Ship motion parameter estimation method based on wake matching

Abstract

The invention relates to the technical field of remote sensing image ship motion, and discloses a wake matching-based ship motion parameter estimation method, which comprises the following steps of: constructing a ship wake model comprising an infrared sea surface model, a ship wake height field and a ship wake temperature field; setting multiple navigational speeds, multiple navigations and multiple sea condition parameters by using the model to generate a ship wake infrared image, and constructing a ship wake database; a deep learning convolutional neural network and a multi-task loss function are adopted to extract a wake to be estimated; and constructing a multi-stage image matching model based on the feature point extraction module, the feature vector description module and the feature point matching module, and matching the database and the batch to-be-estimated trails by using the multi-stage image matching model. According to the method, the infrared sea surface model mixedly calculates two sea surfaces, so that the repeatability of the sea surfaces is reduced while the authenticity is ensured; by simulating the wake temperature field and the wake height field, the authenticity of the wake is improved, the constructed wake data set can be reused, and efficient estimation of ship motion parameters is realized.

Description

Technical Field

[0001] The invention relates to the technical field of remote sensing image ship motion, and in particular to a ship motion parameter estimation method based on wake matching. Background Art

[0002] When a ship sails on the sea, it will inevitably produce a wake, which is one of the important characteristics of a moving target.

[0003] At present, there are many difficulties in the field of ship wake motion parameter estimation at home and abroad. The traditional method of extracting wakes from infrared images and estimating speed and heading by combining empirical formulas is highly dependent on the accuracy of wake extraction. In high sea conditions, wakes are easily disturbed by waves, and when wakes are blocked or absorbed by sea objects or land, fine structure wakes are difficult to obtain, resulting in significant errors in the method of using ship wakes extracted from infrared images and empirical formulas to estimate speed and heading. Summary of the invention

[0004] The purpose of the present invention is to provide a ship motion parameter estimation method based on wake matching, so as to solve the problem in the existing ship wake motion parameter estimation technology that wake extraction is difficult due to factors such as sea conditions, thereby causing serious deviations in estimating speed and heading using infrared images and empirical formulas.

[0005] The purpose of the present invention can be achieved by the following technical solutions: A method for estimating ship motion parameters based on wake matching comprises the following steps: Construct a sea surface ship wake model including an infrared sea surface model, a ship wake height field and a ship wake temperature field, wherein the infrared sea surface model is calculated and rendered in combination with a sine wave and a sea spectrum sea surface, the ship wake height field is constructed based on a slender ship model and an integral domain transformation and attenuation model, and the ship wake temperature field is constructed according to the sea surface temperature characteristics and the ship thermal wake mechanism, and is adjusted according to motion parameters; Based on the constructed sea surface ship wake model, multiple speed, multiple heading, and multiple sea condition parameters are set to generate ship wake infrared images and build a ship wake database; A deep learning convolutional neural network and a multi-task loss function are used to extract the tail to be estimated; A multi-stage image matching model is constructed based on the feature point extraction module, feature vector description module and feature point matching module, and is used to match the database and batches of tails to be estimated to obtain and select the optimal motion parameter estimation result; The steps of constructing the multi-stage image matching model are as follows: Extracting feature points of the image from the template image and the target image based on the feature point extraction module, cropping image blocks around the feature points according to certain specifications with the feature points as the center, and inputting the blocks into the feature vector description module to obtain the feature point description vector; The feature point matching module searches for the template image description vector with the smallest Euclidean distance to the feature point description vector, and calculates whether the Euclidean distance between the two is less than a threshold value to obtain a feature point matching result.

[0006] Preferably, the infrared sea surface model calculates the sea surface of the sine wave and the sea surface of the sea spectrum by sea surface simulation, and the sea surface simulation uses the sea spectrum model and the sine wave with superimposed noise at the near and far sea surface grid points respectively to show details and reduce repetition.

