Ship speed inversion method, device, medium and product based on single-frame ship image

Abstract

The application discloses a kind of based on single-frame ship image's speed inversion method, equipment, medium and product, the method includes: obtaining the target ship remote sensing image to be carried out speed inversion, and identifying target ship in target ship remote sensing image, according to the image area where target ship is located, in the target ship remote sensing image positioning ship wake identification area and water condition identification area;Image recognition is carried out from ship wake identification area and water condition identification area respectively, obtains the ship wake feature and water condition feature matched with current driving scene;According to water condition feature, determine the dynamic weight of each ship wake feature, combine weight and feature input speed inversion model, output predicted speed, the technical scheme of the embodiment of the application can realize only with single-frame image's water condition adaptive speed high-precision inversion, provides real-time, reliable automatic perception ability for maritime supervision, navigation safety and marine target behavior analysis.

Description

Technical Field

[0001] This invention relates to the field of marine remote sensing monitoring technology, and in particular to a method, equipment, medium, and product for retrieving ship speed based on a single frame of ship imagery. Background Technology

[0002] With the booming development of the maritime transport industry and the continuous improvement of maritime safety supervision requirements, real-time and accurate speed monitoring of maritime targets has become a key technology for ensuring navigation safety, maintaining maritime traffic order, and conducting intelligent situational awareness at sea. However, the sea surface environment is dynamic and ever-changing, and traditional speed measurement methods are difficult to achieve high-precision and robust speed perception under single-frame image conditions.

[0003] In existing technologies, ship speed inversion mainly relies on two types of methods: one is signal analysis and trajectory calculation based on the Automatic Identification System (AIS). This method completely fails when the target is not equipped with an AIS, the signal is interfered with, or it is actively turned off, resulting in a blind spot in supervision; the other is target matching and tracking technology based on multi-frame remote sensing image sequences. This method has high requirements for image continuity, imaging quality, and computing resources. In complex scenarios, the stability and timeliness of the algorithm are difficult to guarantee, and its universality is insufficient. Summary of the Invention

[0004] This invention provides a method, device, medium, and product for ship speed inversion based on single-frame ship images, to achieve high-precision, adaptive inversion of ship speed without relying on continuous frames and AIS signals.

[0005] According to one aspect of the present invention, a method for speed inversion based on a single frame of ship imagery is provided, the method comprising:

[0006] Acquire remote sensing images of the target vessel to be used for speed inversion, and identify the target vessel in the remote sensing images;

[0007] Based on the image region where the target vessel is located, the vessel wake recognition region and the water condition recognition region are located in the remote sensing image of the target vessel.

[0008] Image recognition is performed in the ship wake recognition area and the water condition recognition area respectively to obtain the ship wake features and water condition features that match the current driving scene of the target ship.

[0009] Among them, the characteristics of ship wakes include wake length, wake width, diffusion angle and texture clarity, and the characteristics of water conditions include water surface roughness, water surface brightness uniformity and water surface high frequency energy ratio.

[0010] Based on the characteristics of each water condition, dynamic weights corresponding to each of the ship wake features are determined, and each ship wake feature and the matched dynamic weights are input together into a pre-trained speed inversion model to obtain the predicted speed of the target ship in the remote sensing image of the target ship output by the speed inversion model.

[0011] According to another aspect of the present invention, a speed inversion device based on a single frame of ship imagery is provided, the device comprising:

[0012] The image data acquisition module is used to acquire remote sensing images of the target ship to be subjected to speed inversion and to identify the target ship in the remote sensing images.

[0013] The positioning module is used to locate the ship wake recognition area and the water condition recognition area in the remote sensing image of the target ship based on the image area where the target ship is located.

[0014] The feature acquisition module is used to perform image recognition in the ship wake recognition area and the water condition recognition area respectively, and acquire ship wake features and water condition features that match the current driving scene of the target ship. Among them, the ship wake features include wake length, wake width, diffusion angle and texture clarity, and the water condition features include water surface roughness, water surface brightness uniformity and water surface high frequency energy ratio.

[0015] The speed prediction module is used to determine the dynamic weights corresponding to each of the ship wake features based on the water condition characteristics, and input each ship wake feature and the matched dynamic weights into a pre-trained speed inversion model to obtain the predicted speed of the target ship in the target ship remote sensing image output by the speed inversion model.

[0016] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0017] At least one processor; and

[0018] A memory communicatively connected to the at least one processor; wherein,

[0019] The memory stores a computer program that can be executed by the at least one processor, which enables the at least one processor to execute a speed inversion method based on a single frame of ship imagery as described in any embodiment of the present invention.

[0020] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions, the computer instructions being configured to cause a processor to execute and implement a speed inversion method based on a single-frame ship image as described in any embodiment of the present invention.

[0021] According to another aspect of the present invention, a computer program product is also provided, including a computer program / instructions that, when executed by a processor, implement the steps of the method as described in any embodiment of the present invention.

[0022] The technical solution of this invention involves acquiring remote sensing images of the target vessel to be inverted in terms of speed, identifying the target vessel within them, locating the vessel wake recognition region and water condition recognition region in the remote sensing image based on the image region where the target vessel is located, performing image recognition in the wake recognition region and water condition recognition region respectively, acquiring various vessel wake features and water condition features matching the current navigation scene of the target vessel, determining dynamic weights corresponding to each vessel wake feature based on each water condition feature, and inputting each vessel wake feature and the matching dynamic weights into a pre-trained speed inversion model to obtain the predicted speed of the target vessel output by the speed inversion model. This achieves automated and high-precision speed inversion of vessels under complex sea conditions, effectively overcoming the dependence on continuous frame images and prior information about the vessel, and improving the robustness and practicality of the speed inversion method.

