A ship detection method, device and processing equipment based on remote sensing images

Abstract

The application provides a ship detection method and device based on remote sensing images and processing equipment. By optimizing the existing YOLOv8 model twice, the ship detection based on satellite remote sensing images can achieve faster convergence speed and higher detection accuracy, thereby meeting the high-quality ship detection demand in actual situations, and providing better data support for application services such as fishing boat navigation, fishing boat distribution monitoring, fishing boat distribution management, etc.

Description

Technical Field

[0001] This application relates to the field of target detection, specifically to a method, apparatus, and processing equipment for ship detection based on remote sensing images. Background Technology

[0002] In the development of satellite remote sensing technology, it is necessary to consider the diverse application scenarios under actual conditions, and target detection of fishing vessels at sea has its application prospects.

[0003] Specifically, using remote sensing imagery for target detection of fishing vessels helps provide precise navigation support for the safe operation of fishing vessels, especially in severe weather conditions. At the same time, monitoring the distribution of fishing vessels also helps in the protection of fishery resources and avoids intensive fishing in local areas.

[0004] With the development of deep learning, object detection algorithms based on convolutional neural networks are playing an increasingly important role in ship detection. They are mainly divided into two-stage detection algorithms and single-stage detection algorithms.

[0005] Two-stage object detection algorithms, including Faster R-CNN, Mask R-CNN, Cascade R-CNN, and R-FC, are more suitable for complex scenes and applications requiring high precision. They typically require more computing resources and are slightly slower.

[0006] Single-stage object detection algorithms such as YOLO, SSD, RetinaNet, CenterNet, EfficientDet, and DETR offer faster detection speeds with limited computing resources, making them suitable for scenarios with high real-time requirements.

[0007] Regarding the YOLO algorithm or YOLO model, the inventors of this application found that the YOLOv8 model, which is an improvement on YOLOv5, is more suitable for ship detection scenarios. However, its residual network architecture for target detection can only extract image features containing three spectral features: RGB (red, green, blue). This leads to false detections when ship targets with similar spectral features and indistinct edge features are easily detected. Summary of the Invention

[0008] This application provides a method, apparatus, and processing equipment for ship detection based on remote sensing imagery. By performing secondary optimization on the existing YOLOv8 model, the ship detection based on satellite remote sensing imagery achieves faster convergence speed and higher detection accuracy, thereby meeting the high-quality ship detection requirements in actual situations. This can provide better data support for application services such as fishing vessel navigation, fishing vessel distribution monitoring, and fishing vessel distribution management.

[0009] Firstly, this application provides a ship detection method based on remote sensing imagery, the method comprising: Acquire the current remote sensing image to be detected; The current remote sensing image is fed into a pre-configured ship detection model, which is configured based on an improved YOLOv8 model. The ship detection model is used to perform single-stage target detection processing of ship objects based on the remote sensing image input to the model. In the specific working process of the ship detection model, after extracting the edge features and RGB spectral features of the remote sensing image input to the model, specific detection processing is carried out based on the four-channel image features composed of edge features and RGB spectral features. Extract the ship detection results from the current remote sensing image output by the ship detection model.

[0010] Secondly, this application provides a ship detection device based on remote sensing imagery, the device comprising: The acquisition unit is used to acquire the current remote sensing image to be detected; The detection unit is used to send the current remote sensing image into the pre-configured ship detection model. The ship detection model is configured based on the improved YOLOv8 model. The ship detection model is used to perform single-stage target detection processing of ship objects based on the remote sensing image input to the model. In the specific working process of the ship detection model, after extracting the edge features and RGB spectral features of the remote sensing image input to the model, specific detection processing is carried out based on the four-channel image features composed of edge features and RGB spectral features. The extraction unit is used to extract the ship detection results from the current remote sensing image output by the ship detection model.

[0011] Thirdly, this application provides a processing device, including a processor and a memory, wherein a computer program is stored in the memory, and the processor executes the method provided in the first aspect of this application when it invokes the computer program in the memory.

[0012] Fourthly, this application provides a computer-readable storage medium storing a plurality of instructions adapted for loading by a processor to execute the method provided in the first aspect of this application.

