A ship identification method and system based on multi-source data fusion

Abstract

This invention provides a ship identification method and system based on multi-source data fusion. The method includes: acquiring ship navigation data; calculating ship navigation feature data based on the navigation data; processing high-resolution range image data of the ship based on a preset model to obtain a high-resolution range image feature dataset; extracting feature datasets from the synthetic aperture radar remote sensing image; fusing the above data to obtain a target feature dataset; and identifying the target feature dataset based on a ship classification and recognition model to obtain a ship identification result. This invention achieves high-precision ship identification by fusing multi-source ship data.

Description

Technical Field

[0001] This invention relates to the field of ship identification technology, and specifically to a ship identification method and system based on multi-source data fusion. Background Technology

[0002] Currently, ship type identification is mainly achieved through a single data source, such as ship identification based on Synthetic Aperture Radar (SAR) images. While SAR image-based ship identification has the advantages of being unaffected by weather and enabling continuous observation around the clock, it is also susceptible to sea clutter, coastal port facilities, and islands and reefs, resulting in low target detection accuracy. Ship identification based on Automatic Identification System (AIS) can identify ships based on their static information, but AIS data is prone to missing data and errors.

[0003] When sailing at sea, using single-source sensors for environmental perception is not ideal, as each sensor has its own advantages. However, the perception effect is somewhat insufficient. The marine environment is very complex, and the perception effect will drop sharply, especially when encountering extreme weather such as heavy fog or heavy rain.

[0004] Therefore, there is an urgent need for a ship identification method and system based on multi-source data fusion to solve the above problems. Summary of the Invention

[0005] In view of this, it is necessary to provide a ship identification method and system based on multi-source data fusion to solve the problems of poor ship perception and low identification accuracy in traditional methods.

[0006] To address the aforementioned technical problems, this invention provides a ship identification method based on multi-source data fusion, comprising:

[0007] Acquire the ship's navigation data, and calculate the ship's navigation characteristic data based on the navigation data;

[0008] High-resolution range image data of ships is acquired, and the high-resolution range image data of ships is processed based on a preset model to obtain a high-resolution range image feature dataset;

[0009] Extracting feature datasets from synthetic aperture radar remote sensing images;

[0010] The target feature dataset is obtained by fusing the navigation feature data of ships, the high-resolution range image feature dataset, and the feature dataset of synthetic aperture radar remote sensing images.

[0011] The target feature dataset is input into a pre-defined classification model for ship classification and recognition training, resulting in a ship classification and recognition model.

[0012] The target feature dataset is identified based on a ship classification and recognition model to obtain ship identification results.

[0013] In one possible implementation, before acquiring the ship's navigation data, the following steps are also included:

[0014] Acquire synthetic aperture radar (SAR) remote sensing images and identify ship types based on the YoloV5 algorithm to obtain ship identification results.

[0015] In one possible implementation, synthetic aperture radar (SAR) remote sensing images are acquired, and ship types are identified based on the YOLOv5 algorithm to obtain ship identification results, including:

[0016] Acquire synthetic aperture radar remote sensing images;

[0017] Image cropping of synthetic aperture radar remote sensing images is performed based on the sliding window cropping method. The cropped synthetic aperture images are then deleted, deduplicated, and classified.

[0018] The synthetic aperture radar remote sensing images that have been deleted, deduplicated, and classified are labeled to obtain the first target dataset, which includes a training dataset and a test dataset.

[0019] The training dataset is imported into the YOLOv5 algorithm for training, resulting in a ship classification and detection model.

[0020] The test dataset is identified using a ship classification and detection model to obtain ship identification results.

[0021] In one possible implementation, acquiring the ship's navigation data and calculating the ship's navigation characteristic data based on the navigation data includes:

[0022] Initial data about a ship is acquired based on its Automatic Identification Unit (AIU).

[0023] The initial data of the ship is preprocessed to obtain the ship's trajectory data;

[0024] The ship's navigation trajectory data is normalized to extract the ship's navigation characteristic data.

[0025] In one possible implementation, the navigation characteristic data includes at least one of the following: ship speed, ship position, ship course, ship acceleration, and rate of change of ship course.

