Ship target detection method based on attention-enhanced feature pyramid network

Abstract

The invention discloses a ship target detection method based on an attention-enhanced feature pyramid network, and belongs to the technical field of remote sensing image processing. The method comprises the following steps: constructing an attention enhancement module; building an attention enhancement feature pyramid network based on an attention enhancement module and a corresponding ship target detection model; then ship target detection task training is carried out on the ship target detection model; and finally, performing ship target detection on a to-be-detected remote sensing image by using the trained ship target detection model. According to the method, the attention mechanism is fully utilized to capture global semantic information association more fully, fusion features have more complete multi-scale feature information, the model can extract data features more comprehensively on a remote sensing image ship target detection task, the image content is understood, and the target detection efficiency is improved. The method can achieve the accurate recognition and positioning of the ship target in a complex ocean background, and facilitates the improvement of the accuracy and robustness of remote sensing image ship target detection.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing image processing, and in particular to a ship target detection method based on an attention enhanced feature pyramid network. Background Art

[0002] In the field of modern navy and maritime monitoring, ship target detection in remote sensing images is a core technology, which is of great significance for maritime safety, shipping monitoring and marine resource management. Traditional ship detection methods mainly rely on manual interpretation and traditional image processing techniques. These methods are often inefficient when processing large-scale high-resolution remote sensing images and are difficult to adapt to complex and changing marine environments. In recent years, the rapid development of deep learning technology has provided new solutions for ship target detection. Especially in feature extraction and target classification, methods based on convolutional neural networks (CNNs) have achieved remarkable results. However, due to the size diversity and background complexity of ship targets in remote sensing images, traditional single-scale or simple cascade feature fusion networks still face certain limitations. The simple fusion of multi-scale features extracted by the feature extraction network has the problem of missing feature information, which leads to insufficient extraction of key feature information of ships, affecting the detection accuracy of the final ship target detection model.

[0003] Therefore, it is urgent to use appropriate technology to supplement the complete feature information before fusion to the features after fusion by learning the dependency relationship between the features before and after fusion, so as to improve the integrity of feature information, further enhance the representation ability of features, and reduce information loss; it can enable the model to accurately identify and locate ship targets in complex ocean backgrounds, significantly improve the accuracy and robustness of detection, so as to realize accurate ship target detection task identification, provide new perspectives and possibilities for the development of ship target detection technology, and lay a solid foundation for the automated analysis and practical application of marine remote sensing images. Summary of the invention

[0004] In view of this, the present invention provides a ship target detection method and system based on an attention enhanced feature pyramid network that can solve the above-mentioned technical problems, which is conducive to improving the accuracy and robustness of ship target detection in remote sensing images.

[0005] To achieve the above object, the technical solution adopted by the present invention is:

[0006] In a first aspect, an embodiment of the present invention provides a method for detecting ship targets based on an attention-enhanced feature pyramid network, the method comprising the following steps:

[0007] S1, build attention enhancement module;

[0008] S2. Build an attention enhancement feature pyramid network in combination with the attention enhancement module;

[0009] S3. Building a ship target detection model based on the attention enhanced feature pyramid network;

[0010] S4. Training the ship target detection model; using the trained ship target detection model to perform ship target detection on the remote sensing image to be measured.

[0011] Furthermore, in the step S1, the attention enhancement module constructed includes: a feature splicing module, a layer normalization module, a multi-head attention module and a multi-layer perceptron module;

[0012] The attention enhancement module receives the features before and after fusion as input, and supplements the complete information before fusion to the features after fusion by learning the dependency between the features before and after fusion.

[0013] Furthermore, in the multi-head attention module, each attention head performs attention calculation independently, and multiple attention heads are processed in parallel.

[0014] Furthermore, in step S2, the attention-enhanced feature pyramid network constructed receives a multi-scale feature map as input, performs a preliminary fusion on the initial multi-scale feature map to obtain a preliminary fused feature map, and then combines the preliminary fused feature map with the initial multi-scale feature map. Figure 1 The features are sent to the attention enhancement module at the same time, and feature enhancement based on the attention mechanism is performed between the feature maps of the corresponding levels to obtain a fused feature map with feature enhancement.