[0023] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention, 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 the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is a flowchart of a speed inversion method based on a single frame of ship imagery provided according to Embodiment 1 of the present invention;

[0026] Figure 2 This is a flowchart of another method for speed inversion based on a single frame of ship imagery provided in Embodiment 2 of the present invention;

[0027] Figure 3 This is a flowchart of a speed inversion process based on a single frame of ship imagery in a specific scenario applicable to an embodiment of the present invention.

[0028] Figure 4 This is a schematic diagram of ship wake feature extraction in a specific scenario applicable to an embodiment of the present invention;

[0029] Figure 5This is a schematic diagram of the structure of a speed inversion device for a single frame ship image provided in Embodiment 3 of the present invention;

[0030] Figure 6 This is a schematic diagram of the structure of an electronic device for implementing a method for retrieving the speed of a single-frame ship image according to an embodiment of the present invention. Detailed Implementation

[0031] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0032] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention 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 of the invention described herein can be implemented in orders other than those 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 apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0033] Example 1

[0034] Figure 1 This is a flowchart of a speed inversion method based on a single-frame ship image provided in Embodiment 1 of the present invention. This embodiment is applicable to ship speed monitoring scenarios that do not rely on an automatic identification system and continuous image sequences. The method can be executed by a speed inversion device based on a single-frame ship image. This device can be implemented in hardware and / or software and is generally configured in an electronic device.

[0035] Correspondingly, such as Figure 1 As shown, the method includes:

[0036] S110. Acquire remote sensing images of the target vessel to be used for speed inversion, and identify the target vessel in the remote sensing images.

[0037] Speed ​​inversion can be understood as the process of using visual features related to ship motion obtained through indirect observation methods, and applying physical laws or data-driven models to reverse-calculate the actual speed of the ship.

[0038] In this embodiment, a remote sensing image of the target ship to be processed is acquired, and the target ship whose speed needs to be calculated is automatically detected and located from it.

[0039] S120. Based on the image area where the target vessel is located, locate the vessel wake recognition area and the water condition recognition area in the remote sensing image of the target vessel.

[0040] In this embodiment, after the target vessel is identified, two analysis areas are identified based on the target vessel's position in the remote sensing image: one area is located behind the vessel and is specifically used to capture and extract the vessel's wake generated during navigation; the other area is the background water surface around the vessel, used to analyze and evaluate the current water conditions.

[0041] S130. Perform image recognition in the ship wake recognition area and water condition recognition area respectively, and obtain the ship wake features and water condition features that match the current driving scene of the target ship.

[0042] Among them, the characteristics of ship wakes include wake length, wake width, diffusion angle and texture clarity, and the characteristics of water conditions include water surface roughness, water surface brightness uniformity and water surface high-frequency energy ratio.

[0043] Specifically, wake length can be understood as the distance the wake created by a ship's disturbance of the water body extends in the image. Wreath width can be understood as the lateral spread of the ship's wake perpendicular to the direction of navigation. Spread angle can be understood as the opening angle of the two edges of the wake, approaching a constant value under deep water conditions. Texture sharpness can be understood as the image gradient variance calculated by an algorithm, used to quantify the sharpness and clarity of the wake texture. Water surface roughness can be understood as a normalized index calculated by an algorithm, used to quantify the undulation and roughness of sea surface waves. Water surface brightness uniformity can be understood as an index obtained by calculating the ratio of the pixel grayscale mean to the standard deviation, used to reflect the uniformity of brightness and darkness on the sea surface under illumination. High-frequency energy of the water surface can be understood as the proportion of high-frequency energy in the total energy after performing a Fourier transform on the sea surface image, used to characterize the activity level of small-scale ripples or turbulence.

[0044] In this embodiment, based on the located ship wake identification area and water condition identification area, a series of image processing techniques are used to extract ship wake features and water condition features. Specifically, extracting wake features is to establish a direct correlation model with ship speed, which is the core basis for inversion; extracting water condition features is to quantify the degree of interference in the imaging environment, providing data support for subsequent adaptive weighting, thereby improving the model's generalization ability and inversion accuracy under different sea conditions.

[0045] Optionally, based on the above embodiments, image recognition is performed in the ship wake recognition area to obtain ship wake features that match the current navigation scene of the target ship. This may include:

[0046] Edge detection and contour analysis are performed on the ship wake recognition area to determine the maximum connected region corresponding to the target ship wake;

[0047] Calculate the minimum bounding rectangle of the maximum connected region, take the length of the long side of the minimum bounding rectangle as the tail length, the length of the short side of the minimum bounding rectangle as the tail width, and the rotation angle of the minimum bounding rectangle as the diffusion angle.

[0048] The gradient variance of the image corresponding to the largest connected region is calculated based on the Laplacian operator, and the normalized gradient variance is used as the texture sharpness.

[0049] Generally, the first step is to perform preliminary image analysis on the designated ship wake identification area. The purpose is to separate the main wake structure generated by the ship's navigation from the relatively complex water background. Preferably, an edge detection algorithm is used to find all obvious edge contours within the area. Then, contour analysis technology is used to filter out the region with the largest area and best connectivity from all detected contours. This identified largest connected region is determined as the most complete wake body generated by the current target ship, and all subsequent feature calculations will be based on this largest connected region.

[0050] Generally, after determining the largest connected region corresponding to the target ship's wake, it is necessary to quantify several of its geometric properties. Preferably, the minimum bounding rectangle of the region is calculated, which can enclose the entire target ship's wake in the form of a rotated frame. The longer side of this rectangle corresponds to the extension scale of the target ship's wake in the direction of travel, i.e., the wake length; its shorter side corresponds to the spread width of the target ship's wake perpendicular to the direction of travel, i.e., the wake width. At the same time, the rotation angle of this rectangle relative to the horizontal axis of the image directly reflects the opening direction of the V-shaped structure of the target ship's wake and is defined as the spread angle. In this way, several key geometric features of the wake are transformed into clear and measurable values.