[0013] From the above, it can be concluded that this application has the following beneficial effects: For the purpose of vessel detection, this application optimizes the existing YOLOv8 model to achieve faster convergence and higher detection accuracy for vessel detection based on satellite remote sensing imagery. This can meet the high-quality vessel detection requirements in real-world situations and provide better data support for applications such as fishing vessel navigation, fishing vessel distribution detection, and fishing vessel distribution management. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a schematic diagram of a ship detection method based on remote sensing imagery according to this application; Figure 2 This is a schematic diagram of a network architecture for the improved YOLOv8 model in this application; Figure 3 This is a schematic diagram of a ship detection device based on remote sensing imagery according to this application; Figure 4 This is a schematic diagram of one type of processing equipment used in this application. Detailed Implementation

[0016] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0017] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or modules is not necessarily limited to those explicitly listed, but may include other steps or modules not explicitly listed or inherent to such processes, methods, products, or devices. The naming or numbering of steps appearing in this application does not imply that the steps in the method flow must be performed in the chronological / logical order indicated by the naming or numbering. The execution order of named or numbered process steps can be changed according to the desired technical purpose, as long as the same or similar technical effect is achieved.

[0018] The module division described in this application is a logical division. In practical applications, there may be other division methods. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the shown or discussed mutual coupling, direct coupling, or communication connections may be through interfaces, and the indirect coupling or communication connections between modules may be electrical or other similar forms, none of which are limited in this application. Moreover, the modules or sub-modules described as separate components may or may not be physically separated, may or may not be physical modules, or may be distributed across multiple circuit modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution in this application.

[0019] Before introducing the ship detection method based on remote sensing imagery provided in this application, we will first introduce the background information involved in this application.

[0020] The remote sensing image-based ship detection method, apparatus, and computer-readable storage medium provided in this application can be applied to processing equipment to perform secondary optimization on the existing YOLOv8 model, thereby enabling faster convergence speed and higher detection accuracy for ship detection based on satellite remote sensing imagery. This can meet the high-quality ship detection requirements in actual situations and provide better data support for applications such as fishing vessel navigation, fishing vessel distribution detection, and fishing vessel distribution management.

[0021] The ship detection method based on remote sensing imagery mentioned in this application can be implemented by a ship detection device based on remote sensing imagery, or by different types of processing devices such as servers, physical hosts, or user equipment (UE) that integrate the ship detection device based on remote sensing imagery. The ship detection device based on remote sensing imagery can be implemented in hardware or software. The UE can be a terminal device such as a smartphone, tablet, laptop, desktop computer, or personal digital assistant (PDA). The processing devices can be configured in a device cluster.

[0022] It is understandable that the solution in this application is usually based on existing remote sensing images or data processing that has already been acquired. Therefore, the processing equipment that implements the ship detection method based on remote sensing images of this application or that carries the corresponding application service of the ship detection method based on remote sensing images of this application usually only needs to meet the required data processing capabilities, and its specific equipment type and equipment deployment form are quite flexible.

[0023] If the direct acquisition of existing data mentioned above is also involved, then further hardware and software adaptation configurations are obviously required for the processing equipment to enable it to acquire data. For example, if real-time acquisition of remote sensing image data is required, the corresponding remote sensing service platform can be incorporated into the equipment cluster of the processing equipment, or the processing equipment itself can be the control part of the remote sensing service platform. Alternatively, a third-party call can be used to trigger the remote sensing service platform outside the processing equipment to perform real-time acquisition of actual remote sensing images.

[0024] In addition, if there is a need to display the processing progress (including the processing results), the processing device itself can be configured with the required display screen (including touch screen) to display the specific content. Of course, the processing device can also display the specific content through an external display device or other devices with a display screen.

[0025] The following section introduces the ship detection method based on remote sensing imagery provided in this application.

[0026] First, refer to Figure 1 , Figure 1 The diagram illustrates a flowchart of the ship detection method based on remote sensing imagery provided in this application. Specifically, the ship detection method based on remote sensing imagery provided in this application may include the following steps S101 to S103: Step S101: Obtain the current remote sensing image to be detected; It is easy to understand that the acquisition and processing of the current remote sensing image to be detected (hereinafter referred to as the image to be detected) usually involves the extraction and processing of existing remote sensing images. Of course, in actual situations, the possibility of real-time remote sensing images cannot be ruled out.

