Recognizing Fishing Activities via VMS Trace
Analysis Based on Mathematical Morphology
Yuan Zong
1
, Haiguang Huang
12
*, Feng Hong
1
, Yong Zhen
1
and Zhongwen Guo
1
1
Ocean University of China
2
Wenzhou Ocean and Fishery Vessel Safety Rescue Information Center
Qingdao, China
*Corresponding Author E-mail: haiguang2000@gmail.com
Abstract—Recently, the satellite-based Vessel Monitoring
Systems (VMS) have been widely deployed on fishing vessels.
Recognition of fishing activity is the key task for various
applications. Previous approaches are basically according to
change of vessel’s speed; or rely on the validated data from
logbooks or documented observations. In this paper, with a rated
60-second temporal resolution VMS data for two years on 34
vessels (typed as otter trawl) in the East China Sea, we propose a
Fishing Activity Recognition system (FAR). It exploits
Mathematical Morphology for analyzing the VMS trace data to
recognize fishing activities and obtain fishing related metrics.
Different from previous approaches, FAR carries out on VMS
traces only, requiring no other reference like logbooks or any
documented observations.
Keywords—VMS; Mathematical Morphology; fishing activity;
fishery yield; trajectory
I.
I
NTRODUCTION
In the last decade of years, as the Vessel Monitoring
Systems (VMS) have been widely equipped on fishing vessels,
vast VMS trajectory data has been generated. It contains
information of vessel’s ID number, time and location of that
moment, instant speed and heading [1]. With this data,
researches have been done on estimating fishing effort [2,3],
exploring vessels’ behavior and interaction [4,5], assessing
changes of environment pressure, management, routine areas
and fuel costs following with the alteration of fishing activity
[6], studying the fishing waters history evolution [7] and even
providing proof for fishing industry when maritime rights
conflicts [8,9]. Although purpose varies, recognition of vessel
activities, especially fishing activities, is the constant key issue.
However, presented as a series of geographic points
sequentially, VMS data does not indicate vessel behavior
directly. Analysis is necessary in order to distinguish different
vessel behaviors and recognize fishing activity during each
individual trip. To recognize fishing activity, previous
researches have proposed various methods that can be mainly
classified into two categories: experience-based method and
statistic-based method.
The experience-based methods, referred as classic methods
as well [10], are basically based on the fishing production
experience that vessel speed is quite different during different
activities. Usually, vessels slow down during fishing activity
and, in contrast, accelerate during steaming activity as heading
for fishing grounds or ports [11]. Even though simple and
direct, practically, this kind of method easily misjudges speed
reduction (e.g. a turning) as fishing behavior. Moreover,
thresholds vary with the type of vessels [12].
Statistical methods, such as the widely used Hidden
Markov Model (HMM), are based on probabilistic
interpretation on VMS samples to model the vessel activity.
HMM-based methods are generally robust, as they rely on
learning procedures on a large data set, creating a model
accommodating variations within an activity class. However,
these methods require enough amount of labeled data to train
the model, which should come from the other sources, like
from logbooks [13] and documented at-sea-observers [14].
Then the corresponding VMS data can be characterized,
divided into segments according to documented behaviors and
used to train statistical models. Afterwards the test VMS data
can be classified via the train model.
Therefore, to recognize fishing activity by using VMS data
only, accurately and efficiently, with no other references,
remains challenging. One problem is that, the individual trip
needs to segment from the whole trace set for each vessel. This
is challenging for no reference map to label the ports and
sometimes the small ports are not even labeled on maps. The
other challenging problem is how to classify the different
activities from one fishing trip.
In this paper, with a rated 60-second temporal resolution
(instead of a general 2-hour one) VMS data for two years on 34
vessels (typed as otter trawl) in the East China Sea, we propose
a Fishing Activity Recognition system (FAR). FAR only digs
into the collection of VMS data. Without map information, we
locate ports of each vessel where it anchors through experience
values on speed and direction information. Then we separate
individual trips of all the vessels based on their own port
locations. With no priori validated information to get data
labeled, which means no machine learning algorithms are
capable of being employed, we recognize fishing activity by
applying Mathematical Morphology (MM) on each individual
trip per vessel. So FAR overcomes the challenges above and
realizes fishing activity recognition.
As a step forward, we explore the correlation between the
changes of fishing activity and the relative yield through
analyzing whole-trip and fishing duration and distance, which
is 89.80% and 82.11% respectively.
This research is partially supported by the National Science Foundation
of China (NSFC) under Grant Number 61379128 and 61379127.
978-1-5090-2445-2/16/$31.00 ©2016 IEEE
Techno-Ocean 2016
465
3E.2