SEAGULL: Low-Cost Pervasive Sensing for Monitoring and Analyzing Underwater Plastics

ACM/IEEE IoTDI full paper presentation

1. SEAGULL: Low-Cost Pervasive Sensing for
Monitoring and Analyzing Underwater Plastics
Huber Flores, Agustin Zuniga,Marko Radeta, Zhigang Yin, Mohan Liyagane,
Naser Motlagh, Ngoc Nguyen, Sasu Tarkoma, Moustafa Youssef, Petteri Nurmi
1
May 14, 2024, CPS-IoT Week, Hong Kong
2024

2. Importance
•Plastic pollution is a global concern, affecting everything from
marine ecosystems to climate change and even human health.
2
Source: https://www.naturepl.com/stock-photo-divers-swimming-past-a-
plastic-bag-floating-underwater-resembling-a-nature-image01619091.html
Source: https://www.pikrepo.com/frqdg/plastic-bottle-floating-on-water
Floating Underwater
Source: https://www.pinterest.com/pin/484840716132563124/
Preventing the pollution caused by plastics is extremely important
Mixing with the ecosystem

3. Finding plastics underwater is non-trivial
•Visual inspections (humans) and object recognition fail to identify
plastics
3
Source: https://www.independent.co.uk/environment/plastic-pollution-coral-
reefs-disease-damage-seas-oceans-cornell-university-a8178156.html
Source: https://www.independent.co.uk/environment/plastic-pollution-coral-
reefs-disease-damage-seas-oceans-cornell-university-a8178156.html
Coral wrapped in a
plastic bag
Plastic bottle blended
with coral
How to identify and classify plastics underwater accurately?
Source: https://twitter.com/4ocean/status/993913424332673025/photo/1
Degraded plastic
bottle in sea floor

4. Automated solutions
4
Current solutions can identify plastics (to some
extent) but it is not possible to classify them
Source: https://newatlas.com/boxfish-rov-underwater-camera-drone/55612/
Source: https://www.pinterest.com/pin/558094578803524390/
RIC = Resin Identification Code

5. SEAGULL (Contributions)
•Low-cost sensing solution: SEAGULL uses inexpensive light
diodes (or LEDs) and a photo-receptor, together with a pipeline
implementing convolutional sparse coding
•Testbed (Sea and river): Two different testbeds to analyze the
performance of our method
•New insights: Classification performance of identification of plastics
underwater (turbidity, turbulence and so on)
5
SEAGULL correctly
distinguishes between the
main plastic with an 85%
accuracy

6. SEAGULL: Overview
6
•Embedded
into ROVs
and UAVs
•Embedded
into scuba
diving gear

7. SEAGULL: Pipeline
•Simply explained
7

8. SEAGULL: Pipeline
•Simply explained
8

9. SEAGULL: Pipeline
•Simply explained
9
•Input
•Dictionary
•Sparse representation
•Sparse coding optimization and
reconstruction
•Output

10. SEAGULL: Pipeline
•Simply explained
10

11. SEAGULL: Evaluation
•Controlled evaluation
11Plastic characterization using light values
Plastic samples
Plastic characteristics
Samples

12. Light characterization performance
12

13. Sensor performance
13
50 samples
Source: https://www.recycling.com/why-is-plastic-recycling-important/

14. Water conditions
•Luminosity
14
AL = Ambient Light
AD = Ambient Dark
Insight: Luminosity causes
drift in the measurements,
requiring calibration of the
sensor

15. Water conditions
•Turbidity
15
AL = Ambient Light
AD = Ambient Dark
Insight: It does not cause
any drastic change in the
measurements

16. Water conditions
•Frozen water
16
AL = Ambient Light
AD = Ambient Dark
Insight: It changes slightly
fingerprints in transparent
and translucid objects

17. Water conditions
•Turbulence
17
AL = Ambient Light
Insight: Introduces more
variance on sensor data,
requiring sampling from
multiple points of the target
object

18. Water conditions
•Turbulence
18
AL = Ambient Light
Insight: Introduces more
variance on sensor data,
requiring sampling from
multiple points of the target
object
Insight: Combination of
accelerometer and
gyroscope can detect when
to take measurements with
the light sensor

19. Classification performance
19

20. Practicability: Prototypes
20
UAV augmentation
Scuba diver gear

21. Practicability: Experiments
21

22. Practicability: River
22
Insight: Better integration
can improve the results
(upward force)

23. Practicability: Sea
23
Insight: Manual placement
provides better stability of
samples

24. Summary and conclusions
•Low-cost sensing solution: SEAGULL uses inexpensive light
diodes (or LEDs) and a photo-receptor, together with a pipeline
implementing convolutional sparse coding
•Testbed (Sea and river): Two different testbeds to analyze the
performance of our method
•New insights: Classification performance of identification of plastics
underwater
24

25. Questions? Thank you!
Huber Flores([email protected])
Agustin Zuniga ([email protected])
Marko Radeta ([email protected])
Zhigang Yin ([email protected])
Mohan Liyanage ([email protected])
Naser Motlagh ([email protected])
Ngoc Thi Nguyen ([email protected])
Sasu Tarkoma ([email protected])
MoustafaYoussef([email protected])
Petteri Nurmi ([email protected])
25