[0015] Furthermore, in step S3, the constructed ship target detection model includes: an image preprocessing network, a feature extraction network, an attention enhancement feature pyramid network and a detection head connected in sequence, wherein:

[0016] The image preprocessing network receives remote sensing ship images as input and performs image transformation operations. The preprocessed image data is sent to the feature extraction network for multi-scale feature extraction. The obtained multi-scale features are sent to the attention enhancement feature pyramid network for feature enhancement operations to obtain enhanced fused feature maps. The detection head receives the enhanced fused feature maps to locate and classify ship targets.

[0017] Furthermore, the image transformation operations of the image preprocessing network include: mixup strategy, image rotation, image stitching, noise injection, image perturbation, image sharpening and blurring;

[0018] The detection head includes a regression network and a classification network. The detection head uses the regression network to regress the anchor boxes in the feature maps of each level based on the predefined anchor boxes to accurately determine the position of the aircraft detection box. At the same time, the classification network performs classification operations on all feature points in the feature map and evaluates the category probability of each detection box.

[0019] Furthermore, in step S4, the model is trained for the remote sensing image ship target detection task using a cross entropy loss function and a mean square error loss function.

[0020] In a second aspect, an embodiment of the present invention further provides a ship target detection system based on an attention-enhanced feature pyramid network, which applies the above-mentioned ship target detection method based on an attention-enhanced feature pyramid network to achieve accurate ship target detection task recognition, and the system includes:

[0021] A target detection model building module is used to construct an attention enhancement module, build an attention enhancement feature pyramid network in combination with the attention enhancement module, and build a ship target detection model based on the attention enhancement feature pyramid network;

[0022] A target detection model training module, used for training the ship target detection model for ship target detection tasks;

[0023] The target detection model application module is used to use the trained ship target detection model to perform ship target detection on the remote sensing image to be tested.

[0024] Compared with the prior art, the present invention has at least the following beneficial effects:

[0025] 1. The present invention provides a ship target detection method based on an attention enhanced feature pyramid network, in which an attention enhancement module is built, and an attention enhanced feature pyramid network optimized based on the attention enhancement module and its corresponding ship target detection model are constructed. The ship target detection model is used to perform remote sensing image ship target detection, and accurate remote sensing image ship target detection can be achieved without human intervention.

[0026] 2. The ship target detection model based on the attention enhanced feature pyramid network proposed in the present invention makes full use of the characteristic that the attention mechanism can more fully capture the global semantic information association, so that the fusion feature has more complete multi-scale feature information, so that the model can more comprehensively extract data features and understand the image content in the remote sensing image ship target detection task, and can realize accurate recognition and positioning of ship targets in complex ocean backgrounds, significantly improving the detection accuracy and robustness, thereby realizing accurate remote sensing image ship target detection.

[0027] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description and the accompanying drawings.

[0028] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0030] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0031] Figure 1 A flow chart of a ship target detection method based on an attention enhanced feature pyramid network provided in an embodiment of the present invention.

[0032] Figure 2 A schematic diagram of the multi-head attention structure provided for an embodiment of the present invention.

[0033] Figure 3 A schematic diagram of an attention enhancement module provided in an embodiment of the present invention.

[0034] Figure 4 Schematic diagram of the attention-enhanced feature pyramid network provided by an embodiment of the present invention.

[0035] Figure 5 A schematic diagram of a ship target detection model based on an attention-enhanced feature pyramid network provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0036] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments.

[0037] In the description of the present invention, it should be noted that: in some processes described in the specification and drawings of this application, multiple operations appearing in a specific order are included, but it should be clearly understood that these operations may not be performed in the order in which they appear in this document or may be performed in parallel. In addition, various serial numbers are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0038] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. 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.

[0039] See also Figure 1 As shown, the present invention provides a ship target detection method based on an attention enhanced feature pyramid network, and the method mainly includes the following steps:

[0040] S1, build attention enhancement module;

[0041] S2. Combine the attention enhancement module to build an attention enhancement feature pyramid network;

[0042] S3. Build a ship target detection model based on attention-enhanced feature pyramid network;

[0043] S4. Train the ship target detection model; use the trained ship target detection model to perform ship target detection on the remote sensing image to be tested.

