only a logical function division, and there may be other division methods in actual implementation. Each functional unit can be integrated into a processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0119] The units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0120] The computer program includes computer program code, which may be in source code form, object code form, executable file or some intermediate form, etc. Computer readable media may include: any entity or device capable of carrying computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory, random access memory, electric carrier signal, telecommunication signal and software distribution medium, etc.

[0121] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application 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 make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A remote sensing image ship segmentation method, characterized in that: The following steps are involved: S1, obtaining a remote sensing image to be processed and preprocessing it to obtain a preprocessed remote sensing image; S2, training a remote sensing image ship segmentation model, and using the trained remote sensing image ship segmentation model to process the pre-processed remote sensing image to obtain a corresponding remote sensing image ship segmentation result; The processing of the pre-processed remote sensing image includes: obtaining a feature map using a feature map extraction network, obtaining a depth enhanced feature map using a feature map enhancement network, and obtaining a remote sensing image ship segmentation result using a ship segmentation network.

2. A remote sensing image ship segmentation method according to claim 1, characterized in that: In S2, a composite loss function L is used to train the remote sensing image ship segmentation model; the composite loss function L is expressed as: L=λ1L a +λ2L b Among them, λ1 and λ2 are weight coefficients, L a is the cross entropy loss function, L b is the Dice coefficient loss function; P true (i) is the image The true label of the i-th pixel, Predict probabilities for the model.

3. The method for ship segmentation in remote sensing images according to claim 1, characterized in that: In S2, the remote sensing image ship segmentation model training is performed using the Adam optimizer to update parameters; the parameter update rule of the Adam optimizer is: Among them, θ t+1 represents the updated parameters, θ t represents the current parameter, η represents the learning rate, m t and v t denote the estimates of the first-order moment and the second-order moment, respectively, and ε denotes a constant to prevent division by zero.

4. A remote sensing image ship segmentation method according to claim 3, characterized in that: The Adam optimizer also combines the cosine annealing strategy to adjust the learning rate; Among them, η t Represents the learning rate when the training step is t, T max represents the total number of training steps, and η0 represents the initial learning rate.

5. The method for segmenting ships in remote sensing images according to claim 1, characterized in that: In S2, the feature map is obtained by using a feature map extraction network, including: The preprocessed remote sensing image I is processed by 3*3 convolution and activation function to obtain the feature map X; wherein the size of the feature map X is H×W×C, H represents height, W represents width, and C represents the number of channels.

6. The method for ship segmentation in remote sensing images according to claim 1, characterized in that: In S2, a feature map enhancement network is used to obtain a depth enhanced feature map, including: Perform layer normalization on the feature map X to obtain the layer normalized feature map X norm ; The feature map X norm Perform 3*3 depth-separable convolution processing to obtain feature map X1; For feature map X norm Add it to the feature map X1 to get the feature map X ’ 1; The feature map X ’ 1 Perform 1×1 convolution to obtain feature map X o ; For feature map X o Perform 7×7 convolution to obtain feature map X a ; and the feature map X o Perform 3*3 depth separable convolution and upsampling to obtain feature maps The feature map Perform subtraction and multiplication respectively to obtain the feature map and For feature maps and Perform 3*3 depth separable convolution processing respectively, and convert the feature map and The 3*3 depth-separable convolution processing results are spliced ​​to obtain the feature map The feature map Perform 1×1 convolution to obtain the feature map For feature maps and feature map X a Perform splicing to obtain the preliminary enhanced feature map X en ; The initial enhanced feature map X en Perform 1×1 convolution processing and add the result of 1×1 convolution processing to feature map X to obtain feature map X ’ en ; For feature map X ’ en Perform layer normalization to obtain the layer normalized feature map X ’ norm ; The feature map X ’ norm Perform 1×1 convolution to obtain the feature map For feature maps Perform 3*3 depth separable convolution processing to obtain feature maps The feature map and feature map Perform addition processing to obtain the feature map For feature maps Perform 1×1 convolution and combine the 1×1 convolution result with the feature map X ’ norm Perform addition processing to obtain a deep enhanced feature map 7. The method for ship segmentation in remote sensing images according to claim 1, characterized in that: In S2, obtaining a remote sensing image ship segmentation result using a ship segmentation network includes: Deep Enhanced Feature Map Perform 1×1 convolution layer, upsampling layer and activation function processing to obtain the segmentation result image 8. A remote sensing image ship segmentation system, characterized in that: include: Preprocessing module: used to obtain the remote sensing image to be processed and perform preprocessing to obtain the preprocessed remote sensing image; Ship segmentation module: Train the remote sensing image ship segmentation model, and use the trained remote sensing image ship segmentation model to process the pre-processed remote sensing image to obtain the corresponding remote sensing image ship segmentation results; The processing of the pre-processed remote sensing image includes: obtaining a feature map using a feature map extraction network, obtaining a depth enhanced feature map using a feature map enhancement network, and obtaining a remote sensing image ship segmentation result using a ship segmentation network.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the remote sensing image ship segmentation method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the remote sensing image ship segmentation method according to any one of claims 1 to 7 is implemented.

Images