[0007] Preferably, the ship wake database includes simulation data of multiple speeds, multiple headings and multiple sea conditions.

[0008] Preferably, the deep learning convolutional neural network is Faster RCNN, which can reduce computational redundancy and significantly improve speed compared to the traditional RCNN network.

[0009] Preferably, the feature points of the wakes of ships in different motion states are extracted by a feature extraction module, specifically, Gaussian blur is applied to the images using Gaussian kernels with different standard deviations, the difference between the two Gaussian blurred images is calculated, and the local extreme points of the images are screened as feature points.

[0010] Preferably, the feature vector description module adopts a convolutional neural network with non-shared parameters.

[0011] Preferably, the feature point matching module adopts a kd tree structure.

[0012] Beneficial effects of the present invention: The present invention calculates two types of sea surfaces, namely, sine waves and sea spectra, through sea surface simulation, thereby ensuring authenticity while reducing the repeatability of the sea surface, which is more in line with the actual situation. The image matching model is constructed by using the image feature matching method, and image features with characterization capabilities are extracted for matching, which can reduce redundant steps during matching and reduce the impact of noise on the model. The present invention performs wake estimation through the wake template matching method, which breaks through the traditional inversion estimation method. After the database is completed, the ship motion parameters can be estimated in batches, which significantly improves the accuracy and efficiency of ship motion parameter estimation, especially the application effect under complex sea conditions.

[0013] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for describing the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0015] Figure 1 This is a flow chart of a method for estimating ship motion parameters based on wake matching according to the present invention; Figure 2 A flow chart for constructing a ship wake database of the present invention; Figure 3 This is a flow chart of the matching method proposed by the present invention. DETAILED DESCRIPTION

[0016] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0017] A method for estimating ship motion parameters based on wake matching specifically comprises the following steps: Step 1: Construct a ship wake model: like Figure 1 and Figure 2 As shown, the present invention constructs a sea surface ship wake model, which consists of a sea surface model and a ship wake model, wherein the sea surface model simulation hybridly calculates the sea surface based on a sine wave and the sea surface based on a sea spectrum, and the ship wake model simulation calculates the wake height field and the wake temperature field based on the geometric model and temperature model of the wake respectively, and converts the scene radiation calculation result into an image texture for rendering to obtain an infrared sea surface ship wake model; The sea surface can be regarded as a superposition of a series of waves with different amplitudes, different wavelengths, and different phases. The sea spectrum model describes the distribution of wave energy in the frequency domain and the direction domain through mathematical functions. The sea surface model simulation of the present invention uses the sea spectrum model to generate the sea surface height field results at the close sea surface grid points to show the details of the dynamic changes on the sea surface. The long-distance grid points use sine waves with superimposed noise to reduce the repetition of the overall sea surface.

[0018] The ship wake height field is based on the slender ship model. By introducing the integral domain transformation, the ship turns during navigation. The attenuated wake height field is calculated using the empirical model of transverse wave attenuation of the Kelvin wake, and an improved height field model of the ship Kelvin wake is established. The temperature change of the ship wake is because the power system of the ship stirs the low-temperature seawater in the deep layer to the sea surface, forming a temperature difference with the sea surface, which is displayed as a dark band on the infrared detector. The present invention is based on the vertical distribution characteristics of the sea surface temperature, combined with the generation mechanism and characteristics of the ship thermal wake, and calculates the width of the ship thermal wake and the temperature distribution formula, thereby constructing a temperature field model of the ship wake, and by adjusting the motion parameters, the ship wake temperature field under different motion parameters is established; The textures that need to be pre-calculated in the infrared simulation of sea surface wakes include: sky radiation texture, atmospheric radiation texture, and solar radiation texture. The scene radiation calculation results are converted into image textures using MODTRAN software, and the above sea surface and wake are rendered to obtain an infrared sea surface ship wake model, and a sea surface ship wake model that can simulate multiple speeds, multiple headings, and multiple sea conditions is constructed.

