Vessel Identification via Trajectory Spectrogram Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing Automatic Identification System (AIS) data lacks reliability for vessel identification due to issues such as incorrect or spoofed Maritime Mobile Service Identity (MMSI) values, unintentional misconfiguration, and environmental disturbances causing vessel trajectory deviations, making it difficult to accurately track and identify marine vessels.
Innovation Solution
A machine learning model using signal analysis and deep learning techniques, specifically floating-mean Lomb-Scargle periodograms and Residual Networks (ResNet), to analyze vessel trajectory oscillations and generate spectrograms for vessel classification, independent of user-specified data, identifying vessels based on their behavior.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If AIS transceivers are used for vessel identification, then vessel tracking capability is provided, but reliability of vessel identification deteriorates due to spoofing and incorrect MMSI values
Solution Approach 1:
The patent introduces an intermediary identification system that does not rely on the vulnerable MMSI field. Instead, it uses environmental sensors (cameras, LIDAR, acoustic sensors) to capture physical characteristics of the vessel and its surroundings, creating an independent verification layer that mediates between the unreliable AIS data and the true vessel identity.
Solution Approach 2:
The patent replaces the mechanical/electronic identification system (AIS transceivers transmitting MMSI) with a sensor-based observation system. Rather than trusting transmitted electronic identifiers that can be spoofed, the system uses physical sensors to detect and record actual vessel characteristics, substituting one identification mechanism with a more trustworthy physical measurement approach.
2Measurement precision
If environmental sensors are used to capture vessel characteristics, then identification accuracy is improved, but device complexity increases
Solution Approach 1:
The patent designs the sensor system to serve multiple functions simultaneously: environmental cameras capture both vessel visual characteristics and surrounding scenery, LIDAR measures both distance and surface properties, and acoustic sensors detect both engine sounds and ambient noise. This multi-functionality reduces the need for separate specialized sensors, thereby limiting the increase in device complexity while maintaining high identification accuracy.
Solution Approach 2:
The system uses the vessel's own operational environment as part of the identification process. The environmental sensors capture characteristics of the surrounding area (landmarks, water conditions, atmospheric conditions) that are unique to each vessel's location and time, allowing the system to create distinctive identification signatures without requiring additional complex hardware.
3Measurement precision
If machine learning models analyze trajectory oscillations, then vessel classification accuracy is improved, but processing time increases
Solution Approach 1:
The patent performs preliminary processing of sensor data to extract relevant features before applying complex machine learning models. By pre-processing the environmental sensor data to identify key characteristics (such as extracting trajectory oscillation patterns, filtering environmental noise, and preparing feature vectors), the system reduces the computational burden on the classification models, thereby decreasing processing time while maintaining high accuracy.
Solution Approach 2:
The patent extracts only the most relevant features from the extensive sensor data for classification purposes. Rather than feeding all raw sensor data into the machine learning model, the system identifies and extracts specific discriminative features (such as oscillation frequency, amplitude patterns, and environmental signature characteristics), reducing processing complexity and time while preserving classification accuracy.
Data Source
AI summary
Computer implemented methods, systems, and computer-readable media for predicting a vessel identifier are provided. These include providing, at a memory, a vessel identification model, receiving, at a processor in communication with the memory, vessel data from one or more sources, determining, at the processor, a plurality of periodograms from the vessel data, the plurality of periodograms determined for a candidate vessel corresponding to candidate vessel data in the vessel data, determining, at the processor, a spectrogram based on the plurality of periodograms; and predicting, at the processor, a vessel identifier output of the candidate vessel based on the spectrogram and the vessel identification model.