[0051] Generally, besides geometric dimensions, the texture characteristics of a wake also contain important information. To quantify this characteristic, the Laplacian operator is typically used to process the wake region image. This operator can keenly capture rapid changes in image intensity, i.e., edge and texture details. By calculating the gradient variance of the processed image, a statistic characterizing the overall texture variation in the region can be obtained. Subsequently, the variance value is normalized to eliminate the influence of factors such as absolute brightness, and the final standardized value is defined as texture sharpness. The higher this value, the stronger the texture contrast and the sharper the edges of the wake.

[0052] Optionally, based on the above embodiments, image recognition is performed in the water condition recognition area to obtain various water condition features that match the current navigation scene of the target vessel. This may include:

[0053] The Laplacian operator is used to perform convolution processing on the water condition identification region to obtain the response map, and the response variance of the response map is calculated.

[0054] The Sobel operator is used to perform convolution processing on the water condition identification region to obtain the gradient map, and the gradient mean of the gradient map is calculated.

[0055] The normalized value obtained by dividing the response variance by the gradient mean is used as the water surface roughness.

[0056] Calculate the mean gray value and standard deviation of each pixel in the water condition identification area, and calculate the normalized value obtained by dividing the mean gray value by the standard deviation of gray value, which is used as the water surface brightness uniformity.

[0057] A two-dimensional fast Fourier transform is performed on the water condition identification area to obtain a frequency domain map, and low-frequency and high-frequency intervals are identified in the frequency domain map;

[0058] Calculate the low-frequency energy value in the low-frequency range and the high-frequency energy value in the high-frequency range, and calculate the proportion of high-frequency energy on the water surface based on the low-frequency energy value and the high-frequency energy value.

[0059] Generally, the microscopic texture undulations of the water surface are first analyzed collaboratively using two classic image processing operators. The Laplacian operator is used to convolve the water region, aiming to highlight the rapid second-order changes in intensity within the image. Its output response map effectively reflects the intensity of details such as fine ripples and wave crests. The variance of this response map is then calculated to statistically represent the overall intensity of texture changes within the region. Simultaneously, the Sobel operator is used for convolution. The resulting gradient map primarily characterizes the intensity of the first-order directional derivative of the image, and its mean reflects the average edge intensity level of the region. Finally, the variance of the Laplacian response is divided by the mean of the Sobel gradient. This ratio serves as a normalization process, eliminating the influence of overall illumination intensity on the absolute value, thus obtaining a relatively stable comprehensive index characterizing the local irregularities and ripple density of the water surface—i.e., water surface roughness.

[0060] Generally, to assess the distribution of light and shadow on the sea surface after being affected by illumination, basic grayscale statistical analysis is performed on the water condition identification area. Calculating the average grayscale value of all pixels within the area reflects the overall brightness level of that area; while calculating the standard deviation of the grayscale value measures the dispersion of the brightness of each pixel relative to the average value, i.e., non-uniformity. Dividing the grayscale mean by the grayscale standard deviation yields a value that is essentially the reciprocal of the coefficient of variation. When the sea surface is uniformly bright, the standard deviation is small, and the ratio is large; when there are significant bright and dark patches, the standard deviation is large, and the ratio becomes smaller. Therefore, this calculated ratio is defined as the water surface brightness uniformity, used to quantify the uniformity or mottled state of light reflection.

[0061] Generally, to analyze the spatial frequency distribution characteristics of water ripples, a two-dimensional Fast Fourier Transform (FFT) is performed on the image of the water condition identification area. This transforms the image from a spatial domain depicting brightness variations with location to a frequency domain depicting energy distribution with frequency. In the resulting frequency domain image, the low-frequency portion near the center represents the slowly changing large-scale background and main outlines, while the high-frequency portion far from the center corresponds to rapidly changing details, edges, and small-scale textures. By calculating the energy values ​​contained in the low-frequency and high-frequency intervals separately, the proportion of fluctuations at different scales can be quantified. Finally, the energy value of the high-frequency interval is compared with the total energy or the sum of high and low-frequency energies to obtain the proportion of high-frequency energy on the water surface. The higher this proportion, the more active the detail changes caused by small-scale ripples, broken waves, or turbulence in the image.

[0062] S140. Based on the characteristics of each water condition, determine the dynamic weights corresponding to each of the ship wake features, and input each ship wake feature and the matched dynamic weights into the pre-trained speed inversion model to obtain the predicted speed of the target ship in the remote sensing image of the target ship output by the speed inversion model.

[0063] In this embodiment, since different water conditions have different effects on the shape and salience of the wake, appropriate weights are dynamically assigned to each wake feature based on the aforementioned extracted water condition features. These dynamically weighted wake features are then combined into a feature vector and input into a pre-trained speed inversion model. This model analyzes these weighted features and ultimately outputs an accurate prediction of the target vessel's current speed.

[0064] Optionally, based on the above embodiments, determining the dynamic weights corresponding to each of the aforementioned ship wake characteristics according to various water condition characteristics may include:

[0065] Based on each water condition feature and the fusion weight matched with each water condition feature, the target sea state level matched for the current driving scenario is calculated.

[0066] Based on the target sea state level, a search is performed in the pre-weighted allocation strategy to obtain a target weighted allocation strategy that matches the target sea state level, and dynamic weights corresponding to each of the ship wake features are obtained from the target weighted allocation strategy.

