Problem Statement
S No.10
Theme Name Blue Economy
Title of the PS To track, provide alert to ships before piracy attack &
continue to pinpoint track the attacking boat info to CG etc.
Current Situation Piracy, armed robbery & pilferage is a serious concern to
shipping worldwide including coastal waters in ASEAN region
& ports. Though there are statistics /report available on
public domain on area of attack with details, there is no
system which alerts the potential target such as to take early
avoiding action/precautions. Also many a times, the attacking
boat after attack escapes in the darkness.
Solution Needed Using Satellite imaging, AIS data etc a solution can be
provided using AI/ML to provide the much needed pre alert as
well continue track the offenders
Risk Analysis
Analysing riskof attack
associated with a particular
vessel and in specific
areas. Use historical data to
assess risk score and
probability of attack.
7.
High-risk zones
Map
● Worldmap which shows
the past attacks and the
high-risk zones
● Can help the ship’s captain
reroute to another path or
be more careful when
going through the danger
zone.
● Will be achieved by a
clustering algorithm.
● Uses piracy data from
1990-2020
Clustering
Danger zones
Prevention
Re-route
8.
When the
captain is
planningthe
routes before
starting to
sail
Look at the
map of
dangerous
zones
beforehand
Plan for
alternate
routes and
take all
precautions
9.
What affect pirates’
decisionto attack?
● Vessel type
● Vessel size
● Vessel flag
● Vessel geographic location
● Date-Time
10.
Hypothesis 1: Theprobability of a piracy attack varies with vessel type and size[1]
Statistics of past attacks 1990 - 2020
[1] J.M. Shane, S. Magnuson, Successful and unsuccessful pirate attacks worldwide: a
situational analysis, Justice Q. JQ 33 (4) (2016) 682–707.
11.
Hypothesis 2: Openregistry vessels have a higher probability of piracy attacks[2]
Statistics of past attacks 1990 - 2020
[2] G. Kiourktsoglou, A.D. Coutroubis, Is Somali piracy a random phenomenon? WMU
J. Maritime Affairs 11 (1) (2012) 51–70.
12.
Hypothesis 3: Vesselsnavigating off the east coast of Africa have a higher probability of
piracy attacks[3]
Statistics of past attacks 1990 - 2020
[3] S. Bateman, Maritime piracy in the Indo-Pacific region–ship vulnerability issues, Marit. Policy Manag. 37 (7) (2010) 737–751.
13.
Hypothesis 4: Attacksmainly occurred at night and early morning.
Statistics of past attacks 1990 - 2020
14.
Risk assessment
model
● Realdata has shown that there
are a lot of factors that affect
pirates’ decision to attack.
● Each vessel has a different
probability of being attacked.
● Objective: Assess the risk
score/probability of each vessel
when passing a particular area
● Dataset: The piracy and armed
robbery dataset (GISIS, IMO)
● Implementation: Logistic
Regression Logistic Regression
Risk Score
you have 15%
chance of being
attacked if
entering this area
15.
Assign a risk
scorebefore
starting to
sail so crew
can be aware
beforehand
Alert the ship
before it
enters a
high-risk area
Crew will stay
alert and
aware be on
the lookout
for pirates
16.
Attack Prediction
Predicting animpending
attack using a three-fold
system, that focuses on
long-range as well as
short-range, in all visibility
conditions. Ensures that
crew is alerted of any
suspicious activity.
17.
Detecting threats
using Radars
●Using Radar, nearby ships,
their location, velocity and
direction can be obtained.
● Detect if another ship
comes within a certain
radius of our ship
● If that ship is unverified on
the AIS, alert the crew of a
possible threat.
Fault tolerant
Low Latency Centralized
Existing marine Radar present in
the ship
Prerequisites:
18.
Detects any
boat inthe
radius
coming
towards our
ship, using
Radar
Cross-verify
the boat on
the AIS to see
if it’s a
verified ship,
to reduce
false positives
Alert the crew
of the
unidentified
boat, it’s
direction and
velocity
19.
Risk: 15% lon:127, 5
lat: -80
velocity: 50 km/h
50 km/h
45 km/h
22 km/h
49 km/h
70 km/h
50 km/h
There are three sus
ships approaching us in
20 mins (based on our
current direction and
velocity
50 km/h
This ship
approaching us, but
was identified by AIS
data
20.
Up to thispoint
you have 15%
chance of being
attacked if
entering this area
● The risk score based on:
○ Vessel type
○ Vessel size
○ Vessel flag
○ Vessel geographic location
○ Date - Time
○ Historical incidents
● An alert is triggered when:
○ There are unidentified ships approaching us
→ No connection between risk score
and an alert
21.
The crucial problem
youhave 15%
chance of being
attacked if
entering this area
● Risk score do not capture current
circumstances. No integration with
radar information.
● Risk model only depend on historical
data → Maybe it will be biased by
time.