[0026] In one possible implementation, high-resolution range image data of the ship is acquired, and the high-resolution range image data is processed based on a preset model to obtain a high-resolution range image feature dataset, including:

[0027] Acquire high-resolution range image data of ships;

[0028] The initial data of the high-resolution range image of the ship is extracted. The initial data of the high-resolution range image of the ship includes the line-of-sight distance, amplitude value and phase value at different azimuth angles.

[0029] Based on the azimuth angle of the initial data of the high-resolution range image data of the ship, the high-resolution range image data of the ship is divided into several regions;

[0030] Match the same range gate to the high-resolution range image data of ships within each azimuth angle, and perform equidistant interpolation on the actual range unit of each high-resolution range image data of ships.

[0031] The echo data features of high-resolution range image data of ships at different realization distances are obtained, and the echo data features are normalized to obtain a high-resolution range image feature dataset.

[0032] In one possible implementation, acquiring echo data features of high-resolution range image data of ships at different implementation distances includes:

[0033] Extract high-resolution range image data of the ship within a preset azimuth angle;

[0034] The average value of the echo data for each azimuth angle within the preset azimuth angle sampling interval is calculated to obtain the echo data characteristics of the ship's high-resolution range image data at different realization distances.

[0035] In one possible implementation, a feature dataset of synthetic aperture radar remote sensing images is extracted, including:

[0036] Extracting gradient orientation histograms of local regions from synthetic aperture radar remote sensing images;

[0037] Convert synthetic aperture radar remote sensing images into an image matrix;

[0038] Perform gamma processing on the pixel values ​​in the image matrix;

[0039] Calculate the horizontal and vertical gradients of pixels in the image matrix after gamma processing to obtain the gradient direction and gradient magnitude of each pixel.

[0040] The overall features of the synthetic aperture radar remote sensing image are extracted. Based on the overall features of the synthetic aperture radar remote sensing image, the horizontal and vertical gradients of the pixels, and the gradient direction and magnitude of the pixels, the initial feature values ​​of the synthetic aperture radar remote sensing image are obtained.

[0041] The initial feature values ​​of the synthetic aperture radar (SAR) remote sensing image are normalized to obtain the feature dataset of the SAR remote sensing image.

[0042] In one possible implementation, the target feature dataset is input into a classification model for ship classification and recognition training, resulting in a ship classification and recognition model, including:

[0043] The target feature dataset is divided into a target training set and a test dataset according to a preset ratio;

[0044] The classification model is trained based on the target training set to obtain the ship classification and recognition model.

[0045] To address the aforementioned problems, the present invention also provides a ship identification system based on multi-source data fusion, comprising:

[0046] The navigation data acquisition module is used to acquire the navigation data of the vessel and calculate the navigation characteristic data of the vessel based on the navigation data;

[0047] The high-resolution range image module for ships is used to acquire high-resolution range image data of ships and process the high-resolution range image data of ships based on a preset model to obtain a high-resolution range image feature dataset.

[0048] The synthetic aperture radar remote sensing image extraction module is used to extract the feature dataset of synthetic aperture radar remote sensing images;

[0049] The fusion processing module is used to fuse the navigation feature data of ships, the high-resolution range image feature dataset, and the feature dataset of synthetic aperture radar remote sensing images to obtain the target feature dataset.

[0050] The training module is used to input the target feature dataset into a preset classification model for ship classification and recognition training, thereby obtaining a ship classification and recognition model.

[0051] The processing module is used to identify the target feature dataset based on the ship classification and recognition model and obtain the ship identification results.

[0052] The beneficial effects of the above embodiments are as follows: by acquiring ship navigation feature data; processing the ship high-resolution range image data based on a preset model to obtain a high-resolution range image feature dataset; extracting feature datasets from synthetic aperture radar remote sensing images; fusing the above data to obtain a target feature dataset; and identifying the target feature dataset based on a ship classification and recognition model to obtain ship identification results. This invention achieves high-precision ship identification by fusing multi-source ship data through data association. Attached Figure Description

[0053] 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.

[0054] Figure 1 This is a flowchart illustrating an embodiment of a ship identification method based on multi-source data fusion provided by the present invention.

[0055] Figure 2 This is a flowchart illustrating an embodiment of the method for obtaining ship identification results based on synthetic aperture radar remote sensing images provided by the present invention.