[0067] Generally, each water condition characteristic, such as surface roughness and brightness uniformity, is pre-assigned a fusion weight that reflects its importance. Then, using a pre-defined comprehensive calculation rule, all water condition characteristics are combined with their corresponding fusion weights to calculate a comprehensive score. Based on which pre-defined range this score falls within, the predefined sea state level can be determined, such as calm, moderate, or severe sea state; this level is the target sea state level.

[0068] Generally, after determining the target sea state level, it is necessary to assign appropriate importance weights to different wake features based on that level. Before implementing the method, a weight allocation strategy table is pre-built and stored, which specifies a particular weight allocation scheme for each possible sea state level. This scheme is trained based on prior knowledge or historical data, and it clarifies the credibility or importance ratio of the contribution of each of the four features—wake length, width, spread angle, and texture sharpness—to speed inversion at a specific sea state level. Therefore, based on the obtained target sea state level, the row of weight allocation schemes that perfectly matches it is found in the strategy table. From this scheme, the specific weight values ​​pre-set for each wake feature are directly read and output; these values ​​are the dynamic weights ultimately used to adjust the feature inputs.

[0069] The technical solution of this invention involves acquiring remote sensing images of the target vessel to be inverted in terms of speed, identifying the target vessel within them, locating the vessel wake recognition region and water condition recognition region in the remote sensing image based on the image region where the target vessel is located, performing image recognition in the wake recognition region and water condition recognition region respectively, acquiring various vessel wake features and water condition features matching the current navigation scene of the target vessel, determining dynamic weights corresponding to each vessel wake feature based on each water condition feature, and inputting each vessel wake feature and the matching dynamic weights into a pre-trained speed inversion model to obtain the predicted speed of the target vessel output by the speed inversion model. This achieves automated and high-precision speed inversion of vessels under complex sea conditions, effectively overcoming the dependence on continuous frame images and prior information about the vessel, and improving the robustness and practicality of the speed inversion method.

[0070] Example 2

[0071] Figure 2 This is a flowchart of a speed inversion method based on a single-frame ship image provided in Embodiment 2 of the present invention. This embodiment is an optimization based on the above embodiments. Specifically, the speed inversion method based on a single-frame ship image has been refined.

[0072] like Figure 2 As shown, the method includes:

[0073] S210. Acquire remote sensing images of the target vessel to be used for speed inversion, and identify the target vessel in the remote sensing images.

[0074] S220. Based on the image area where the target vessel is located, locate the vessel wake recognition area and the water condition recognition area in the remote sensing image of the target vessel.

[0075] S230. Perform image recognition in the ship wake recognition area and water condition recognition area respectively, and obtain the ship wake features and water condition features that match the current driving scene of the target ship.

[0076] Among them, the characteristics of ship wakes include wake length, wake width, diffusion angle and texture clarity, and the characteristics of water conditions include water surface roughness, water surface brightness uniformity and water surface high-frequency energy ratio.

[0077] S240. Based on the image region where the target ship is located, obtain the bounding rectangle of the target ship in the remote sensing image of the target ship, and obtain the pixel width and pixel length of the bounding rectangle.

[0078] In this embodiment, in order to eliminate the inherent influence of ships of different sizes on the wake features, it is necessary to first estimate the size of the target ship itself. In the image, a minimum horizontal rectangle that can completely enclose the identified target ship body is drawn, and the width and length of this rectangle (in pixels) are recorded.

[0079] S250. Calculate the physical dimensions of the target ship based on the ground sampling distance matched to the remote sensing image of the target ship, the pixel width, and the pixel length.

[0080] In this embodiment, the obtained ship pixel size is only a measurement on the image plane and must be converted into its actual physical size in geographic space. This conversion relies on the data matched by the remote sensing image—the ground sampling distance. The ground sampling distance defines the actual ground distance corresponding to a single pixel in the image. By multiplying the pixel width and pixel length of the target ship's bounding rectangle by this ground sampling distance, the actual physical width and physical length of the target ship's hull can be estimated.

[0081] S260. Based on the physical dimensions of the target vessel, the wake length and wake width are normalized to eliminate feature deviations caused by differences in vessel dimensions.

[0082] In this embodiment, after obtaining the ship's actual physical dimensions, the extracted wake length and wake width are divided by representative dimensions such as the ship's physical length or width to perform a proportional normalization process. The purpose of this step is to normalize the wake dimensions generated by ships of different sizes to a comparable benchmark, thereby eliminating the bias caused by differences in the ship's physical size, such as "large ships naturally have large wakes and small ships naturally have small wakes," so that the wake features can more purely reflect the speed information.

[0083] S270. Based on the characteristics of each water condition, determine the dynamic weights corresponding to each of the ship wake features, and input each ship wake feature and the matched dynamic weights into the pre-trained speed inversion model to obtain the predicted speed of the target ship in the remote sensing image of the target ship output by the speed inversion model.

[0084] Furthermore, regarding the construction of the model training dataset, in this embodiment of the invention, it is first necessary to obtain a sample set containing multiple sample ship remote sensing images and the corresponding real speed labels for each image. Specifically, all sample images in the sample set are read in batches, and the aforementioned preprocessing and ship wake feature extraction steps are sequentially performed on each sample image to obtain a four-dimensional feature vector for each image, consisting of wake length, wake width, spread angle, and texture sharpness. Simultaneously, a label file recorded in a structured format is read, which at least contains the identifier of the sample image (such as "image name") and its corresponding real speed label. Subsequently, using the image identifier as the matching key, each extracted four-dimensional feature vector is precisely associated and paired with the corresponding real speed label in the label file. Finally, all successfully paired data are integrated to construct a structured model training dataset, which contains a feature matrix composed of all feature vectors and a label vector composed of all real speed labels.