[0044] The specific implementation methods of the method of the present invention are described in detail below:

[0045] In an embodiment of the present invention, an efficient attention enhancement module is designed, an attention enhancement feature pyramid network and a corresponding ship target detection model are constructed, and the complete feature information before fusion is supplemented to the fused features by learning the dependency relationship between the features before and after fusion, so as to improve the integrity of the feature information, further enhance the characterization ability of the features, and reduce information loss; the model can be enabled to accurately identify and locate ship targets in complex ocean backgrounds, significantly improve the accuracy and robustness of detection, so as to realize accurate ship target detection task recognition. The specific implementation methods of the present invention are as follows:

[0046] 1. Build an attention enhancement module:

[0047] In this embodiment, the Attention Enhancement Module (AEM) receives the features before and after fusion as input, and supplements the complete information before fusion to the features after fusion by learning the dependency relationship between the features before and after fusion, so as to improve the integrity of the feature information and further enhance the representation ability of the features. Specifically, let the features before fusion be C i , where i is the feature at different levels, and its corresponding fused feature is set to P i , since the fusion process will change the number of channels of the feature, C i With P i The channel dimensions of are different. Therefore, Figure 3 As shown, in order to better enhance the features, we first need to i With P i Perform the splicing operation on the channel dimension to obtain preliminary splicing features. The specific operation can be formalized as follows:

[0048] CP i =Concat(C i ,P i )

[0049] Among them, Concat represents the concatenation operation. Then, CP i Transformed into image embedding vector X through deformation operation i , to become a data format suitable for attention mechanism processing, as shown in the following formula:

[0050]

[0051] Among them, ReshapeA means to transform the three-dimensional feature map CP i Convert to image embedding vector X i Then, the embedding vector is normalized by the layer normalization function, that is, The normalized vector is sent to the multi-head attention mechanism for attention calculation. For ease of description, in the multi-head attention mechanism calculation, X i Unified into X E form, Expressed as

[0052] like Figure 2 As shown in the figure, in the multi-head attention mechanism, each "head" performs an independent attention calculation process, and multiple heads can be processed simultaneously. In this object detection model, the number of heads is set to 8. This parallel computing method enables the model to extract input features from different angles, and finally integrate these diverse perspective information to generate richer and more comprehensive attention features.

[0053] For the jth attention head, the input vector X E Through three independent fully connected layers (linear transformation), they are mapped to query (Query, Q), key (Key, K) and value (Value, V). The specific calculation formula for this mapping is as follows:

[0054]

[0055] in, and Represents the independent learnable weight matrix in each attention head, and j represents different attention heads. In the attention mechanism, we first calculate the query Q j With key K j The dot product of is calculated and scaled to get the attention score. Then, these scores are normalized and then the normalized attention score is compared with the corresponding value V j Multiply to complete further calculations. The attention calculation process can be formalized as follows:

[0056]

[0057] Among them, d k Represents the dimension of the key vector. This scaling operation helps prevent the generation of too large values ​​before applying the softmax function, thereby maintaining numerical stability. The softmax function is a normalized exponential function that is used to convert attention scores into probability distributions that represent the degree of attention each output element pays to each element in the input sequence. According to the obtained probability distribution, the value vector is weighted and summed to generate the final output of the attention mechanism. This process can be viewed as extracting and combining key feature information from the input elements according to the attention weights. Finally, the outputs of all heads are concatenated and processed through a linear transformation layer to integrate the information extracted by each head. The computational formalization of the above process is as follows:

[0058] MultiHead(X E )=Concat(head1, head2,..., head h )W O

[0059] Among them, W O is a learnable weight matrix.

[0060] like Figure 3As shown in the figure, after the multi-head attention is calculated, its output is added to the original input through residual connection and layer normalization is performed. Then, the normalized result is input into the multi-layer perceptron (MLP), and residual connection and layer normalization are performed again. The expression of this process is as follows.

[0061]

[0062] AttentionOutput=LayerNorm(Output+MLP(Output))

[0063] Finally, the obtained features need to be converted back to the three-dimensional feature map form through the deformation operation ReshapeB. ReshapeB is the inverse process of ReshapeA, and then the dimension is adjusted through a 1×1 convolution operation. The specific operation can be formalized as follows:

[0064] A=Conv 1×1 (ReshapeB(AttentionOutput))

[0065] Among them, A represents the output of the attention enhancement module. So far, the attention enhancement calculation process has been described and can be summarized as follows:

[0066] A i =AAEM(C i , P i )

[0067] AEM is the complete attention enhancement calculation process. i is the output feature of the attention enhancement module, and its shape is similar to P i The same, i represents features at different levels.