[0019] Step 2: Build a database, including a ship wake template database and a ship wake real-time database Based on the ship wake model, a ship wake template database was established. Different motion parameters were set to generate infrared images of ship wakes and construct a ship wake database.

[0020] Construct a ship wake template database: In this embodiment, 31 different speeds with wakes ranging from 0m / s to 30m / s, 10 different headings with turning angles ranging from 0° to 45°, and 7 different sea conditions ranging from calm sea surface to wave height of 6 meters were selected. With the help of Unity3D rendering engine, 2170 1200×1200 images were simulated as template images for subsequent matching estimation.

[0021] Constructing a real-time database of ship wakes: Data was collected through drones, and a total of about 20 minutes of video records were obtained. The video was deframed with the help of the processing tool ffmpeg, and a total of 11,495 real-shot image data were obtained as the target image for subsequent matching estimation.

[0022] Step 3: Extract the tail to be estimated: The deep learning convolutional neural network Faster RCNN is used as the ship wake extraction algorithm, and a multi-task loss function is adopted to achieve fast and accurate extraction of ship wakes.

[0023] In this embodiment, the ship training data comes from the public training data set SeaShips Dataset, which is specially designed for ship detection and contains 6 types of ships and about 9,000 annotated images. In the experiment, the number of training iterations is set to 8x104 and 4x104 respectively, and the batch size is set to 400. After the training is completed, the wake extraction is performed on the wake real-shot database collected by the present invention, and the area with the wake in the image is detected to facilitate subsequent matching.

[0024] Step 4: Construct feature point extraction module: The present invention designs a feature point extraction module, which is constructed by the DOG operator and is used to extract the feature points of the template image and the target image. The module first uses Gaussian kernels with different standard deviations to apply Gaussian blur to the image, then subtracts the two Gaussian blurred images to obtain a DOG response image, and detects extreme points in the DOG response image as image feature points. The module has scale invariance and strong anti-interference ability, thereby realizing accurate positioning of feature points.

[0025] Step 5: Construct feature vector description module: The present invention improves the L2-Net network model and proposes the EL2-Net (Effective L2-Net) model. Several key improvements have been made to the ship wake image: a dual-branch non-shared parameter structure is used to independently extract image pair features to enhance adaptability to differences in viewing angle, illumination, etc.; a fully convolutional network is used to retain spatial information and extract common features to improve feature repeatability; instance normalization is introduced to replace batch normalization to accelerate convergence while maintaining independence between feature instances; the Leaky ReLU activation function is used to alleviate the gradient vanishing problem and enhance nonlinear fitting capabilities. These improvements significantly improve the robustness of the feature description of ship wake images, making its key point matching accuracy and stability in complex scenes better than the original model.

[0026] The feature point vector description module cuts a 32x32 image block around the feature point extracted in step 4 as the center, inputs the EL2-Net feature point vector extraction model proposed by the present invention, extracts descriptors for the key point positions in the image, and obtains a 1x128-dimensional feature vector description. In view of the significant nonlinear differences between different images, the present invention uses a convolutional neural network with unshared parameters to extract feature point vector descriptions for each image.

[0027] Step 6: Construct a feature point matching module. The specific steps are as follows: The feature point matching module uses the BBF algorithm to find the nearest neighbor descriptor and its Euclidean distance, extracts nodes from the queue, and calculates the Euclidean distance to the target point. If the distance is greater than the current closest distance, the point is ignored and the next node is calculated. If the distance is less than the current closest distance, the nearest distance is updated and the nearby child nodes of the node are added to the queue until all nodes are traversed to obtain the node closest to the target point. It also determines whether the Euclidean distance between the node and the target is less than the threshold. If it is greater than the threshold, the match fails. If it is less than the threshold, the feature point matching result is obtained.

[0028] Step 7: Estimation of ship motion parameters. The specific steps are as follows: like Figure 3 As shown in the figure, the ship wake template database and the real-shot ship wake data are input into the multi-stage remote sensing image matching model. After feature point extraction, feature point vector description and feature point matching, the wake template matching result is output. The motion parameters of the ship can be estimated through the best matching result.

