logo SBA

ETD

Digital archive of theses discussed at the University of Pisa

 

Thesis etd-03142024-170556


Thesis type
Tesi di dottorato di ricerca
Author
PULPITO, OSVALDO
URN
etd-03142024-170556
Thesis title
TARGET DETECTION IN MARITIME INFRARED SURVEILLANCE SYSTEMS
Academic discipline
ING-INF/03
Course of study
INGEGNERIA DELL'INFORMAZIONE
Supervisors
tutor Prof. Acito, Nicola
relatore Prof. Corsini, Giovanni
relatore Prof. Diani, Marco
commissario C.V. (AN) Reversi, Stefano
Keywords
  • automatic surveillance systems
  • data-driven
  • dim target detection
  • Dynamic Mode Decomposition
  • electro-optical
  • extended target detection
  • infrared
  • machine learning
  • maritime scenario
  • moving target detection
  • naval targets
  • real-time
  • Robust Principal Component Analysis
  • saliency
Graduation session start date
05/04/2024
Availability
Withheld
Release date
05/04/2064
Summary
MILITARY ships are very complex and highly technological means. To gain situational awareness, they take advantage of surveillance systems made of diverse sensors, including Radars, Sonars, and Electro-Optical (EO), each one with its advantages and disadvantages. One of the main advantages of EO sensors is that they enable the identification of the targets, which is an arduous task for systems based on the other types of sensors. Furthermore, they are passive, meaning they do not need to emit any ElectroMagnetic (EM) radiation. In the military field, where reducing your visibility is crucial, active sensors, which, on the opposite, need to emit some signal to do their job, are easily interceptable by opponents, and, for this reason, their use is not always feasible. EO sensors are classified, based on the EM band they can detect, into visible and InfraRed (IR). While visible sensors detect the energy of an external light source (like the Sun), which is then reflected by the elements composing the scenario, IR sensors can detect the radiated energy autonomously emitted by hot bodies. Such difference is the reason for the main advantages of IR sensors: they do not need sunlight, meaning they enable night vision; they reduce the representation dynamic of the regions of the scene characterized by homogeneous temperature (like the sea); they make the bodies at different temperatures more easily distinguishable, regardless of their mimetic abilities.
Automatic surveillance is the engineering branch that aims to monitor and analyze events occurring in a specific area without continuous human intervention. Of course, different sensors require different approaches. In the EO field, a lot of research has been and still is conducted, led by the diffusion of low-cost sensors and by the spread of Artificial Intelligence (AI). Historically, due to the expansiveness of IR sensors, most of the research was made on visible sensors, and the IR surveillance systems were mainly deployed for military purposes. Nowadays, the cost of IR sensors is also lowering, encouraging research in this field as well.
The main features of an automatic surveillance system are the sensor, the detector, the tracker, and the classifier. After giving a brief overview of each contributor, this thesis focuses on the detector and examines different algorithms based on spatial and temporal features. This work also contributes by presenting two Moving Target Detection (MTD) algorithms. One of them is based on a relatively recent data-driven technique called Dynamic Mode Decomposition (DMD), which is implemented in a spatial-multiscale form to help detect both small and extended targets. The other one is based on another data-driven technique called Robust Principal Component Analysis (RPCA) implemented in an online fashion, which uses saliency maps to take advantage of both spatial and temporal features.
Finally, this thesis introduces specific IR datasets collected by the Italian Navy and by the Centre for Maritime Research and Experimentation (CMRE) of the NATO’s Science and Technology Organization (STO), which have been used to evaluate the performance of the examined detector algorithms.
File