Trans RINA, Vol 157, Part A3, Intl J Maritime Eng, Jul-Sep 2015
1.3 PREDICTIONS AND EARLY NOTIFICATION
One of the aims of the anomaly detection process is to detect a threatening situation as it develops. This helps the operator to prevent such a situation from occurring or, if that can’t be done, to prepare a response to it.
The prediction must be made as early as possible to give time for the operator to process the related information. For example, travelling at high speed in the wrong direction in a waterway may lead to a collision and the warning alert should be raised as soon as the situation is detected.
1.4 MAINTAINING A COMPLETE AND CONTINUOUS SURFACE PICTURE
The limitations of the human brain have been proven to have a significant impact
dynamic environments with large
for humans operating in quantities of
information available [10, 11]. It is difficult for a human to maintain and monitor the surveillance system at all times. This kind of task can be done better by machines. The reasoning of a computer is deterministic and it has an exact memory with a large capacity. Machines can process data to maintain a constant
and complete
monitoring picture and notify the operator when an anomaly is detected.
Here, we used the BNs approach to identify the anomaly behaviour by developing a model that represents the normal situation. In this case, the extent of the anomaly situation is defined by the degree to which a vessel deviates from the normal situation.
In this paper, we focus on the vessel anomaly behaviour based on speed data. With an appropriate speed limit, a vessel can take effective action to avoid collision. The speed limit will also reduce emissions by up to 70% [12].
This paper is organized as follows. Sec. II is a literature review on anomaly detection. Sec. III is an overview of our approach with the theory of BNs, while Sec. IV presents the BNs based approaches to anomaly detection. Sec. V describes the simulation process. Sec. VI describes data used for the simulation process. Sec. VII presents the experimental results with discussion, and Sec. VIII presents the conclusions.
2. LITERATURE REVIEW ON ANOMALY DETECTION
In computer science, anomaly detection has been an active research topic for a long time. The advancement of research in this area provides a variety of practical examples and studies
concerning classifications of
techniques and research challenges. Even if anomaly detection is an immature field of
research in other domains, we believe that this body of knowledge, at least at
its core, can be used and applied in many areas, including maritime surveillance.
Anomaly detection is extensively studied in areas such as network security, road surveillance, video surveillance, and military surveillance [13]. The non-consistent pattern is given various names such as anomaly, outliers, exceptions, etc. [2]. In the case of the maritime surveillance domain, the diverged patterns are referred to as anomaly.
According to Kazemi (2013), the anomaly detection techniques can be divided into two groups, namely the data-driven techniques and knowledge-driven techniques [2].
Data-driven techniques determine the normal
situation using machine learning or statistic algorithm to analyze the historical data [14]. Meanwhile, knowledge- driven techniques encode the expert knowledge into the system [15, 16].
2.1 DATA-DRIVEN TECHNIQUE
The data-driven technique is based on a classification that
learns a normal situation using unsupervised or
supervised learning. Data are considered as anomalous when they do not match the model. The data-driven technique is divided into two kinds of approach:
machine learning and statistical approaches. 2.1 (a) Machine Learning
The machine learning approach makes decisions based on the learning process from data. In the case of maritime surveillance, after the learning process, the machine learning model can be used to classify the normal and anomaly situation.
Some popular techniques belonging to this field include clustering,
neural Clustering networks (NNs), support
machines (SVMs), Bayesian Networks (BNs), etc.
A study done by Dahlbom and Niklasson (2007) [17] used the trajectory clustering technique for anomaly detection. In this case, the anomaly is determined by looking at the probability that the path that does not match the normal trajectory.
Neural Networks (NNs)
The NNs use a learning algorithm inspired by the structure and function of the neuron. They are often referred to as black boxes. Some previous studies have used anomaly detection using NNs in the maritime domain [18-22]. The NNs
predict the behaviour of various phenomena through the learning process.
According to Patcha and Park (2007) [22], the main advantage of NNs is the inclusion of tolerance when it
vector the
A-146
©2015: The Royal Institution of Naval Architects
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50 |
Page 51 |
Page 52 |
Page 53 |
Page 54 |
Page 55 |
Page 56 |
Page 57 |
Page 58 |
Page 59 |
Page 60 |
Page 61 |
Page 62 |
Page 63 |
Page 64 |
Page 65 |
Page 66 |
Page 67 |
Page 68 |
Page 69 |
Page 70 |
Page 71 |
Page 72 |
Page 73 |
Page 74 |
Page 75 |
Page 76