CN-121811697-B - Ship traffic situation awareness method, device, equipment and storage medium

CN121811697BCN 121811697 BCN121811697 BCN 121811697BCN-121811697-B

Abstract

The application provides a ship traffic situation awareness method, device, equipment and storage medium, wherein the method comprises the steps of acquiring radar characteristics and AIS characteristics of a target ship and determining attention weight; according to the time change rate of the attention weight and the time sequence differential error of the radar feature and the AIS feature, constructing a time sequence differential consistency constraint, wherein the time sequence differential consistency constraint is used for forcing the time change rate to be inversely proportional to the time sequence differential error, generating a fusion feature according to the constructed time sequence differential consistency constraint and the attention weight, and generating the ship track prediction of the target ship according to the fusion feature. According to the application, by introducing a time sequence differential consistency constraint and delay perception attention module, the radar and AIS characteristics are subjected to accurate time alignment and anti-vibration fusion, so that stable fusion characteristics are generated, accurate ship track prediction is realized, and situation perception precision and navigation safety are effectively improved.

Inventors

SONG YINGLIN
LONG XIAOJING
YANG CHANGYI
GE BING
Sui jilin
YANG YUSEN
MENG ZHAOYI
ZHANG JINKAI
WANG XIAOYU
WU QINGWEI

Assignees

中交第一航务工程勘察设计院有限公司

Dates

Publication Date: 20260508
Application Date: 20260306

Claims (10)

1. The ship traffic situation awareness method is characterized by comprising the following steps of: acquiring radar characteristics and AIS characteristics of a target ship; determining attention weights according to probability distributions of the radar features and the AIS features; Constructing a time sequence differential consistency constraint according to the time change rate of the attention weight and the time sequence differential error of the radar characteristic and the AIS characteristic, wherein the time sequence differential consistency constraint is used for forcing the time change rate to be inversely proportional to the time sequence differential error; Generating fusion features according to the constructed time sequence differential consistency constraint and the attention weight; and generating a ship track prediction of the target ship according to the fusion characteristics.
2. The method of claim 1, wherein determining the attention weight from the probability distribution of the radar signature and the AIS signature comprises: splicing the radar feature and the AIS feature to form a spliced feature; inputting the spliced characteristic into a linear transformation matrix to generate a linear transformed result; A softmax normalization is applied to the linearly transformed results to determine an attention weight.
3. The method of claim 1, wherein constructing a time-series differential consistency constraint comprises: constructing a loss term which is the product of the norm of the attention weight time derivative and the square norm of the time sequence differential error; and constructing a time sequence differential consistency constraint according to the loss term.
4. The ship traffic situation awareness method according to claim 1, wherein the time sequence differential error comprises the following steps of: acquiring a feature average value of the radar feature and the AIS feature in a past time window; And calculating a difference according to the characteristic average value and the current value of the radar characteristic or the AIS characteristic so as to determine a time sequence differential error.
5. The method of claim 1, wherein constructing a time-series differential consistency constraint comprises: performing a fast fourier transform on the attention weighting sequence to obtain a frequency spectrum; performing a power-rejection operation on high-frequency components of the spectrum to generate a modified spectrum; Performing an inverse fast fourier transform on the modified spectrum to obtain a modified attention weight sequence; and constructing a time sequence differential consistency constraint according to the corrected attention weight sequence.
6. The method of claim 1, wherein generating fusion features from the constructed time-series differential consistency constraints and the attention weights comprises: splicing the radar feature and the AIS feature to form a spliced feature; the stitched features are weighted using the attention weights to generate fusion features.
7. The ship traffic situation awareness method according to claim 1, wherein the calculating step includes: setting the width of the integration window as a maximum level delay time constant of the sensor, wherein the time constant is determined based on the characteristics of the sensor; A first order low pass filtering is performed according to the integration window width to calculate a time differential error.
8. A ship traffic situation awareness apparatus, comprising: The acquisition module acquires radar characteristics and AIS characteristics of the target ship; the weight module is used for determining attention weight according to probability distribution of the radar features and the AIS features; a building module for building a time sequence differential consistency constraint according to the time change rate of the attention weight and the time sequence differential error of the radar feature and the AIS feature, wherein the time sequence differential consistency constraint is used for forcing the time change rate to be inversely proportional to the time sequence differential error; The fusion module generates fusion characteristics according to the constructed time sequence differential consistency constraint and the attention weight; and the generation module is used for generating the ship track prediction of the target ship according to the fusion characteristics.
9. An electronic device comprising a memory and a processor, the memory having stored therein a computer program, the processor implementing the method of any of claims 1-7 when the computer program is executed.
10. A computer readable storage medium, having stored thereon a computer program which, when executed in a computer, causes the computer to perform the method of any of claims 1-7.