[0056] Figure 3 Provided by the present invention Figure 1 A flowchart illustrating an embodiment of the method for obtaining a high-resolution range image feature dataset in step S103;

[0057] Figure 4 This is a flowchart illustrating another embodiment of a ship identification method based on multi-source data fusion provided by the present invention.

[0058] Figure 5 This is a schematic diagram of an embodiment of a ship identification method based on multi-source data fusion provided by the present invention. Detailed Implementation

[0059] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0060] Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor systems and / or microcontroller systems.

[0061] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0062] This invention provides a ship identification method and system based on multi-source data fusion, which will be described below.

[0063] Figure 1 This is a flowchart illustrating an embodiment of a ship identification method based on multi-source data fusion provided by the present invention.

[0064] Reference Figure 1 This invention provides a ship identification method based on multi-source data fusion, comprising:

[0065] S101. Obtain the ship's navigation data and calculate the ship's navigation characteristic data based on the navigation data;

[0066] S102. Obtain high-resolution range image data of ships, and process the high-resolution range image data of ships based on a preset model to obtain a high-resolution range image feature dataset.

[0067] S103. Extract the feature dataset of synthetic aperture radar remote sensing images;

[0068] S104. The navigation feature data of the ship, the high-resolution range image feature dataset, and the feature dataset of the synthetic aperture radar remote sensing image are fused to obtain the target feature dataset.

[0069] S105. Input the target feature dataset into the preset classification model to train the ship classification and recognition model, and obtain the ship classification and recognition model.

[0070] S106. Based on the ship classification and recognition model, identify the target feature dataset and obtain the ship recognition results.

[0071] The beneficial effects of the above embodiments are: acquiring ship navigation feature data; processing the ship's high-resolution range image data based on a preset model to obtain a high-resolution range image feature dataset; extracting feature datasets from synthetic aperture radar remote sensing images; fusing the above data to obtain a target feature dataset; and identifying the target feature dataset based on a ship classification and recognition model to obtain ship identification results. This invention achieves high-precision ship identification by fusing multi-source ship data through data association.

[0072] Synthetic Aperture Radar (SAR) images are unaffected by weather, enabling continuous observation around the clock. However, they are susceptible to sea clutter, coastal port facilities, and islands, resulting in lower target detection accuracy. While Automatic Identification System (AIS) has lower accuracy than SAR, it can identify ships based on their static information and is less affected by sea clutter, coastal port facilities, and islands. High-resolution range profile (HRRP) of marine radar can acquire information such as the linear distance, amplitude, and phase of ships, thus supplementing the data of SAR and AIS.

[0073] In one embodiment, before acquiring the ship's navigation data, the method further includes:

[0074] Acquire synthetic aperture radar (SAR) remote sensing images and identify ship types based on the YoloV5 algorithm to obtain ship identification results.

[0075] Understandably, when there is no interference from sea clutter, coastal port facilities, islands, or other surrounding environment on the identification of synthetic aperture radar (SAR) remote sensing images, ships can be identified using SAR remote sensing images. However, when sea clutter, coastal port facilities, islands, or other surrounding environment interfere with ship identification under SAR remote sensing images, it is necessary to combine data fusion of automatic identification units (AIS) and high-resolution range images from marine radar to identify and judge ships.

[0076] Figure 2 This is a flowchart illustrating an embodiment of the method for obtaining ship identification results based on synthetic aperture radar remote sensing images provided by the present invention.

[0077] Reference Figure 2 In one embodiment, a synthetic aperture radar (SAR) remote sensing image is acquired, and the ship type is identified based on the YOLOv5 algorithm. The ship identification result includes:

[0078] S201. Acquire synthetic aperture radar remote sensing images;

[0079] S202. Based on the sliding window cropping method, perform image cropping on synthetic aperture radar remote sensing images, and perform deletion, deduplication and classification processing on the cropped synthetic aperture images;

[0080] S203. Label the synthetic aperture radar remote sensing images that have been deleted, deduplicated, and classified to obtain the first target dataset, which includes a training dataset and a test dataset.