[0027] It should be noted that although it refers to the current remote sensing image, it does not mean that the remote sensing image is the remote sensing image at the current time. Rather, it means the remote sensing image that needs to be detected by the scheme of this application at the current time. It can be a remote sensing image from a historical time period, which corresponds to the flexible and ever-changing application needs in actual situations.

[0028] In practice, the proposed solution can also be initiated as a related remote sensing image processing task. The current remote sensing images can be directly included in the task information, or the acquisition method or collection method can be included in the task information, thus playing an indirect role.

[0029] As an example of remote sensing image processing tasks, when a task of monitoring the distribution of fishing boats in a specific season is triggered, fishing boats in the target area can be identified by real-time remote sensing images. Then, the number and location of the identified fishing boats can be combined to provide data reference for the management of fishing boat distribution by relevant management agencies.

[0030] Furthermore, the current remote sensing image can be either a single point in time or multiple consecutive points in time, i.e., an impact sequence, which corresponds to the configuration of the processing logic of the subsequent ship detection model.

[0031] Step S102: The current remote sensing image is sent to the pre-configured ship detection model. The ship detection model is configured based on the improved YOLOv8 model. The ship detection model is used to perform single-stage target detection processing of ship objects based on the remote sensing image input to the model. In the specific working process of the ship detection model, after extracting the edge features and RGB spectral features of the remote sensing image input to the model, specific detection processing is carried out based on the four-channel image features composed of edge features and RGB spectral features. As can be easily seen, this application performs secondary optimization on the working logic of the existing or conventional YOLOv8 model, thereby obtaining an improved YOLOv8 model.

[0032] Specifically, the existing YOLOv8 model's residual network architecture for object detection can only extract image features containing three spectral features: RGB (red, green, and blue). This can lead to false detections when there are ship targets with similar spectral features and indistinct edge characteristics. To address this issue, the inventors of this application have performed a secondary optimization on the existing YOLOv8 model. By adding edge features to the YOLOv8 model and optimizing its network structure and upgrading the feature fusion mechanism, a precise feature extraction architecture has been built for the efficient use of edge features, which can avoid detection box offset or missed detection.

[0033] By constructing a new channel, which together with the original three RGB channels forms a four-channel system, the single-stage target detection processing for ship objects is advanced through the four-channel image features composed of edge features and RGB spectral features. In this process, the addition of edge detail information of the target can effectively reduce the occurrence of missed detections and false detections, thereby improving detection accuracy.

[0034] More specifically, the four-channel image features can not only extract the spectral and texture features of the target, but also provide pixel-level positioning points for the detection box through precise edge features, thereby effectively avoiding the offset of the detection box and background interference, and greatly improving the target detection performance of the algorithm.

[0035] It is worth mentioning that the target detection of vessels can not only involve the target detection of the entire vessel, but also the target detection of specific objects (objects, structures, etc.) on the vessel, such as net hooking points, hook handling areas, anchor piles, cranes, feed delivery devices, net cage traction piles, winches, etc. Specific specific objects can also be distinguished according to type indicators such as the vessel's operating mode. Among them, the vessel's operating mode can be divided into trawlers, hoisting vessels, aquaculture vessels, and purse seine vessels.

[0036] In terms of specific model architecture, the improved YOLOv8 model can include three structures: backbone network structure, neck network structure, and head network structure.

[0037] 1) Backbone network structure The backbone network structure adopts the CSPDarknet-53 architecture, which consists of 53 convolutions, including residual blocks and C2f modules. The residual blocks are used to help capture details and contextual information in the image, while the C2f modules are used to fuse feature maps of different scales to extract rich feature information and have rich gradient flow.

[0038] 2) Neck network structure The neck network structure consists of a PAN-FPN structure. The upsampled convolution is removed from the PAN structure to achieve a lightweight effect without changing the original performance. Compared with the traditional FPN structure, which uses a top-down approach to transmit deep semantic information and may lose some target localization information, the PAN structure added to the FPN structure enhances path information through the complementarity of shallow and deep information.