Description

Ship traffic situation awareness method, device, equipment and storage medium Technical Field The application belongs to the field of ship traffic, and particularly relates to a ship traffic situation awareness method, device, equipment and storage medium. Background The existing ship traffic situation awareness method based on multi-source data fusion generally comprises the key steps of data acquisition, space-time alignment, feature fusion, situation awareness generation and the like. The system synchronously collects target sea area data from radar sensors, AIS receivers, hydrological sensors and other devices, processes the data to output a structured data set, processes multi-source data through a space-time alignment engine to complete space-time alignment and data association and output a space-time alignment associated data set, builds various feature vectors for the associated data set through a multi-mode fusion module, generates fusion feature vectors through an attention mechanism to obtain deep fusion features of each ship target, and finally a situation awareness engine predicts ship tracks, calculates nearest meeting distance (CPA) and nearest meeting Time (TCPA) based on the fusion features and generates collision risk early warning when meeting conditions. However, in practice, the prior art exposes a series of problems, especially in the multi-modal data fusion process. Due to the difference between the sampling frequencies of the radar and the AIS sensor, a millisecond-level time delay may be caused, so that the weights allocated to the characteristics of the same target in adjacent time steps fluctuate drastically. This time-sensitive amplification directly leads to oscillation phenomena during feature fusion, which are manifested as high-frequency jumps of the fused features over successive time steps. The method is characterized in that the method is obviously dithered in ship track prediction, prediction errors are increased, and false triggering is easily caused in the aspect of conflict early warning, so that unnecessary alarms are caused, and normal ship traffic management work is interfered. These problems are more hidden, are hardly visible when the data synchronization is good, and only appear when the asynchronization exceeds the system tolerance threshold. Moreover, the performance form of the sensor is similar to the noise of the sensor, so that the sensor is easy to misjudge, the problem is solved in a delayed manner, and the hidden danger of safe burying during ship navigation is provided. These problems severely restrict the performance improvement of the existing ship traffic situation awareness method, and cannot meet the increasing shipping safety requirements of high precision and high reliability. Disclosure of Invention The application aims to overcome the defects in the prior art and provide a ship traffic situation sensing method, a device, equipment and a storage medium. The application provides a ship traffic situation awareness method, which comprises the following steps: acquiring radar characteristics and AIS characteristics of a target ship; determining attention weights according to probability distributions of the radar features and the AIS features; Constructing a time sequence differential consistency constraint according to the time change rate of the attention weight and the time sequence differential error of the radar characteristic and the AIS characteristic, wherein the time sequence differential consistency constraint is used for forcing the time change rate to be inversely proportional to the time sequence differential error; Generating fusion features according to the constructed time sequence differential consistency constraint and the attention weight; and generating a ship track prediction of the target ship according to the fusion characteristics. Optionally, determining the attention weight according to the probability distribution of the radar feature and the AIS feature includes: splicing the radar feature and the AIS feature to form a spliced feature; inputting the spliced characteristic into a linear transformation matrix to generate a linear transformed result; A softmax normalization is applied to the linearly transformed results to determine an attention weight. Optionally, constructing the time-series differential consistency constraint includes: constructing a loss term which is the product of the norm of the attention weight time derivative and the square norm of the time sequence differential error; and constructing a time sequence differential consistency constraint according to the loss term. Optionally, the time sequence differential error comprises calculation through first-order low-pass filtering, and the specific steps comprise: acquiring a feature average value of the radar feature and the AIS feature in a past time window; And calculating a difference according to the characteristic average value and the current value of