[0081] S204. Import the training dataset into the YOLOV5 algorithm for training to obtain the ship classification and detection model;

[0082] S205. Based on the ship classification and detection model, identify the test dataset and obtain the ship identification results.

[0083] The process involves acquiring SAR data of ships using the GF-3 satellite. GF-3 is my country's first C-band multi-polarization SAR imaging satellite with a resolution of 1 meter, an observation swath width of 10-650 km, and a spatial resolution of 1-500 m. SAR image data is then cropped using a sliding window method. By setting the window size and overlap rate, an image of the desired size can be obtained, for example, an image with 512 pixels * 512 pixels. The cropped images are then processed through deletion, deduplication, and classification. Deletion removes images where ship targets are cropped, and also removes images containing multiple ship targets. Deduplication ensures that each image contains only one ship and no duplicates. The images of ships are repeated, and the processed ship target images are classified, numbered, and saved. Ship targets are labeled on the cropped classified images to obtain the first target dataset. The labeling method can use the LabelImg annotation tool to mark the minimum bounding box of the ship target, and the labeling method used is the rectangular box annotation method. Then, the first target dataset is divided into training dataset and test dataset according to a preset ratio. Then, the training dataset is imported into the training module of the YOLOv5 model for training. The number of training epochs can be set to 300, and the batch-size of each data reading is set according to the computer performance (here it is set to 3). Other parameters can be set according to the default parameters, thus obtaining the ship classification and detection model.

[0084] In one embodiment, acquiring ship's navigation data and calculating the ship's navigation characteristic data based on the navigation data includes:

[0085] Initial data about a ship is acquired based on its Automatic Identification Unit (AIU).

[0086] Preprocess the initial data of the ship to extract its trajectory data;

[0087] The ship's trajectory data is normalized to extract its navigation characteristic data.

[0088] Ship navigation characteristic data include the average speed, standard deviation, maximum speed, minimum speed, and quantiles.

[0089] The formula for calculating the average speed is:

[0090] Where D and T represent the distance and time difference between the first and last points of the ship's AIS data sorted by time.

[0091] Formula for calculating the standard deviation of velocity

[0092] Where n represents the number of AIS data points for the ship (n≥2), v i Let i be the speed of the ship at the AIS track point (i = 1, 2, ..., n). This represents the average ship speed of the ship's AIS trajectory.

[0093] Formulas for calculating maximum and minimum speeds:

[0094] Where n represents the number of AIS data points for the ship (n≥2), v i Let i be the speed of the ship at the AIS track point (i = 1, 2, ..., n).

[0095] Quantile calculation formula:

[0096] Where n represents the number of AIS data points for the ship (n≥2), Q is the percentile, R is the index of percentile Q, k is the integer down from the index of percentile Q, and v k and v k+1 This refers to the speeds at the k-th and k+1-th points after sorting the AIS data points by speed from smallest to largest.

[0097] Among them, AIS data is preprocessed to extract acceleration characteristics of ships, including mean, standard deviation, maximum, minimum, quantile, etc.

[0098] Acceleration calculation formula:

[0099] Where n represents the number of AIS data points for the ship (n≥2), v i and v i+1 The speeds at positions i and i+1 in the ship's AIS data, sorted by time; t i and t i+1 The times at positions i and i+1 in the AIS data of the ship, sorted by time (i = 1, 2, ..., n-1).

[0100] Formula for calculating average acceleration

[0101] Where n represents the number of AIS data points for the ship (n≥2), a i Let be the acceleration (i = 1, 2, ..., n-1).

[0102] Furthermore, the formulas for calculating the standard deviation, maximum value, minimum value, and quantiles of acceleration can be based on the aforementioned velocity formulas. The heading characteristics of AIS data include the mean, standard deviation, maximum value, minimum value, and quantiles. The characteristics of the heading rate of change of AIS data also include the mean, standard deviation, maximum value, minimum value, and quantiles. The formulas for calculating the heading characteristics and the heading rate of change of AIS data can refer to the velocity calculation formulas, which will not be elaborated here.

[0103] The navigation characteristic data of ships is normalized to extract the navigation characteristic data. The normalization calculation formula is as follows:

[0104]

[0105] Where x represents the value of each feature class, x min Let x represent the minimum value of each feature class. max This represents the maximum value of each type of feature.