[0068] 2. Build an attention-enhanced feature pyramid network:

[0069] In a specific embodiment, the Attention Enhancement Feature Pyramid Network (AE-FPN) receives a multi-scale feature map as input. Figure 4 As shown, let the input multi-scale feature map be C iIn this target detection, there are 4 feature maps of different scales, which are recorded as C1, C2, C3, and C4. The scale of the C3 feature map is twice that of C4, the scale of the C2 feature map is twice that of C3, and the scale of the C1 feature map is twice that of C2. The initial feature map needs to be initially fused. Specifically, these four feature maps of different scales are adjusted through a 1×1 convolution operation to obtain feature maps F1, F2, F3, and F4 respectively. The specific operation can be formalized as the following formula:

[0070] F i =Conv 1×1 (C i )

[0071] The obtained feature maps F1, F2, F3, and F4 will be used for feature fusion between adjacent feature maps. First, the smallest scale feature map F4 is upsampled twice, and its scale after upsampling is the same as F3. It is then added to F3, and then a 3×3 convolution operation is performed to obtain the initial fused feature map P3. Similarly, after F3 is upsampled twice, it is added to the feature map F2, and then a 3×3 convolution is performed to obtain the initial fused feature map P2. After F2 is upsampled twice, it is added to the feature map F1, and then a 3×3 convolution is performed to obtain the initial fused feature map P1. The above operations can be expressed as the following formula:

[0072] P3=Conv 3×3 (Add(F3+Upsampling(F4)))

[0073] P2=Conv 3×3 (Add(F2+Upsampling(F3)))

[0074] P1=Conv 3×3 (Add(F1+Upsampling(F2)))

[0075] Among them, Upsampling represents the upsampling operation, and Add represents the feature map addition operation. For feature map P4, it is obtained by directly performing a 3×3 convolution operation on F4, which can be expressed as the following formula:

[0076] P4=Conv 3×3 (F4)

[0077] Therefore, the obtained preliminary fusion feature maps include P1, P2, P3, and P4. Subsequently, the preliminary fusion feature maps and the initial multi-scale feature maps C1, C2, C3, and C4 are sent to the attention enhancement module, and further feature enhancement based on the attention mechanism is performed between the feature maps of the corresponding levels, as shown in the attention enhancement calculation formula.

[0078] Finally, the calculation in the complete attention-enhanced feature pyramid network can be expressed as AEFPN, so the complete calculation operation can be expressed by the following formula:

[0079] A1, A2, A3, A4 = AEFPN (C1, C2, C3, C4)

[0080] A1, A2, A3, and A4 are the feature map outputs of the attention-enhanced feature pyramid network.

[0081] 3. Build a ship target detection model:

[0082] like Figure 5 As shown in the figure, the complete ship target detection model mainly consists of image preprocessing, feature extraction network, attention enhanced feature pyramid network, and detection head.

[0083] First, the image preprocessing network receives the remote sensing ship image as input and performs a series of image transformation operations, including mixup strategy (randomly selecting two images and mixing them in a certain proportion to generate a new image), image rotation, image stitching, noise injection, image perturbation (randomly changing the brightness, contrast, saturation and hue of the image), image sharpening and blurring, etc. Through these operations, the data is enhanced to obtain enhanced image data I.

[0084] Then, the preprocessed image data is sent to the feature extraction network for multi-scale feature extraction. This step can select a common feature extraction network, such as the ResNet network. Using BackboneNet to represent the feature extraction network, the multi-scale feature extraction can be expressed as:

[0085] C1, C2, C3, C4=BackboneNet(I)

[0086] The obtained multi-scale features C1, C2, C3, and C4 will be sent to the attention-enhanced feature pyramid network to perform feature enhancement operations to obtain enhanced fused feature maps A1, A2, A3, and A4. The enhanced feature maps have their own scale sizes to adapt to ship targets of different scales.

