| KEYWORD |
Image compression and deep learning onboard Earth observation satellites
Parole chiave COMPRESSIONE, COMPUTER VISION, DEEP LEARNING
Riferimenti ENRICO MAGLI, DIEGO VALSESIA
Gruppi di ricerca ICT4SS - ICT FOR SMART SOCIETIES, Image Processing Lab (IPL)
Tipo tesi RESEARCH
Descrizione This thesis is about the use of deep learning on board satellites in order to improve the effectiveness of image compression. Roughly half of the images acquired by Earth observation satellite missions are significantly covered by clouds, making them of little use. Recently planned missions such as the CHIME hyperspectral and likely the Sentinel-2 Next Generation multispectral constellations will include a "cloud screening" algorithm to be run onboard the satellite, which generates a cloud classification map. Such map is employed as input to the image compression algorithm in order to represent with lower quality the cloudy pixels, thereby obtaining a significant amount of data volume reduction.
This thesis has two objectives: 1) to extend the current standard for onboard image compression so that it can employ the compression map, also performing lossless compression of the map itself, and 2) to test the performance of the compression algorithm with classification maps generated by deep learning methods.
This activity is performed under a contract funded by the Italian Space Agency. Full-time work on the project is required during the thesis, and a monetary compensation is foreseen. For more information and to apply, send an email to enrico.magli@polito.it including a cv and list of L3 and LM exams with grades. Only students with GPA >= 27/30 will be considered for this thesis.
Conoscenze richieste Basic deep learning skills, including Python and usage of Pytorch or Tensorflow. Ability to write programs in C language.
Scadenza validita proposta 29/03/2025
PROPONI LA TUA CANDIDATURA