[0085] Furthermore, regarding the training and validation of the speed inversion model, this embodiment of the invention employs the random forest regression algorithm as the core model. First, the model training dataset is divided into a training subset and a test subset. The random forest regression model is trained using the training subset to obtain an initial speed inversion model, and the performance of the model is initially evaluated using the test subset. To further scientifically evaluate the model's generalization ability and stability, K-fold cross-validation is used for validation. Specifically, the data used for model training (usually the training subset) is randomly and uniformly divided into K mutually exclusive data subsets. Then, K rounds of training and validation are executed. In each round, one data subset is used as the validation set, and the remaining K-1 subsets are combined as the training set for that round. The model is retrained and validated to obtain a validation result. All K validation results obtained from the K rounds are collected, and a comprehensive evaluation index of model performance (such as mean absolute error) is calculated based on these results, serving as the basis for evaluating the model's robustness and generalization ability. Based on the evaluation results of cross-validation, the model parameters can be optimized and finalized.

[0086] Optionally, based on the above embodiments, the wake features of each ship and the matched dynamic weights are input together into a pre-trained speed inversion model to obtain the predicted speed of the target ship in the remote sensing image of the target ship output by the speed inversion model, which may include:

[0087] Each ship wake feature is assigned a matching dynamic weight, and a weighted calculation is performed to obtain a weighted ship wake feature vector.

[0088] The weighted ship wake feature vector is input into a feature attention module to calculate the attention score corresponding to each weighted ship wake feature.

[0089] The weighted ship wake feature vector and its matching attention score are input together into the pre-trained speed inversion model, wherein the speed inversion model is a random forest regression model, and its multiple decision trees perform differentiated sensitivity learning on different feature dimensions based on the attention score when splitting.

[0090] The predicted speed of the target ship is obtained based on the output of all decision trees in the random forest regression model.

[0091] Generally, the extracted and size-normalized ship wake features are first combined with dynamic weights assigned to them based on water condition information. This combination process is usually a multiplication operation, the purpose of which is to perform an initial scaling adjustment on the original feature values ​​based on water condition adaptation, thereby generating a weighted feature vector. This step can be understood as, based on the assessment of the current sea surface environment, highlighting wake features that may be more reliable under specific conditions, while appropriately weakening features that may be more susceptible to interference.

[0092] Generally, after obtaining the weighted feature vector of the ship's wake, a dedicated mechanism is used to evaluate the relative importance of these features in the specific input. This mechanism, the feature attention module, analyzes the weighted feature vector. Through internal learnable parameters and nonlinear transformations, it simulates an "attention allocation" process, calculating a new score for each feature dimension in the vector. This score, called the attention score, quantifies the contribution or criticality of each feature to making a correct speed prediction within the specific context of the current weighted feature combination.

[0093] Generally, the weighted feature vectors and their corresponding attention scores are then fed into the random forest regression model. During the training and inference process of this model, the numerous decision trees within it do not treat all feature dimensions equally when splitting nodes. Instead, they refer to the incoming attention scores. Specifically, when deciding which feature to use and where to split, feature dimensions with higher attention scores receive higher priority or are assigned a greater selection probability. This allows each decision tree to dynamically focus its learning process on the more important features in the current input, achieving a differentiated, data-driven, and sensitive learning process.

[0094] Generally, a random forest regression model eventually aggregates the speed predictions made independently by all its decision trees. This aggregation process typically uses an averaging method to combine the outputs of all trees, resulting in a stable and robust ensemble prediction. This final value is the method's estimate of the target ship's speed in the current frame of imagery, i.e., the predicted speed.

[0095] Optionally, based on the above embodiments, the weighted ship wake feature vector is input into a feature attention module to calculate the attention score corresponding to each weighted ship wake feature, which may include:

[0096] The weighted ship wake feature vector is input into a nonlinear transformation layer, and the initial importance score of each weighted ship wake feature is calculated through a learnable weight matrix and activation function.

[0097] The initial importance scores of all weighted ship wake features are input into a normalization layer. The initial importance scores are converted into a probability distribution form through a normalization function to obtain the normalized attention weights.

[0098] The normalized attention weights are used as the attention scores output corresponding to each weighted ship wake feature.

[0099] Generally, after obtaining the weighted feature vector of the ship's wake, it is first fed into a learnable transformation structure to assess the relative importance of each feature in the current context. This structure typically contains a set of trainable parameter matrices and a nonlinear activation function. Its working principle is to use these parameters to perform linear combinations and mathematical transformations on the input feature vector, and then introduce nonlinear mapping capabilities through the activation function. After this series of calculations, each input feature dimension is transformed and output as a scalar value, called the initial importance score. This score initially expresses the original importance of the corresponding feature before standardization comparison.

[0100] Generally, since the initial importance scores are in an unscaled, raw state with potentially large differences in absolute values, they cannot be directly used to measure the relative importance between features. Therefore, these scores need to be fed into a normalization layer. This layer processes all the initial importance scores using a specific normalization function (such as the Softmax function). The core function of this process is to map all scores to a positive interval with a fixed sum (e.g., a sum of 1), thereby transforming these raw scores into a vector that conforms to a probability distribution. This transformed vector is called the normalized attention weights.

[0101] Generally, the attention weights obtained after normalization represent the proportion of "attention" the model allocates to each wake feature in this specific input scenario. Therefore, this weight vector is directly used as the final output of the feature attention module, which is the formal attention score corresponding one-to-one with each weighted ship wake feature. These scores will serve as key metadata, passed to the subsequent speed inversion model to guide the model in dynamically adjusting its internal attention to and utilization of different feature dimensions during this calculation.