[0039] 3) Head Network Structure The head network structure employs a decoupled head structure, specifically using two independent branches for target classification and prediction / regression, and then each branch is further used... The convolutional layers complete the classification and localization tasks, while anchor-free detection is used to improve convergence speed and accuracy.

[0040] Among them, it can also be combined Figure 2 The diagram shown below illustrates a network architecture of the improved YOLOv8 model of this application for a more intuitive understanding. Figure 2 The paper further demonstrates the Conv structure in the YOLOv8 model, the Bottleneck structure in the C2f module, and the SPPF structure of the pooling layer.

[0041] At the same time, in order to apply the model, the corresponding model training work will also be involved in the early stage.

[0042] In this regard, as an exemplary embodiment, the method of this application may further include: Acquire remote sensing images of sample vessels and configure corresponding annotations; A ship detection model is trained based on sample ship remote sensing images.

[0043] It is understood that the remote sensing images of sample vessels can be real images, images modified from real images, or directly generated images, thus meeting diverse sample configuration requirements. During or after the acquisition of remote sensing images of sample vessels, corresponding data preprocessing may also be involved to further improve the sample size or data quality.

[0044] As an example, Mosaic data augmentation can be used to enrich the samples, thereby enhancing the generalization performance of the model.

[0045] The corresponding annotation work can be handled manually or by corresponding automated annotation tools. Such automated annotation tools need to be pre-configured with corresponding automated annotation logic / strategies. The annotation content can involve specific information such as object category, object location, and object size. The annotation format can be either XML or TXT.

[0046] After completing model initialization tasks such as cropping size, optimizer, and number of iterations, the actual model training can proceed. During model training, the following main approach can be followed: A sample remote sensing image of a ship is fed into the model for detection to perform forward propagation. Then, based on the detection results output by the model, the loss function is calculated by combining the annotations, and the model parameters are optimized according to the loss function calculation results to achieve backpropagation. After continuous iterative training, when the preset model training requirements such as training time, number of training sessions or detection accuracy are met, the model training is completed, and a ship detection model that can be put into practical use is obtained.

[0047] The specific training scheme (e.g., dividing the samples into training, validation, and test sets in an 8:1:1 ratio) and the specific loss function can obviously be either an existing scheme, a further optimization of the existing scheme, or a novel self-developed scheme. This can be flexibly configured according to actual needs.

[0048] As an example, this application may use specific metrics such as precision, recall, F1 score, and mean average precision (mAP) to evaluate model performance.

[0049] The formula for calculating Precision is as follows: , The formula for calculating recall is as follows: , Wherein, TP: label value is True, model prediction is Positive; FN: label value is False, model prediction is Negative; FP: label value is False, model prediction is Positive; TN: label value is True, model prediction is Negative.

[0050] Average Precision (AP) is calculated based on the true label and predicted probability of each category, and the corresponding Precision and Recall values ​​form the region. mAP is the average of the AP values ​​of all categories, and its value ranges from [0,1]. The higher the values ​​of these two indicators, the better the detection performance.

[0051] Step S103: Extract the ship detection results of the current remote sensing image output by the ship detection model.

[0052] Understandably, after the ship detection model completes the ship detection processing of the current remote sensing image, the corresponding ship detection results output by the ship detection model can be extracted.

[0053] In this case, the output can be generated according to preset or real-time configured output requirements to meet the diverse application needs in specific applications.

[0054] For example, the system can store the vessel's inspection results locally or remotely, display the results, output a notification that the inspection is complete, forward the results, or perform further data analysis.

[0055] As an example of the results presentation, the method of this application may also include: Based on the ship detection results, the ship distribution is displayed in a visualization interface.

[0056] Taking fishing vessel targets as an example, it can be understood that in fishing vessel distribution detection or management application scenarios, different detected vessels can be displayed or updated in a customized visualization interface, allowing users to easily view the current distribution of fishing vessels and providing good visualization data support for decision-making in fishing vessel distribution management.

