| KEYWORD |
Deep learning for onboard satellite image processing
Parole chiave 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 Satellites for Earth Observation are increasingly being equipped with image processing capabilities, in order to extract information from images directly onboard the satellite, without waiting for image download and preparation at the ground segment. The objective is to employ deep neural networks for two tasks: image compression and image analysis, each of which may be the topic of a thesis. Image compression can be performed using autoencoders or predictors, and may generate features that are useful not only for compression, but also for image analysis. The image analysis task may involve various applications including cloud detection/segmentaiton, as well as object detection and anomaly detection.
We envisage the use of a new neural network architecture that works in a line-based fashion (it only requires the previous line of the image instead of a large number of lines or the whole image), which is very suitable for satellite onboard processing, especially for imaging sensors based on a linear array, which generate one image line at a time.
Conoscenze richieste Basic deep learning skills, including Python and usage of Pytorch or Tensorflow
Scadenza validita proposta 13/11/2025
PROPONI LA TUA CANDIDATURA