[0102] The technical solution of this invention acquires and identifies remote sensing images of the target vessel, and locates the vessel wake recognition area and water condition recognition area based on the image region where the target vessel is located. Image recognition is performed in the two areas to obtain vessel wake features and water condition features. Simultaneously, the physical dimensions of the vessel are calculated based on the bounding rectangle and pixel size of the target vessel in the image, combined with the ground sampling distance of the image. The wake length and width are then normalized to eliminate feature deviations caused by differences in vessel size. Finally, the dynamic weights of each wake feature are determined based on the water condition features. The normalized wake features and dynamic weights are combined and input into a pre-trained speed inversion model to output a predicted speed. This solves the technical challenge of real-time and accurate speed inversion for target vessels without Automatic Identification System (AIS) signals or lacking continuous frame sequences under complex and variable sea conditions. It achieves high-precision and robust speed measurement based on only a single frame image, effectively overcoming differences in vessel size and environmental interference.

[0103] To facilitate understanding, the specific application scenarios applicable to each embodiment of the present invention are described below. In this specific embodiment, for sea areas without Automatic Identification System (AIS) signals and with complex and variable imaging conditions, the present invention designs a complete high-precision inversion scheme for single-frame images of ship speed based on water condition adaptive feature weighting and attention guidance.

[0104] Specifically, Figure 3 Here is a flowchart of a method for speed inversion based on single-frame ship imagery, as shown below. Figure 3As shown, the process begins with acquiring a single-frame satellite image of the target sea area. First, the original satellite images are preprocessed to remove noise and enhance image quality. Then, the process moves to the core feature extraction stage. Based on the preprocessed image, key geometric and textural features of the ship's wake are identified and extracted, specifically including wake length, wake width, spread angle, and texture clarity representing water surface texture fluctuations. These extracted multi-dimensional features are combined to form a feature vector. After extracting the wake feature vector, to achieve high-precision speed inversion under supervised learning, a mapping relationship needs to be established between the extracted objective physical features and the ship's actual motion state. Specifically, the extracted four-dimensional feature vector is read from a structured data table (such as a CSV file) containing image names and corresponding real speed labels. Data pairing and merging operations are performed using the image name as the unique matching key, thereby constructing a standard model training dataset. This dataset is further input into a random forest regression model for iterative training, and the model's generalization ability and stability are comprehensively evaluated through a cross-validation mechanism. Finally, using a mature model that has undergone rigorous evaluation and verification, feature extraction and model inference are performed on satellite images of new input samples to achieve automated and high-precision inversion calculation of ship speed.

[0105] Figure 4 This is a schematic diagram for extracting ship wake features, such as... Figure 4 As shown in the diagram, this schematic visually illustrates the morphology of a ship's wake during navigation and its corresponding four core quantitative features. In this embodiment, the ship's body is represented by a solid red graphic, and the wake generated by the ship during navigation is represented by a white-filled wedge-shaped area. To establish a precise mapping relationship between wake features and ship speed, the process focuses on extracting three key geometric parameters of the wake: first, the wake length, which is the horizontal distance between the furthest boundaries extending from the stern to both sides; second, the wake width, which is the maximum lateral span at the stern measured perpendicular to the wake centerline; and third, the diffusion angle, which is the angle between the two boundary lines of the wake at the stern. In addition, to more comprehensively characterize the physical properties of the wake, a texture feature called texture clarity is introduced. In the specific embodiment of this invention, this feature is represented by the variance of the second derivative of the image, used to quantify the intensity of the water surface grayscale changes within the wake area, thereby effectively reflecting the turbulence intensity information of the water flow.

[0106] Example 3

[0107] Figure 5 Embodiment 3 of the present invention provides a speed inversion device based on a single frame of ship imagery, such as... Figure 5 As shown, the device includes:

[0108] The image data acquisition module 510 is used to acquire remote sensing images of the target ship to be subjected to speed inversion and to identify the target ship in the remote sensing images.

[0109] The positioning module 520 is used to locate the ship wake recognition area and the water condition recognition area in the remote sensing image of the target ship based on the image area where the target ship is located.

[0110] The feature acquisition module 530 is used to perform image recognition in the ship wake recognition area and the water condition recognition area respectively, and acquire ship wake features and water condition features that match the current driving scene of the target ship. Among them, the ship wake features include wake length, wake width, diffusion angle and texture clarity, and the water condition features include water surface roughness, water surface brightness uniformity and water surface high frequency energy ratio.

[0111] The speed prediction module 540 is used to determine the dynamic weights corresponding to each of the ship wake features based on the water condition characteristics, and input each ship wake feature and the matched dynamic weights into the pre-trained speed inversion model to obtain the predicted speed of the target ship in the target ship remote sensing image output by the speed inversion model.

[0112] The technical solution of this invention involves acquiring remote sensing images of the target vessel to be inverted in terms of speed, identifying the target vessel within them, locating the vessel wake recognition region and water condition recognition region in the remote sensing image based on the image region where the target vessel is located, performing image recognition in the wake recognition region and water condition recognition region respectively, acquiring various vessel wake features and water condition features matching the current navigation scene of the target vessel, determining dynamic weights corresponding to each vessel wake feature based on each water condition feature, and inputting each vessel wake feature and the matching dynamic weights into a pre-trained speed inversion model to obtain the predicted speed of the target vessel output by the speed inversion model. This achieves automated and high-precision speed inversion of vessels under complex sea conditions, effectively overcoming the dependence on continuous frame images and prior information about the vessel, and improving the robustness and practicality of the speed inversion method.