[0106] In one embodiment, the navigation characteristic data includes at least one of ship speed, ship position, ship heading, ship acceleration, and rate of change of ship heading.

[0107] Figure 3 Provided by the present invention Figure 1 A flowchart illustrating an embodiment of the method for obtaining a high-resolution distance image feature dataset in step S103.

[0108] Reference Figure 3 In one embodiment, high-resolution range image data of a ship is acquired, and the high-resolution range image data of the ship is processed based on a preset model to obtain a high-resolution range image feature dataset, including:

[0109] S301. Acquire high-resolution range image data of ships;

[0110] S302. Extract the initial data of the high-resolution range image data of the ship. The initial data of the high-resolution range image data of the ship includes the line-of-sight distance, amplitude value and phase value at different azimuth angles.

[0111] S303. Based on the azimuth angle of the initial data of the high-resolution range image data of the ship, the high-resolution range image data of the ship is divided into several regions;

[0112] S304. Match the same range gate to the high-resolution range image data of the ship in each azimuth angle, and perform equidistant interpolation on the realization range unit of each high-resolution range image data of the ship.

[0113] S305. Obtain echo data features of high-resolution range image data of ships at different realization distances, and normalize the echo data features to obtain a high-resolution range image feature dataset.

[0114] When using HRRP for target classification and recognition, reducing the azimuth sensitivity of the target distance image can improve the stability of HRRP data.

[0115] HRRP data can exhibit significantly different responses under large azimuth variations. Therefore, for the same target, a large number of feature vectors with azimuth as the index are needed to represent it in the target database. To reduce the number of feature vectors used to represent targets in the database, for each target, the echo data is divided into several regions based on azimuth (the degree of division can be determined according to the recognition rate requirements). The amplitude of each range cell on the range image is equal to the coherent sum of the echoes from all scattering centers within that range cell. When the target attitude changes relative to the radar line of sight, HRRP amplitude fluctuations will occur. Signal multi-resolution decomposition based on wavelet theory can effectively solve this problem. Multi-resolution decomposition approximates the signal at different resolutions; in fact, it is the only orthogonal decomposition with minimum error in the low-dimensional space of the signal at higher resolution. Because it is an orthogonal decomposition, multi-resolution decomposition has a good decorrelation effect. This invention uses multi-resolution decomposition to achieve efficient feature extraction and feature dimension compression.

[0116] In one embodiment, acquiring echo data features of high-resolution range image data of a ship at different implementation distances includes:

[0117] Extract high-resolution range image data of the ship within a preset azimuth angle;

[0118] The average value of the echo data for each azimuth angle within the preset azimuth angle sampling interval is calculated to obtain the echo data characteristics of the ship's high-resolution range image data at different line-of-sight distances.

[0119] In one embodiment, the feature dataset of the synthetic aperture radar remote sensing image is extracted, including:

[0120] Extracting gradient orientation histograms of local regions from synthetic aperture radar remote sensing images;

[0121] Convert synthetic aperture radar remote sensing images into an image matrix;

[0122] Perform gamma processing on the pixel values ​​in the image matrix;

[0123] Calculate the horizontal and vertical gradients of pixels in the image matrix after gamma processing to obtain the gradient direction and gradient magnitude of each pixel.

[0124] The overall features of the synthetic aperture radar remote sensing image are extracted. Based on the overall features of the synthetic aperture radar remote sensing image, the horizontal and vertical gradients of the pixels, and the gradient direction and magnitude of the pixels, the initial feature values ​​of the synthetic aperture radar remote sensing image are obtained.

[0125] The initial feature values ​​of the synthetic aperture radar (SAR) remote sensing image are normalized to obtain the feature dataset of the SAR remote sensing image.

[0126] Histogram of Oriented Gradients (HOG) is a feature descriptor used in computer vision and image processing for object detection. HOG features are constructed by calculating and statistically analyzing the gradient orientation histograms of local regions of an image. Since SAR images are grayscale images, there is no need to perform grayscale processing on the image; the image is directly converted into an image matrix, where the values ​​in the matrix correspond to the grayscale value of each pixel.