[0057] In addition, the visual interface can be configured with a user-friendly and interactive page design, allowing users to zoom in, zoom out, and adjust the position of the content displayed on the interface to better view the distribution of fishing boats within the selected area.

[0058] For further data analysis, as an example, the method of this application may also include: Adaptive response processing based on ship detection results.

[0059] Understandably, the response processing involved here corresponds to the relevant data analysis in order to meet the closed-loop output of the automated system. Taking fishing vessel distribution detection as an example, if a fishing vessel is detected approaching or appearing in a dangerous sea area, such as a sea area prone to running aground, a sea area where multiple accidents have recently occurred, or a sea area where there may be swimmers due to a large number of tourists on the coast recently, an early warning can be issued to the relevant management personnel to remind them to pay attention to the situation or to take further action in a timely manner.

[0060] Understandably, the specific response handling strategy is quite flexible, and can be adapted to the actual situation in practical applications.

[0061] Thus, from Figure 1 As can be seen from the embodiments shown, for the target of vessel detection, this application optimizes the existing YOLOv8 model to achieve faster convergence speed and higher detection accuracy for vessel detection based on satellite remote sensing images. This can meet the high-quality vessel detection requirements in actual situations and provide better data support for application services such as fishing vessel navigation, fishing vessel distribution detection, and fishing vessel distribution management.

[0062] Continue with the above Figure 1 The steps of the illustrated embodiment and their possible implementation methods in practical applications are described in detail.

[0063] Regarding the additional influencing factor of edge features introduced in this application, as an exemplary embodiment, the ship detection model can extract the edge features of the remote sensing image input to the model through the pre-configured edge feature function Sobel operator to obtain the corresponding edge features.

[0064] Understandably, the Sobel operator has the advantages of being simple and fast, having smooth and continuous edges, strong noise resistance, clear edge direction information, and good robustness, which can well meet the edge feature extraction requirements of this application.

[0065] Continuing our focus on the loss function used during model training, as an example, the loss function involved in model training includes angle loss. Distance loss Shape loss IoU loss and SIoU loss.

[0066] It is understandable that these five loss types are the existing loss function types involved in the existing YOLOv8 model. The corresponding quantization formulas will be introduced in detail below.

[0067] 1) Angle loss : , in, This represents the angle between the line connecting the center points of the ground truth box (GT Box) and the center points of the prediction box (Anchor Box, Pred Box) and the horizontal direction. This represents the x-coordinate of the center point of the truth box. This represents the y-coordinate of the center point of the truth box. This represents the x-coordinate of the center point of the prediction box. This represents the ordinate of the center point of the prediction box. This represents a preset minimum constant, for example, it can be taken as... .

[0068] 2) Distance loss : , , , in, This represents the normalized offset of the center points of the truth box and the prediction box in the horizontal / vertical direction. Indicates the attenuation coefficient. express Normalized partial variables under given conditions The time indicates the horizontal direction. When indicates the vertical direction, This represents the x-coordinate of the center point of the prediction box. This represents the ordinate of the center point of the prediction box. This represents the x-coordinate of the center point of the truth box. This represents the y-coordinate of the center point of the truth box. This represents the width of the smallest bounding box between the prediction box and the truth box. This represents the length of the smallest bounding box between the prediction box and the truth box.

[0069] 3) Shape loss : , , in, This represents the shape difference between the truth box and the prediction box. This indicates the shape deviation under condition t2. Time represents the width dimension. Time represents the height dimension. , The importance of shape cost is determined. Indicates the width of the prediction box. Indicates the height of the truth box. This represents the width of the truth box. This indicates the height of the truth box.

[0070] 4) IoU loss : , in, The loss represents the intersection-union ratio (IoU) between the predicted bounding box and the ground truth bounding box. Indicates the prediction box. Represents a real bounding box.

[0071] 5) SIoU loss : .

[0072] At the same time, the inventors of this application discovered that, although SIoU loss Through distance loss and shape loss Geometric constraints can effectively improve detection performance, but in real-world scenarios, there may be some small objects that may lead to deviations in the detection capability for small objects. Therefore, this application further introduces the idea of ​​size loss, which calculates IoU loss by generating auxiliary bounding boxes to improve the model's ability to locate small objects and accelerate the regression of predicted bounding boxes.