● For example, the situation, in which
there are 3 suspect ships
approaching us, has more risk than
no one around us.
→ The problem is we have no data that
can capture both current circumstances
and the risk score
22.
Our proposed solution
-online learning
you have 15%
chance of being
attacked if
entering this area
● We can deploy our system in the real
ship and capture data (including data
obtained by radar, and current state
of the ship) from each trip.
● The data of successful trips will be
labeled as 0.
● If a ship is being attacked by pirates,
the system will capture all data and
label as 1.
→ The risk assessment model will be
better by time.
23.
Our proposed solution
-online learning
you have 85%
chance of being
attacked if
entering this area
● We can deploy our system in the real
ship and capture data (including data
obtained by radar, and current state
of the ship) from each trip.
● The data of successful trips will be
labeled as 0.
● If a ship is being attacked by pirates,
the system will capture all data and
label as 1.
→ The risk assessment model will be
better by time.
24.
Infrared Sensors to
detectthreats
● Use a rotating infrared sensor to
get 360 degree view
● Detect nearby ships and boat, alert
the crew in case it’s an unverified
vessel
● Added layer of security, other than
Radar
● Advantages:
○ Works in low visibility
○ Can detect small vessels,
wooden boats and special
vessels even in high waves
○ Better results than Radar
○ Compact, cost effective
Rotating Infrared sensor :
● on highest point of the ship
for unrestricted 360 ° view
● Or 4 sensors along the 4
directions, to get a
unrestricted 180 ° view
Prerequisites:
25.
Detects any
boat inthe
radius coming
towards our
ship, using the
rotating
Infrared Sensor
Cross-verify
the boat on
the AIS to see
if it’s a
verified ship,
to reduce
false positives
Alert the crew
of the
unidentified
boat, it’s
direction and
velocity
26.
Low visibility
human detection
●Used for shorter range
● Pirates take advantage of low
visibility or low light conditions
to attack ships.
● An AI enabled human detection
network can be integrated with
thermal imaging to counter this
practise.
CNN
Fault tolerant
Low Latency
Centralized
Thermal camera (FLIR) at major blind
spots and intrusion prone zones
Prerequisites:
● Use night vision cameras
(less expensive)
Alternatives:
● Major functions:
○Detect human activity
○ Identify presence of weapons,
ladders (pirate equipment)
○ Can be equipped with defense
measures to reduce mobility
of the pirates.
● Optimised for longer distances
Detecting Threats
using cameras
Human
Activity
Weapons,
suspicious
tools
29.
Post Attack
Steps tobe undertaken after an
attack on a ship has occurred.
Includes tracking the pirate
ships and gathering evidence to
report the event and claim
insurance.
30.
● Tracking pirateships after pirate attacks
○ Using available satellite images of the time of attack to
analyze the pirate attack and track the motion of the pirate
ships
● Evidence for Insurance Claims
○ Pirate attack details like time, location and direction of attack
can be verified using available satellite images paired with
available AIS data
Multi-spectral Satellite Imagery
31.
Access Satellite
Images fora
particular
region and
time
Identify &
Locate ships
using ship
segmentation
models
Cross-reference
the ships with the
AIS coordinates to
identify suspected
ships
Track the
motion of such
ships or find
evidence of the
pirate attack
32.
Segmentation Model
Model: Pre-trainedDeepLabv3+
with ResNet152 encoder with
IoU+BCE Loss
Trained on around 40000
satellite images
Accuracy = 99.89%
F1 = 78.90%
F2 = 79.22%
Precision = 78.38%
Recall = 79.43%
Post-attack
tracking
● Using satelliteimaginary with ship
segmentation model, a ship’s
location and route could be
determined.
● With this, an hijacked ship could
be tracked even if communication
systems are compromised.
CNN
GPS enabled
Low Latency
Centralized
Live satellite feed access.
Prerequisites:
35.
● Use satelliteimagery and CCTV footage
to evaluate the legitimacy of a insurance
claim
● Record of Navigational and Pilotage
events during watch
● Record of ship position and behaviour of
ship at regular intervals
● Information on Cargo work and activities in
port
Evidence for insurance claim
CCTV footage
Satellite imagery for observing the ship
36.
Component Function PlacingConditions Price
Radar (X-band,
S-band)
To detect suspicious
boats around a radius
Placed on top of the
ship to get 360°
view
Can work in all
weather conditions
Already present
on ship
Infrared Sensors
(long range)
To detect suspicious
boats in a radius, can
detect small vessels,
wooden boats
4x IR sensors, one
for each direction
for a 360°
unrestricted view
Works in all weather,
better than radar in
cloudy & storms.
Works upto 12 km.
Around INR
10k-20k per
sensor
Night vision
Camera
To detect suspicious
access around a
radius
Placed all around
the ship for 360°
unrestricted view
Works in all weather,
switches to thermal
mode at night and in
low visibility
May be present
on the ship,
INR 5,000
Structure of Analysis