[0127] Gamma processing is performed on each pixel value A(x,y) in the image matrix A to increase or decrease the overall brightness of the image, adjust the image contrast, and reduce the image non-uniformity.

[0128] The formula for Gamma processing is: A′(x,y)=A(x,y) Gamma ;

[0129] Where A′(x,y) represents the image pixel value after Gamma processing, and A(x,y) represents the image pixel value before Gamma processing. Gamma is a correction value, for example, Gamma = 0.5.

[0130] Gradient calculation: The horizontal and vertical gradients of each pixel in the image matrix after Gamma processing are calculated to obtain the gradient direction and magnitude of each pixel, thus obtaining the contour information of the image. The image can be filtered using a horizontal template K = [1,0,1] and its transpose to complete the calculation of the horizontal and vertical gradients.

[0131] The formula for calculating the gradient in the horizontal direction at pixel A′(x,y) is: gx=A′(x+1,y)A′(x1,y).

[0132] The gradient in the vertical direction at pixel A′(x,y) is calculated as follows: gy=A′(x,y+1)A′(x,y1).

[0133] The formula for calculating the gradient magnitude at pixel A′(x,y) is:

[0134] The formula for calculating the gradient direction at pixel A′(x,y) is:

[0135] The process of constructing an oriented gradient histogram for each cell unit of a synthetic aperture radar remote sensing image includes: decomposing the image into several cells, where a cell is the smallest statistical and processing unit in the image decomposition process of the algorithm. Each cell unit contains m×m pixels. The gradient direction is quantized into 9 histogram channels, representing 9 different gradient directions. Based on these 9 histogram channels, each pixel in the cell unit is weighted and voted on. The magnitude of the gradient is used as the projection, and the projection values ​​are successively superimposed onto the histograms represented by these 9 gradient directions.

[0136] Specifically, the overall features of the synthetic aperture radar (SAR) remote sensing image are extracted. Based on the overall features of the SAR image, the horizontal and vertical gradients of pixels, and the gradient direction and magnitude of pixels, the initial feature values ​​of the SAR image are obtained. This includes: extracting the features of each cell unit of the ship image, and concatenating the features of each cell unit to obtain the overall features of the image. First, n×n cell units need to be combined into large, spatially connected blocks. The gradients of the nine histogram channels in each cell unit are concatenated to obtain a multidimensional vector of n×n×9, which serves as the local feature of that block. For example, if the size of the block is set to 2×2 cell units, the feature vector of that block will have a dimension of 2×2×9. Second, local features are collected from all overlapping blocks in the detection window (i.e., the image with a size normalized to 512*512), and the collected local features are combined to obtain the initial feature values ​​of the SAR image.

[0137] Furthermore, the steps for normalizing the initial feature values ​​of the synthetic aperture radar remote sensing image are the same as those for normalizing the initial feature values ​​of the synthetic aperture radar remote sensing image, and will not be repeated here.

[0138] In one embodiment, the target feature dataset is input into a classification model for ship classification and recognition training to obtain a ship classification and recognition model, including:

[0139] The target feature dataset is divided into a target training set and a test dataset according to a preset ratio;

[0140] The classification model is trained based on the target training set to obtain the ship classification and recognition model.

[0141] Feature layer fusion is an intermediate level of fusion, which involves extracting features from raw information from sensors (features can be the edges, orientation, velocity, etc. of a target), and then comprehensively analyzing and processing these features. Feature layer fusion can achieve considerable information compression, which is beneficial for real-time processing. Furthermore, since the extracted features are directly related to decision analysis, the fusion result can provide the feature information needed for decision analysis to the greatest extent possible, thereby improving the efficiency of ship identification.

[0142] In one embodiment, the preset classification model employs a random forest multi-classification algorithm.

[0143] It should be noted that the preset classification algorithm can also adopt other classification methods such as linear discriminant analysis, distance discriminant analysis, Bayesian classification, decision tree, neural network, support vector machine, etc.

[0144] For example, the ship identification results based on different methods are shown in Tables 1 to 4 below:

[0145] Table 1: Statistics of target detection results based on a single SAR image:

[0146]

[0147] Table 2: Ships that failed to be identified in SAR images – Statistical analysis of target detection results based on single AIS data

[0148]

[0149] Table 3: Ships that failed to be identified in SAR images – Statistical analysis of target detection results based on single HRRP data.

