Back to Search
Start Over
An AIS-based deep learning model for multi-task in the marine industry.
- Source :
-
Ocean Engineering . Feb2024, Vol. 293, pN.PAG-N.PAG. 1p. - Publication Year :
- 2024
-
Abstract
- Automatic Identification System (AIS) plays an essential role in maritime surveillance and the global shipping industry. Thanks to Artificial Intelligence (AI) advances, almost all marine-related application needs can be realized by mining AIS Big Data. Most of the research, however, is still developing an AIS-based AI model for a specific task. When facing complex maritime applications, a large number of AI models are required. How to deploy these single-function AI models has become extremely challenging. Considering the distinctive characteristics of AIS data, this paper develops an AIS-based deep learning model for forecasting, classification, anomaly detection, and imputation in the maritime industry. This paper uses hierarchical blocks to ensure the expressive power and generalization ability, employs signal decomposition methods and processes the decomposed data like image processing. Finally, experiments have been conducted in both publicly available datasets and the dataset established in this paper. The performance of state-of-the-art models has been compared with the proposed model and the results shows that our approach can yield optimal results, including the optimal mean squared error in the forecasting task and imputation task, the accuracy in the anomaly detection task, and the category accuracy in the classification task, on our curated dataset. • We created a labeled dataset on vessel traffic in the upper reaches of the Yangtze River by comparing data through standard screening, trajectory tracing, and manual verification. • To minimize redundancy in AIS sequence information, we employ a whitening extraction method to capture vessel behavior modes. • This paper propose a "Meta" model to cater to different task requirements concerning AIS data. [ABSTRACT FROM AUTHOR]
Details
- Language :
- English
- ISSN :
- 00298018
- Volume :
- 293
- Database :
- Academic Search Index
- Journal :
- Ocean Engineering
- Publication Type :
- Academic Journal
- Accession number :
- 175032069
- Full Text :
- https://doi.org/10.1016/j.oceaneng.2024.116694