[0087] Finally, the detection head receives the multi-scale enhanced feature map to locate and classify the ship target. Specifically, based on the predefined anchor frame, the regression network in the detection head regresses the anchor frame on the feature map at each level to obtain an accurate ship detection frame. The classification network classifies all feature points on the feature map and calculates the classification probability of the category to which each detection frame belongs. Finally, the low-confidence detection frame is removed through confidence filtering, and the non-maximum suppression (NMS) technology is used to remove redundant detection frames. The two are combined to output the final target positioning and classification results. The specific operation can be expressed as follows:

[0088] y=Score(ClsNet(A1,A2,A3,A4))

[0089] bbox=NMS(RegNet(A1,A2,A3,A4))

[0090] Among them, RegNet is a regression network and ClsNet is a classification network. Both are composed of a series of convolutional layers. The classification network has a Sigmoid function at the end to obtain a probability value of 0-1. Score represents confidence filtering, which discards results with probabilities lower than the confidence threshold, and NMS filters overlapping prediction boxes. Using DetHead to represent the detection head, the complete prediction calculation can be expressed as:

[0091] y,bbox=DetHead(A1,A2,A3,A4)

[0092] Among them, y corresponds to bbox one by one, bbox is the predicted detection box, and y is the classification probability of the detection box.

[0093] 4. Training and testing the model on remote sensing image ship data:

[0094] In a specific embodiment, after constructing a ship target detection model based on an attention-enhanced feature pyramid network, a cross-entropy loss function (Cross-entropy Loss, L CE ) and the mean square error loss function (Mean SquaredError, L MSE ) The model is trained for the task of ship target detection in remote sensing images. Finally, the loss function used by the model is the sum of the two losses, which can be specifically expressed as:

[0095] L CE =L CE (y,y label )+L MSE (bbox,bboX tabel )

[0096] Among them, bbox label With y label is the true value of the label. When the model is trained until the loss function no longer decreases significantly, it means that the model has reached a stable state. At this time, the training is completed, and the trained ship target detection model based on the attention-enhanced feature pyramid network is obtained.

[0097] Subsequently, the trained ship target detection model based on the attention-enhanced feature pyramid network is used to perform ship target detection tests on the test images. The process can be formally expressed as:

[0098]

[0099] in, (Attention-Enhancement-Feature-Pyramid based DetectionModel) represents the trained ship target detection model based on the attention-enhanced feature pyramid network. Test Represents the remote sensing image to be tested, y test and bbox test Indicates the ship target detection result obtained by the test.

[0100] From the description of the above embodiments, those skilled in the art can know that the present invention provides a ship target detection method based on an attention-enhanced feature pyramid network, wherein an attention-enhanced module is built, and an attention-enhanced feature pyramid network optimized based on the attention-enhanced module and its corresponding ship target detection model are constructed. The ship target detection model is used to detect ship targets in remote sensing images without manual intervention, and accurate remote sensing image ship target detection can be achieved. In the method of the present invention, by learning the dependency between the features before and after fusion, the complete feature information before fusion is supplemented to the features after fusion to improve the integrity of the feature information, further enhance the representation ability of the features, and reduce information loss. The model can realize accurate recognition and positioning of ship targets in complex ocean backgrounds, and significantly improve the accuracy and robustness of detection.

[0101] The method of the present invention provides a more reliable and efficient solution for the field of ship target detection, provides a new perspective and possibility for the development of ship target detection technology, and helps to lay a solid foundation for the automated analysis and practical application of marine remote sensing images.

[0102] Furthermore, the present invention also provides a ship target detection system based on an attention-enhanced feature pyramid network, which is applied to a ship target detection method based on an attention-enhanced feature pyramid network in the above embodiment to improve the performance of remote sensing image target detection and facilitate accurate ship target detection task recognition; the system includes:

[0103] A target detection model building module is used to construct an attention enhancement module, build an attention enhancement feature pyramid network in combination with the attention enhancement module, and build a ship target detection model based on the attention enhancement feature pyramid network;

[0104] A target detection model training module, used for training the ship target detection model for ship target detection tasks;

[0105] The target detection model application module is used to use the trained ship target detection model to perform ship target detection on the remote sensing image to be tested.

[0106] The implementation principle and technical effects of the system provided in the embodiment of the present invention are the same as those of the aforementioned method embodiment. For the sake of brief description, for matters not mentioned in the system embodiment, reference can be made to the corresponding contents in the aforementioned method embodiment, which will not be repeated here.