[0029] In summary, the present invention provides a method for estimating ship motion parameters based on wake matching. The infrared sea surface model of the method hybridly calculates two types of sea surfaces, which ensures the authenticity while reducing the repeatability of the sea surface; the authenticity of the wake is improved by simulating the wake temperature field and the wake height field; the multi-stage remote sensing image matching algorithm of the present invention can extract the feature vector of the target wake and achieve accurate matching; the constructed wake data set can be reused to achieve efficient estimation of ship motion parameters.

[0030] The invention described above can also be extended to other similar ship motion parameter estimation tasks. It only needs to set appropriate parameters, establish a ship wake template database according to step 2, and match the target according to step 7 to estimate the ship motion parameters based on the template matching method.

[0031] The above contents are merely examples and explanations of the concept of the present invention. Those skilled in the art may make various modifications or additions to the specific embodiments described or replace them in a similar manner. As long as they do not deviate from the concept of the invention or exceed the scope defined by the claims, they shall all fall within the protection scope of the present invention.

Claims

1. A ship motion parameter estimation method based on wake matching, characterized in that: The following steps are involved: Construct a sea surface ship wake model including an infrared sea surface model, a ship wake height field and a ship wake temperature field, wherein the infrared sea surface model is calculated and rendered in combination with a sine wave and a sea spectrum sea surface, the ship wake height field is constructed based on a slender ship model and an integral domain transformation and attenuation model, and the ship wake temperature field is constructed according to the sea surface temperature characteristics and the ship thermal wake mechanism, and is adjusted according to motion parameters; Based on the constructed sea surface ship wake model, multiple speed, multiple heading, and multiple sea condition parameters are set to generate ship wake infrared images and build a ship wake database; A deep learning convolutional neural network and a multi-task loss function are used to extract the tail to be estimated; A multi-stage image matching model is constructed based on the feature point extraction module, feature vector description module and feature point matching module, and is used to match the database and batches of trails to be estimated to obtain and select the optimal motion parameter estimation result; The steps of constructing the multi-stage image matching model are as follows: Extracting feature points of the image from the template image and the target image based on the feature point extraction module, cropping image blocks around the feature points according to preset specifications with the feature points as the center, and inputting the image blocks into the feature vector description module to obtain the feature point description vector; The feature point matching module searches for the template image description vector with the smallest Euclidean distance to the feature point description vector, and calculates whether the Euclidean distance between the two is less than a threshold value to obtain a feature point matching result.

2. A method for estimating ship motion parameters based on wake matching according to claim 1, characterized in that: The infrared sea surface model uses a sea surface simulation to mix and calculate a sine wave sea surface and a sea spectrum sea surface. The sea surface simulation uses a sea spectrum model and a sine wave with superimposed noise at near and far sea surface grid points to display details and reduce repetition.

3. The method for estimating ship motion parameters based on wake matching according to claim 1, characterized in that: The ship wake database includes simulation data of multiple speeds, multiple headings and multiple sea conditions.

4. The method for estimating ship motion parameters based on wake matching according to claim 1, characterized in that: The deep learning convolutional neural network is Faster RCNN.

5. The method for estimating ship motion parameters based on wake matching according to claim 1, characterized in that: The feature extraction module is used to extract the feature points of ship wakes in different motion states. Specifically, Gaussian blur is applied to the image using Gaussian kernels with different standard deviations, and then the difference between the two Gaussian blurred images is calculated to screen the local extreme points of the image as feature points.

6. A method for estimating ship motion parameters based on wake matching according to claim 5, characterized in that: The feature vector description module adopts a convolutional neural network with unshared parameters.

7. The method for estimating ship motion parameters based on wake matching according to claim 5, characterized in that: The feature point matching module adopts a kd tree structure.

Images