[0113] Based on the above embodiments, the feature acquisition module 530 is specifically used for:

[0114] Edge detection and contour analysis are performed on the ship wake recognition area to determine the maximum connected region corresponding to the target ship wake;

[0115] Calculate the minimum bounding rectangle of the maximum connected region, take the length of the long side of the minimum bounding rectangle as the tail length, the length of the short side of the minimum bounding rectangle as the tail width, and the rotation angle of the minimum bounding rectangle as the diffusion angle.

[0116] The gradient variance of the image corresponding to the largest connected region is calculated based on the Laplacian operator, and the normalized gradient variance is used as the texture sharpness.

[0117] Based on the above embodiments, the feature acquisition module 530 is specifically used for:

[0118] The Laplacian operator is used to perform convolution processing on the water condition identification region to obtain the response map, and the response variance of the response map is calculated.

[0119] The Sobel operator is used to perform convolution processing on the water condition identification region to obtain the gradient map, and the gradient mean of the gradient map is calculated.

[0120] The normalized value obtained by dividing the response variance by the gradient mean is used as the water surface roughness.

[0121] Calculate the mean gray value and standard deviation of each pixel in the water condition identification area, and calculate the normalized value obtained by dividing the mean gray value by the standard deviation of gray value, which is used as the water surface brightness uniformity.

[0122] A two-dimensional fast Fourier transform is performed on the water condition identification area to obtain a frequency domain map, and low-frequency and high-frequency intervals are identified in the frequency domain map;

[0123] Calculate the low-frequency energy value in the low-frequency range and the high-frequency energy value in the high-frequency range, and calculate the proportion of high-frequency energy on the water surface based on the low-frequency energy value and the high-frequency energy value.

[0124] Based on the above embodiments, the speed prediction module 540 is specifically used for:

[0125] Based on each water condition feature and the fusion weight matched with each water condition feature, the target sea state level matched for the current driving scenario is calculated.

[0126] Based on the target sea state level, a search is performed in the pre-weighted allocation strategy to obtain a target weighted allocation strategy that matches the target sea state level, and dynamic weights corresponding to each of the ship wake features are obtained from the target weighted allocation strategy.

[0127] Based on the above embodiments, the speed inversion device based on a single frame of ship imagery may further include:

[0128] The pixel scale measurement module is used to obtain the bounding rectangle of the target ship in the remote sensing image of the target ship according to the image region where the target ship is located, and to obtain the pixel width and pixel length of the bounding rectangle before inputting the wake features of each ship and the matched dynamic weights into the pre-trained speed inversion model.

[0129] The ship physical size inversion module is used to calculate the physical size of the target ship based on the ground sampling distance matched with the remote sensing image of the target ship, the pixel width, and the pixel length.

[0130] The feature scale normalization module is used to perform scale normalization processing on the wake length and wake width according to the physical dimensions of the target ship, so as to eliminate feature deviations caused by differences in ship size.

[0131] Based on the above embodiments, the speed prediction module 540 may further include:

[0132] The dynamic weighted feature vector generation submodule is used to apply matching dynamic weights to each of the ship wake features and perform weighted calculations to obtain the weighted ship wake feature vector.

[0133] The attention module is used to input the weighted ship wake feature vector into a feature attention module to calculate the attention score corresponding to each weighted ship wake feature.

[0134] The model calculation submodule is used to input the weighted ship wake feature vector and its matching attention score into the pre-trained speed inversion model, wherein the speed inversion model is a random forest regression model, and its multiple decision trees perform differentiated sensitivity learning on different feature dimensions based on the attention score when splitting.

[0135] The predicted speed output submodule is used to obtain the predicted speed of the target ship based on the output of all decision trees in the random forest regression model.

[0136] Based on the above embodiments, attention is focused on the molecular module, specifically used for:

[0137] The weighted ship wake feature vector is input into a nonlinear transformation layer, and the initial importance score of each weighted ship wake feature is calculated through a learnable weight matrix and activation function.

[0138] The initial importance scores of all weighted ship wake features are input into a normalization layer. The initial importance scores are converted into a probability distribution form through a normalization function to obtain the normalized attention weights.

[0139] The normalized attention weights are used as the attention scores output corresponding to each weighted ship wake feature.

[0140] The speed inversion device based on single-frame ship image provided in this embodiment of the invention can execute the speed inversion method based on single-frame ship image provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0141] The collection, storage, use, processing, transmission, provision, and disclosure of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0142] Example 4

[0143] Figure 6 A schematic diagram of an electronic device 10, which can be used to implement embodiments of the present invention, is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0144] like Figure 6 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0145] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0146] Processor 11 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, central processing unit (CPU), graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as performing a speed inversion method based on a single frame of ship imagery as described in any embodiment of the present invention, i.e.:

[0147] Acquire remote sensing images of the target vessel to be used for speed inversion, and identify the target vessel in the remote sensing images;

[0148] Based on the image region where the target vessel is located, the vessel wake recognition region and the water condition recognition region are located in the remote sensing image of the target vessel.

[0149] Image recognition is performed in the ship wake recognition area and the water condition recognition area respectively to obtain the ship wake features and water condition features that match the current driving scene of the target ship. Among them, the ship wake features include wake length, wake width, diffusion angle and texture clarity, and the water condition features include water surface roughness, water surface brightness uniformity and water surface high frequency energy ratio.

[0150] Based on the characteristics of each water condition, dynamic weights corresponding to each of the ship wake features are determined, and each ship wake feature and the matched dynamic weights are input together into a pre-trained speed inversion model to obtain the predicted speed of the target ship in the remote sensing image of the target ship output by the speed inversion model.

[0151] In some embodiments, a speed inversion method based on a single-frame ship image, as described in any one of the embodiments of the present invention, can be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the speed inversion method based on a single-frame ship image, as described above, can be performed. Alternatively, in other embodiments, processor 11 can be configured by any other suitable means (e.g., by means of firmware) to perform the speed inversion method based on a single-frame ship image, as described in any one of the embodiments of the present invention.