[0073] Correspondingly, as an exemplary embodiment, the loss function involved in the model training process may specifically include the improved IoU loss, and the corresponding quantization formula is: , , , , in, Indicates improved IoU loss. Indicates the existing Losses (already described above), This represents the existing distance loss (as previously explained). For size loss, express Balance ratio parameters under certain conditions Time represents the lower left dimension. Time represents the lower right dimension. express The Euclidean distance between the predicted bounding box and the ground truth bounding box under condition 3. This indicates the coordinates of the top-left corner of the prediction box. This indicates the coordinates of the top-left corner of the bounding box. This indicates the coordinates of the bottom right corner of the prediction box. This indicates the coordinates of the bottom right corner of the bounding box. This represents random error.

[0074] In layman's terms, this application introduces a scale loss mechanism, using the coordinates of the top-left and bottom-right corners of the ground truth bounding box and the top-left and bottom-right corners of the predicted bounding box to quantify the size loss. This scale loss function optimizes the shape loss between the ground truth and predicted bounding boxes, and balances the scaling parameters. It can adaptively adjust the changes in parameter values, allowing it to focus more quickly on identifying difficult target samples, thereby improving the algorithm's accuracy in recognizing target objects.

[0075] In summary, the ship inspection results achieved by this application can be demonstrated in the following three aspects: 1) Highly portable, applicable to different types of remote sensing images; 2) It has strong versatility and can be applied to target recognition in various complex scenarios; 3) High accuracy, with high accuracy in identifying small targets.

[0076] The above is an introduction to the ship detection method based on remote sensing images provided in this application. To facilitate better implementation of the ship detection method based on remote sensing images provided in this application, this application also provides a ship detection device based on remote sensing images from the perspective of functional modules.

[0077] See Figure 3 , Figure 3This is a schematic diagram of a ship detection device based on remote sensing imagery according to this application. In this application, the ship detection device 300 based on remote sensing imagery may specifically include the following structure: Acquisition unit 301 is used to acquire the current remote sensing image to be detected; The detection unit 302 is used to send the current remote sensing image into a pre-configured ship detection model. The ship detection model is configured based on the improved YOLOv8 model. The ship detection model is used to perform single-stage target detection processing of ship objects based on the remote sensing image input to the model. In the specific working process of the ship detection model, after extracting the edge features and RGB spectral features of the remote sensing image input to the model, specific detection processing is carried out based on the four-channel image features composed of edge features and RGB spectral features. Extraction unit 303 is used to extract the ship detection results of the current remote sensing image output by the ship detection model.

[0078] In one exemplary embodiment, the improved YOLOv8 model includes a backbone network structure, a neck network structure, and a head network structure. The backbone network structure adopts the CSPDarknet-53 architecture, which consists of 53 convolutions, including residual blocks and C2f modules. The residual blocks are used to help capture details and contextual information in the image, and the C2f modules are used to fuse feature maps of different scales to extract feature information and have gradient flow. The neck network structure consists of a PAN-FPN structure. The upsampled convolution is removed from the PAN structure to achieve a lightweight effect without changing the original performance. The PAN structure added to the FPN structure enhances the path information through the complementarity of shallow and deep information. The head network structure employs a decoupled head structure, specifically using two independent branches for target classification and prediction / regression, and then each branch is further used... The convolutional layers complete the classification and localization tasks, while anchor-free detection is used to improve convergence speed and accuracy.

[0079] In another exemplary embodiment, the ship detection model specifically extracts edge features from the remote sensing image input to the model using a pre-configured Sobel edge feature function to obtain the corresponding edge features.

[0080] In yet another exemplary embodiment, the apparatus further includes a training unit 304 for: Acquire remote sensing images of sample vessels and configure corresponding annotations; A ship detection model is trained based on sample ship remote sensing images.

[0081] In yet another exemplary embodiment, the loss function involved in the model training process includes angular loss. Distance loss Shape loss IoU loss and SIoU loss .