[0150]

[0151] Table 4: Statistical Analysis of Target Detection Results Based on SAR-AIS-HRRP Feature Fusion for Ships Failing to be Identified in SAR Images

[0152]

[0153] Therefore, the present invention provides a ship identification method based on multi-source data fusion that can effectively improve the ship identification efficiency.

[0154] Figure 5 This is a schematic diagram of an embodiment of a ship identification method based on multi-source data fusion provided by the present invention.

[0155] Reference Figure 5 The present invention also provides a ship identification system based on multi-source data fusion, comprising:

[0156] The driving data acquisition module 501 acquires the driving data of the ship and calculates the ship's navigation characteristic data based on the driving data;

[0157] The high-resolution range image module 502 is used to acquire high-resolution range image data of ships and process the high-resolution range image data of ships based on a preset model to obtain a high-resolution range image feature dataset.

[0158] The feature dataset extraction module 503 is used to extract the feature dataset of synthetic aperture radar remote sensing images;

[0159] The fusion processing module 504 is used to fuse the navigation feature data of the ship, the high-resolution range image feature dataset, and the feature dataset of the synthetic aperture radar remote sensing image to obtain the target feature dataset.

[0160] Training module 505 is used to input the target feature dataset into a preset classification model for ship classification and recognition training, and to obtain a ship classification and recognition model;

[0161] Processing module 506 is used to identify the target feature dataset based on the ship classification and recognition model and obtain the ship recognition results.

[0162] The beneficial effects of the above embodiments are as follows: The present invention uses a driving data acquisition module 501 to acquire ship driving data; a ship high-resolution range image module 502 to process the ship high-resolution range image data based on a preset model to obtain a high-resolution range image feature dataset; a synthetic aperture radar remote sensing image extraction module 503 to extract the feature dataset of the synthetic aperture radar remote sensing image; a fusion processing module 504 to fuse the above data to obtain a target feature dataset; and a training module 505 and a processing module 506 to identify the target feature dataset based on a ship classification and recognition model to obtain ship identification results. The present invention performs data association and fusion of multi-source ship data, enabling high-precision ship identification.

[0163] The above provides a detailed description of the ship identification method and system based on multi-source data fusion provided by the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A ship identification method based on multi-source data fusion, characterized in that, include: Acquire the ship's navigation data, and calculate the ship's navigation characteristic data based on the navigation data; Acquiring high-resolution range image data of ships and processing the data based on a preset model to obtain a high-resolution range image feature dataset includes: acquiring high-resolution range image data of ships; extracting initial data from the high-resolution range image data of ships, the initial data of which includes line-of-sight distance, amplitude value, and phase value at different azimuth angles; dividing the high-resolution range image data of ships into several regions based on the azimuth angles of the initial data of the high-resolution range image data of ships; matching the same range gate to the high-resolution range image data of ships within each azimuth angle, and performing equidistant interpolation on the realization distance unit of each high-resolution range image data of ships; acquiring echo data features of the high-resolution range image data of ships at different realization distances, and normalizing the echo data features to obtain a high-resolution range image feature dataset; Extracting feature datasets from synthetic aperture radar remote sensing images; The navigation feature data of the ship, the high-resolution range image feature dataset, and the feature dataset of the synthetic aperture radar remote sensing image are fused to obtain the target feature dataset; The target feature dataset is input into a preset classification model for ship classification and recognition training to obtain the ship classification and recognition model; The target feature dataset is identified based on the ship classification and recognition model to obtain ship recognition results.

2. The ship identification method based on multi-source data fusion according to claim 1, characterized in that, Before obtaining the ship's navigation data, the following steps are also included: Acquire synthetic aperture radar (SAR) remote sensing images, and identify ship types based on the YOLOv5 algorithm to obtain ship identification results.