[0152] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in a computer program product comprising one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0153] The computer programs included in the computer program product for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, partially on a remote machine as a standalone software package, or entirely on a remote machine or server.

[0154] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0155] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0156] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0157] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0158] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0159] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for speed inversion based on single-frame ship imagery, characterized in that, The method includes: Acquire remote sensing images of the target vessel to be used for speed inversion, and identify the target vessel in the remote sensing images; Based on the image region where the target vessel is located, the vessel wake recognition region and the water condition recognition region are located in the remote sensing image of the target vessel. Image recognition is performed in the ship wake recognition area and the water condition recognition area respectively to obtain the ship wake features and water condition features that match the current driving scene of the target ship. Among them, the characteristics of ship wakes include wake length, wake width, diffusion angle and texture clarity, and the characteristics of water conditions include water surface roughness, water surface brightness uniformity and water surface high frequency energy ratio. Based on the characteristics of each water condition, dynamic weights corresponding to each of the ship wake features are determined, and each ship wake feature and the matched dynamic weights are input together into a pre-trained speed inversion model to obtain the predicted speed of the target ship in the target ship remote sensing image output by the speed inversion model. Specifically, image recognition is performed in the ship wake recognition area to obtain the wake features of each ship that match the current navigation scene of the target ship, including: Edge detection and contour analysis are performed on the ship wake recognition area to determine the maximum connected region corresponding to the target ship wake; Calculate the minimum bounding rectangle of the maximum connected region, take the length of the long side of the minimum bounding rectangle as the tail length, the length of the short side of the minimum bounding rectangle as the tail width, and the rotation angle of the minimum bounding rectangle as the diffusion angle. The gradient variance of the image corresponding to the largest connected region is calculated based on the Laplacian operator, and the normalized gradient variance is used as the texture sharpness. Specifically, image recognition is performed in the water condition identification area to obtain various water condition features that match the current navigation scene of the target vessel, including: The Laplacian operator is used to perform convolution processing on the water condition identification region to obtain the response map, and the response variance of the response map is calculated. The Sobel operator is used to perform convolution processing on the water condition identification region to obtain the gradient map, and the gradient mean of the gradient map is calculated. The normalized value obtained by dividing the response variance by the gradient mean is used as the water surface roughness. Calculate the mean gray value and standard deviation of each pixel in the water condition identification area, and calculate the normalized value obtained by dividing the mean gray value by the standard deviation of gray value, which is used as the water surface brightness uniformity. A two-dimensional fast Fourier transform is performed on the water condition identification area to obtain a frequency domain map, and low-frequency and high-frequency intervals are identified in the frequency domain map; Calculate the low-frequency energy value in the low-frequency range and the high-frequency energy value in the high-frequency range, and calculate the proportion of high-frequency energy on the water surface based on the low-frequency energy value and the high-frequency energy value.

2. The method according to claim 1, characterized in that, Based on the characteristics of each water condition, determine the dynamic weights corresponding to each of the aforementioned ship wake characteristics, including: Based on each water condition feature and the fusion weight matched with each water condition feature, the target sea state level matched for the current driving scenario is calculated. Based on the target sea state level, a search is performed in the pre-weighted allocation strategy to obtain a target weighted allocation strategy that matches the target sea state level, and dynamic weights corresponding to each of the ship wake features are obtained from the target weighted allocation strategy.

3. The method according to any one of claims 1-2, characterized in that, Before inputting the wake features of each ship and the matched dynamic weights into the pre-trained speed inversion model, the following steps are also included: Based on the image region where the target vessel is located, obtain the bounding rectangle of the target vessel in the remote sensing image of the target vessel, and obtain the pixel width and pixel length of the bounding rectangle; The physical dimensions of the target ship are calculated based on the ground sampling distance matched with the remote sensing image of the target ship, the pixel width, and the pixel length. Based on the physical dimensions of the target vessel, the wake length and wake width are normalized to eliminate feature deviations caused by differences in vessel dimensions.

4. The method according to any one of claims 1-2, characterized in that, The wake features of each vessel and their matched dynamic weights are input into a pre-trained speed inversion model to obtain the predicted speed of the target vessel in the remote sensing image of the target vessel output by the speed inversion model, including: Each ship wake feature is assigned a matching dynamic weight, and a weighted calculation is performed to obtain a weighted ship wake feature vector. The weighted ship wake feature vector is input into a feature attention module to calculate the attention score corresponding to each weighted ship wake feature. The weighted ship wake feature vector and its matching attention score are input together into the pre-trained speed inversion model, wherein the speed inversion model is a random forest regression model, and its multiple decision trees perform differentiated sensitivity learning on different feature dimensions based on the attention score when splitting. The predicted speed of the target ship is obtained based on the output of all decision trees in the random forest regression model.

5. The method according to claim 4, characterized in that, The weighted ship wake feature vector is input into a feature attention module to calculate the attention score corresponding to each weighted ship wake feature, including: The weighted ship wake feature vector is input into a nonlinear transformation layer, and the initial importance score of each weighted ship wake feature is calculated through a learnable weight matrix and activation function. The initial importance scores of all weighted ship wake features are input into a normalization layer. The initial importance scores are converted into a probability distribution form through a normalization function to obtain the normalized attention weights. The normalized attention weights are used as the attention scores output corresponding to each weighted ship wake feature.

6. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the speed inversion method based on single-frame ship imagery as described in any one of claims 1-5.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute the speed inversion method based on a single-frame ship image as described in any one of claims 1-5.

8. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the speed inversion method based on a single-frame ship image according to any one of claims 1-5.

Images