[0082] In yet another exemplary embodiment, the loss function involved in the model training process includes an improved IoU loss, the corresponding quantization formula of which is: , , , , in, Indicates improved IoU loss. Indicates the existing loss, This indicates the existing distance loss. For size loss, express Balance ratio parameters under certain conditions Time represents the lower left dimension. Time represents the lower right dimension. express The Euclidean distance between the predicted bounding box and the ground truth bounding box under the given conditions. This indicates the coordinates of the top-left corner of the prediction box. This indicates the coordinates of the top-left corner of the bounding box. This indicates the coordinates of the bottom right corner of the prediction box. This indicates the coordinates of the bottom right corner of the bounding box. This represents random error.

[0083] In yet another exemplary embodiment, the apparatus further includes a response unit 305, configured to: Adaptive response processing based on ship detection results.

[0084] This application also provides a processing device from a hardware architecture perspective. As mentioned earlier, in practice, a processing device may exist as a device cluster. In this case, each device in the device cluster can also be referred to as a processing device. See [reference needed]. Figure 4 , Figure 4 This diagram illustrates a structural schematic of the processing device of this application. Specifically, the processing device may include a processor 401, a memory 402, and an input / output device 403. The processor 401 executes the computer program stored in the memory 402 to implement, for example... Figure 1The corresponding steps of the ship detection method based on remote sensing imagery in the embodiment; or, when the processor 401 executes the computer program stored in the memory 402, it implements as follows: Figure 3 Corresponding to the functions of each unit in the embodiment, the memory 402 is used to store the functions executed by the processor 401 as described above. Figure 1 The computer program required for the ship detection method based on remote sensing imagery in the corresponding embodiment.

[0085] For example, a computer program may be divided into one or more modules / units, one or more of which are stored in memory 402 and executed by processor 401 to complete this application. One or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in a computer device.

[0086] The processing device may include, but is not limited to, processor 401, memory 402, and input / output device 403. Those skilled in the art will understand that the illustrations are merely examples of the processing device and do not constitute a limitation on the processing device. It may include more or fewer components than illustrated, or combine certain components, or different components. For example, the processing device may also include network access devices, buses, etc., and processor 401, memory 402, input / output device 403, etc., are connected via a bus.

[0087] Processor 401 can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the processing device, connecting various parts of the device through various interfaces and lines.

[0088] The memory 402 can be used to store computer programs and / or modules. The processor 401 implements various functions of the computer device by running or executing the computer programs and / or modules stored in the memory 402 and by calling data stored in the memory 402. The memory 402 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for at least one function, etc.; the data storage area may store data created according to the use of the processing device, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device or other volatile solid-state storage device.

[0089] When processor 401 executes a computer program stored in memory 402, it can specifically perform the following functions: Acquire the current remote sensing image to be detected; The current remote sensing image is fed into a pre-configured ship detection model, which is configured based on an improved YOLOv8 model. The ship detection model is used to perform single-stage target detection processing of ship objects based on the remote sensing image input to the model. In the specific working process of the ship detection model, after extracting the edge features and RGB spectral features of the remote sensing image input to the model, specific detection processing is carried out based on the four-channel image features composed of edge features and RGB spectral features. Extract the ship detection results from the current remote sensing image output by the ship detection model.

[0090] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the ship detection device, processing equipment, and its corresponding units based on remote sensing imagery described above can be found in [reference to...]. Figure 1 The description of the ship detection method based on remote sensing imagery in the corresponding embodiment will not be repeated here.

[0091] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be performed by instructions, or by instructions controlling related hardware. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0092] Therefore, this application provides a computer-readable storage medium storing a plurality of instructions that can be loaded by a processor to execute the present application. Figure 1 The steps of the ship detection method based on remote sensing imagery in the corresponding embodiment can be referred to as follows for specific operations. Figure 1 The description of the ship detection method based on remote sensing imagery in the corresponding embodiments will not be repeated here.

[0093] The computer-readable storage medium may include: read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0094] Because of the instructions stored in the computer-readable storage medium, the present application can be executed as described above. Figure 1 The steps of the ship detection method based on remote sensing imagery in the corresponding embodiment can therefore achieve the results of this application. Figure 1 The beneficial effects that the ship detection method based on remote sensing images can achieve in the corresponding embodiments are detailed in the preceding description and will not be repeated here.

