Please use this identifier to cite or link to this item: https://hdl.handle.net/10216/115333
Full metadata record
DC FieldValueLanguage
dc.creatorJosé Pedro Vieira Gomes
dc.date.accessioned2022-09-07T16:49:27Z-
dc.date.available2022-09-07T16:49:27Z-
dc.date.issued2018-07-10
dc.date.submitted2018-09-05
dc.identifier.othersigarra:282843
dc.identifier.urihttps://hdl.handle.net/10216/115333-
dc.descriptionThis project revolves around the detection of anomalous behaviors, more specifically on maritime vessels and using the Automatic Identification System (AIS) data. To accomplish the intended goal while having several restrictions like the large amounts of data which imply performance issues or the lack of labelled data and restrictions on creating one since anomalous behavior should be detected even without a priori characterization. To achieve the detection of anomalies Hierarchical Temporal Memory (HTM) based algorithms were used. HTM theory core is based on cortical research, the algorithms used on this project have as focus the ability to learn temporal sequences, since the behavior of maritime vessels is described by a data stream of AIS data which includes fields like the geolocation, speed over ground (SOG), course over ground (COG), vessel type or identification. This project starts with the expectation that an algorithm that can use the sequences of behavior instead of the behavior at any point in time will present better results. The first step of the project was to understand how to use HTM algorithms to detect anomalous behavior. The basic model used to process the data is composed of several components. The first is a Geospatial Encoder which transforms coordinates data into Sparse Distributed Representations (SDRs) used on all other steps of the process. SDRs are basically a list of bits, each being a 0 or 1 depending on the input data but with semantic meaning which produces differences from simple binary data. The second being a Spatial Pooler algorithm able to make use of the SDRs semantics and able to normalize the data while maintaining encoded information and presenting the first step on the general learning capabilities, being able to learn and predict the next step based only on the current information. The last component is the Temporal Memory, it extends the capability of the learning algorithm, it enables the prediction of the future data based on the sequence of data previously learned. All these components allow the ability to learn complex sequences of data, in this project the ability to learn sequences was first applied to learn sequences of positions. The ability to learn sequences and later perform predictions allows a simple way of detecting anomalous behavior. After the creation of a model capable of having as input the current position and making predictions, an anomalous behavior should be detected when a prediction is very different from the actual next value. In this case if there's no close prediction to a position after the sequence of positions being previously assessed then an abnormal behavior is detected. The next phase of the project was about improving the characterization of the sequences, the vessel trajectories are not only characterized by positions but also by other information like speed or the timestamp at each new data point. This data could be used directly or by derivation of new information, e.g. timestamp can be converted into time from sequence start. To use these data as input was also important to choose the right encoder and pertinent parameters. To accomplish these a better understanding of the vessels' trajectories is fundamental to which an analysis of the data was conducted. Still, the main concern on this stage was to try different data with different HTM algorithm parameters to understand the impact of different information on the ability to detect anomalies while trying to improve the performance both in terms of learning sequences, reducing false anomalies, and improving the amount of information describing a sequence, boosting the ability to find true anomalies. This is the main phase and involves a multitude of experiments and the definition of some metrics and data samples to allow the quantification of the results obtained in order to compare different models.
dc.description.abstractWith an increasingly number of ships equipped with an Automatic Identification System more and more data is being generated and creating an opportunity for new studies of the maritime vessel behaviors. The problem of securing and protecting vast expanses of the maritime zone could use the help of an automatic vessel anomalous behavior system. With that objective in mind the development of this project took the capabilities of Hierarchical Temporal Memory theory and respective algorithms to help identify anomalous behaviors on vessel trajectories improving sea monitoring capabilities and the possibility of more opportune warnings and subsequent action plans from simple contact or vessel identification to arrest or rescue missions.
dc.language.isoeng
dc.rightsopenAccess
dc.rights.urihttps://creativecommons.org/licenses/by-nc/4.0/
dc.subjectEngenharia electrotécnica, electrónica e informática
dc.subjectElectrical engineering, Electronic engineering, Information engineering
dc.titleOVERSEE: Identification Of Anomalous Vessel Behaviour
dc.typeDissertação
dc.contributor.uportoFaculdade de Engenharia
dc.identifier.tid202116514
dc.subject.fosCiências da engenharia e tecnologias::Engenharia electrotécnica, electrónica e informática
dc.subject.fosEngineering and technology::Electrical engineering, Electronic engineering, Information engineering
thesis.degree.disciplineMestrado Integrado em Engenharia Electrotécnica e de Computadores
thesis.degree.grantorFaculdade de Engenharia
thesis.degree.grantorUniversidade do Porto
thesis.degree.level1
Appears in Collections:FEUP - Dissertação

Files in This Item:
File Description SizeFormat 
282843.pdfOVERSEE: Identification Of Anomalous Vessel Behaviour9.74 MBAdobe PDFThumbnail
View/Open


This item is licensed under a Creative Commons License Creative Commons