[0107] In addition, an embodiment of the present invention also provides a storage medium on which one or more programs readable by a computing device are stored, and the one or more programs include instructions. When the instructions are executed by the computing device, the computing device executes a ship target detection method based on an attention enhanced feature pyramid network in the above-mentioned embodiment.

[0108] In the embodiment of the present invention, the storage medium may be, for example, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination thereof. More specific examples of storage media (a non-exhaustive list) include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, and any suitable combination thereof.

[0109] Those skilled in the art should understand that the embodiments of the present invention can be provided as methods, systems or computer program products, etc. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects.

[0110] It should be noted that the word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention can be implemented by means of hardware comprising several distinct components, and by means of a suitably programmed computer.

[0111] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0112] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A ship target detection method based on attention enhanced feature pyramid network, characterized in that: The method comprises the following steps: S1, build attention enhancement module; S2. Build an attention enhancement feature pyramid network in combination with the attention enhancement module; S3. Building a ship target detection model based on the attention enhanced feature pyramid network; S4. Training the ship target detection model; using the trained ship target detection model to perform ship target detection on the remote sensing image to be measured.

2. According to claim 1, a ship target detection method based on attention enhanced feature pyramid network is characterized in that: In the step S1, the attention enhancement module constructed includes: a feature splicing module, a layer normalization module, a multi-head attention module and a multi-layer perceptron module; The attention enhancement module receives the features before and after fusion as input, and supplements the complete information before fusion to the features after fusion by learning the dependency between the features before and after fusion.

3. According to claim 2, a ship target detection method based on attention enhanced feature pyramid network is characterized in that: In the multi-head attention module, each attention head performs attention calculation independently, and multiple attention heads are processed in parallel.

4. According to claim 3, a ship target detection method based on attention enhanced feature pyramid network is characterized in that: In step S2, the attention enhancement feature pyramid network constructed receives a multi-scale feature map as input, performs a preliminary fusion on the initial multi-scale feature map to obtain a preliminary fused feature map, and then sends the preliminary fused feature map and the initial multi-scale feature map together to the attention enhancement module, performs feature enhancement based on the attention mechanism between the feature maps of corresponding levels, and obtains a feature enhanced fused feature map.

5. According to claim 4, a ship target detection method based on attention enhanced feature pyramid network is characterized in that: In step S3, the constructed ship target detection model includes: an image preprocessing network, a feature extraction network, an attention enhancement feature pyramid network and a detection head connected in sequence, wherein: The image preprocessing network receives remote sensing ship images as input and performs image transformation operations. The preprocessed image data is sent to the feature extraction network for multi-scale feature extraction. The obtained multi-scale features are sent to the attention enhancement feature pyramid network for feature enhancement operations to obtain enhanced fused feature maps. The detection head receives the enhanced fused feature maps to locate and classify ship targets.

6. According to claim 5, a ship target detection method based on attention enhanced feature pyramid network is characterized in that: The image transformation operations of the image preprocessing network include: mixup strategy, image rotation, image stitching, noise injection, image perturbation, image sharpening and blurring; The detection head includes a regression network and a classification network. The detection head uses the regression network to regress the anchor boxes in the feature maps of each level based on the predefined anchor boxes to accurately determine the position of the aircraft detection box. At the same time, the classification network performs classification operations on all feature points in the feature map and evaluates the category probability of each detection box.

7. The ship target detection method based on attention enhanced feature pyramid network according to claim 1 is characterized in that: In step S4, the model is trained for the remote sensing image ship target detection task using a cross entropy loss function and a mean square error loss function.

8. A ship target detection system based on attention-enhanced feature pyramid network, characterized in that: The ship target detection method based on the attention enhanced feature pyramid network as described in any one of claims 1 to 7 is applied to realize accurate ship target detection task recognition, and the system comprises: A target detection model building module is used to construct an attention enhancement module, build an attention enhancement feature pyramid network in combination with the attention enhancement module, and build a ship target detection model based on the attention enhancement feature pyramid network; A target detection model training module, used for training the ship target detection model for ship target detection tasks; The target detection model application module is used to use the trained ship target detection model to perform ship target detection on the remote sensing image to be tested.

Images