[0095] The foregoing has provided a detailed description of the ship detection method, apparatus, processing equipment, and computer-readable storage medium based on remote sensing imagery provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are merely for the purpose of helping to understand the core ideas of this application; furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A ship detection method based on remote sensing imagery, characterized in that, The method includes: Acquire the current remote sensing image to be detected; The current remote sensing image is fed into a pre-configured ship detection model, which is configured based on an improved YOLOv8 model. The ship detection model is used to perform single-stage target detection processing of ship objects based on the remote sensing image input to the model. In the specific working process of the ship detection model, after extracting the edge features and RGB spectral features of the remote sensing image input to the model, specific detection processing is carried out based on the four-channel image features composed of the edge features and the RGB spectral features. Extract the ship detection results from the current remote sensing image, which are output by the ship detection model.

2. The method according to claim 1, characterized in that, The improved YOLOv8 model includes a backbone network structure, a neck network structure, and a head network structure. The backbone network structure adopts the CSPDarknet-53 architecture, which consists of 53 convolutions, including residual blocks and C2f modules. The residual blocks are used to help capture details and contextual information in the image, and the C2f modules are used to fuse feature maps of different scales to extract feature information and have gradient flow. The neck network structure consists of a PAN-FPN structure. The upsampled convolution is removed from the PAN structure to achieve a lightweight effect without changing the original performance. The PAN structure added to the FPN structure enhances the path information through the complementarity of shallow and deep information. The head network structure adopts a decoupled head structure, specifically using two independent branches for target classification and prediction regression, and then each branch is further used... The convolutional layers complete the classification and localization tasks, while anchor-free detection is used to improve convergence speed and accuracy.

3. The method according to claim 1, characterized in that, The ship detection model specifically extracts edge features from the remote sensing image input to the model using a pre-configured Sobel edge feature function to obtain the corresponding edge features.

4. The method according to claim 1, characterized in that, The method further includes: Acquire remote sensing images of sample vessels and configure corresponding annotations; The ship detection model is trained based on the sample ship remote sensing images.

5. The method according to claim 4, characterized in that, The loss functions involved in model training include angle loss. Distance loss Shape loss IoU loss and SIoU loss .

6. The method according to claim 1, characterized in that, The loss functions involved in model training include improved IoU loss, and the corresponding quantization formula is as follows: ， ， ， ， in, This represents the improved IoU loss. Indicates the existing loss, This indicates the existing distance loss. For size loss, express Balance ratio parameters under certain conditions Time represents the top left dimension. Time represents the lower right dimension. express The Euclidean distance between the predicted bounding box and the ground truth bounding box under the given conditions. This indicates the coordinates of the top-left corner of the prediction box. This indicates the coordinates of the top-left corner of the actual bounding box. This indicates the coordinates of the lower right corner of the prediction box. This indicates the coordinates of the lower right corner of the actual bounding box. This represents random error.

7. The method according to claim 1, characterized in that, The method further includes: Based on the vessel detection results, an appropriate response process is performed.

8. A ship detection device based on remote sensing imagery, characterized in that, The device includes: The acquisition unit is used to acquire the current remote sensing image to be detected; The detection unit is used to send the current remote sensing image into a pre-configured ship detection model. The ship detection model is configured based on an improved YOLOv8 model. The ship detection model is used to perform single-stage target detection processing of ship objects based on the remote sensing image input to the model. In the specific working process of the ship detection model, after extracting the edge features and RGB spectral features of the remote sensing image input to the model, specific detection processing is carried out based on the four-channel image features composed of the edge features and the RGB spectral features. The extraction unit is used to extract the ship detection results of the current remote sensing image output by the ship detection model.

9. A processing device, characterized in that, It includes a processor and a memory, wherein the memory stores a computer program, and the processor executes the method as described in any one of claims 1 to 7 when it invokes the computer program in the memory.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a plurality of instructions adapted for loading by a processor to perform the method of any one of claims 1 to 7.

Images