3. The ship identification method based on multi-source data fusion according to claim 2, characterized in that, The process of acquiring synthetic aperture radar (SAR) remote sensing images and identifying ship types based on the YoloV5 algorithm to obtain ship identification results includes: Acquire synthetic aperture radar remote sensing images; The synthetic aperture radar remote sensing image is cropped using the sliding window cropping method, and the cropped synthetic aperture image is then deleted, deduplicated, and classified. The synthetic aperture radar remote sensing images that have undergone deletion, deduplication, and classification are labeled to obtain a first target dataset, which includes a training dataset and a test dataset. The training dataset is imported into the YOLOV5 algorithm for training to obtain a ship classification and detection model. The test dataset is identified based on the ship classification and detection model to obtain ship identification results.

4. The ship identification method based on multi-source data fusion according to claim 1, characterized in that, The acquisition of ship navigation data and the calculation of ship navigation characteristic data based on the navigation data include: Initial data about a ship is acquired based on its Automatic Identification Unit (AIU). The initial data of the vessel is preprocessed to obtain the vessel's trajectory data; The trajectory data of the vessel is normalized to extract the vessel's navigation characteristic data.

5. The ship identification method based on multi-source data fusion according to claim 4, characterized in that, The navigation characteristic data includes at least one of the following: ship speed, ship position, ship course, ship acceleration, and rate of change of ship course.

6. The ship identification method based on multi-source data fusion according to claim 1, characterized in that, The features of the echo data of the high-resolution range image data of the ship at different realization distances include: Extract high-resolution range image data of the ship within a preset azimuth angle; The average value of the echo data for each azimuth angle within the preset azimuth angle sampling interval is calculated to obtain the echo data characteristics of the ship's high-resolution range image data at different realization distances.

7. The ship identification method based on multi-source data fusion according to claim 1, characterized in that, The feature dataset extracted from the synthetic aperture radar remote sensing image includes: Extract gradient orientation histograms of local regions based on the synthetic aperture radar remote sensing images; The synthetic aperture radar remote sensing image is converted into an image matrix; Gamma processing is performed on the pixel values ​​in the image matrix; Calculate the horizontal and vertical gradients of pixels in the image matrix after gamma processing to obtain the gradient direction and gradient magnitude of the pixel. Extract the overall features of the synthetic aperture radar remote sensing image, and obtain the initial feature values ​​of the synthetic aperture radar remote sensing image based on the overall features of the synthetic aperture radar remote sensing image, the horizontal and vertical gradients of the pixels, and the gradient direction and gradient magnitude of the pixels. The initial feature values ​​of the synthetic aperture radar remote sensing image are normalized to obtain the feature dataset of the synthetic aperture radar remote sensing image.

8. The ship identification method based on multi-source data fusion according to claim 1, characterized in that, The step of inputting the target feature dataset into a classification model for ship classification and recognition training to obtain a ship classification and recognition model includes: The target feature dataset is divided into a target training set and a test dataset according to a preset ratio; The classification model is trained based on the target training set to obtain the ship classification and recognition model.

9. A ship identification system based on multi-source data fusion, characterized in that, include: The navigation data acquisition module is used to acquire the navigation data of the vessel and calculate the navigation characteristic data of the vessel based on the navigation data; A high-resolution range image module for ships is used to acquire high-resolution range image data of ships and process the high-resolution range image data of ships based on a preset model to obtain a high-resolution range image feature dataset. The module includes: acquiring high-resolution range image data of ships; extracting initial data of the high-resolution range image data of ships, the initial data of which includes line-of-sight distance, amplitude value, and phase value at different azimuth angles; dividing the high-resolution range image data of ships into several regions based on the azimuth angles of the initial data of the high-resolution range image data of ships; matching the same range gate to the high-resolution range image data of ships within each azimuth angle, and performing equidistant interpolation on the realization distance unit of each high-resolution range image data of ships; acquiring echo data features of the high-resolution range image data of ships at different realization distances, and normalizing the echo data features to obtain a high-resolution range image feature dataset. A synthetic aperture radar remote sensing image extraction module is used to extract the feature dataset of the synthetic aperture radar remote sensing image; The fusion processing module is used to fuse the navigation feature data of the ship, the high-resolution range image feature dataset, and the feature dataset of the synthetic aperture radar remote sensing image to obtain the target feature dataset. The training module is used to input the target feature dataset into a preset classification model for ship classification and recognition training, thereby obtaining a ship classification and recognition model. The processing module is used to identify the target feature dataset based on the ship classification and identification model and obtain the ship identification result.

Images