IEEE - Aerospace and Electronic Systems - March 2023 - 16

A Drone-Based Multisensory Payload for Maritime Pollutants Detections
Table 6.
Reference Scenarios
Feature
Purpose
Operational scenarios
Near-shore routine inspection
Surveillance flights are scheduled periodically on
areas at a short or medium range from the shore
that are particularly sensitive to pollution.
Target areas Fisheries, areas subject to environmental
protection.
Surveilled
area size
Range from
coast
Required
system
availability
Required
response
time
Long (offline processing).
Very short (online processing).
Small-medium.
Near-medium.
Mainly daytime and good weather conditions.
After-accident assessment
Assessing the situation
immediately after a maritime
pollution accident is reported.
Dense maritime traffic, prospection
areas, near shore areas.
Small-medium.
Near-medium.
24/7, all weather.
Moreover, Figure 16 shows all the EO/IR and radar
detections in a single image. The colored polygons in the
bottom left hand corner of the image are the detections
provided by the EO/IR sensors but are clearly due to the
presence of the rocks. The radar does not detect targets in
this area because its field of view is narrower than that of
the camera. Hence, in this particular case, the fusion algorithm
tags these detections as false positives.
CONCLUSION
This article shows the main achievements of the first
POSEIDON system test. The POSEIDON system is a
drone based multisensory system conceived for oil spill
and sea debris detection. The multisensory payload has
been integrated on a small drone and tested in a realistic
scenario. Results have shown that the use of multiple sensors
provides the overall system with redundancy and reliability,
complementarity, and resilience.
The on board sensors have shown good capabilities to
detect the aggregated sea litter. Particularly, from the multispectral
perspective the CNN-based detection algorithm
proved to be suitable to detect sea debris and exhibits a
detection rate greater than 80%, although to be fair, the
sea was in ideal conditions. On the other hand, however,
the orthorectification of the detections was not good
enough, because of both the noise generated by the drone
movements in the captures and the oblique captures.
16
The radar system also experienced some difficulties in
achieving good performance in the SAR image formation,
due to the drone instability during the flight and due to the
drone navigation data rate, which was much lower than
the radar PRF.
Despite this, the targets were clearly visible in the
SAR images even if part of the target was made of plastic.
The mean measured localization accuracy is 4.7 m (evaluated
as the mean value of the distances between the detections
and the USV GPS coordinates). The target has been
detected in all the SAR images. Overall, however, we
have experienced a false negative and two false positives
among the 15 processed SAR images and 31 detections.
Overall, measurement results have proven that:
The data fusion algorithm manages to fuse the inputs
of the EO/IR and RADAR sensor successfully in a
high percentage at each sensor and globally. In fact,
the information fusion algorithm has been applied first
to radar detections and multispectral detections separately.
Then, the same algorithm has been used to fuse
all the detections in a common operation picture.
The detection tracking algorithm in the fusion module
is capable of grouping detections from the same
pollutant.
The radar unit was tested successfully and was able
to detect the targets, even the aggregated plastic
objects.
IEEE A&E SYSTEMS MAGAZINE
MARCH 2023
IEEE - Aerospace and Electronic Systems - March 2023

Table of Contents for the Digital Edition of IEEE - Aerospace and Electronic Systems - March 2023
