Deep Temporal Convolutional Autoencoder for Unsupervised Representation Learning of Incoherent Polsar Time-Series

Thomas Di Martino (1) , Regis Guinvarc'H (2) , Laetitia Thirion-Lefevre (2) , Elise Koeniguer (1)

1 DTIS, ONERA, Université Paris Saclay [Palaiseau]
2 SONDRA - Sondra, CentraleSupélec, Université Paris-Saclay

Temporal Convolutional AutoEncoders are used as feature extractors to project time series onto a latent space where similarity detection can be easily performed. This model can generate accurate descriptors of the temporal profile of the input time-series. We apply this algorithm to PolSAR S1 uncoherent SAR time series where the model learns highly discriminative data representations. This reduction method is compared to others such as PCA or Temporal Averaging and is shown to outperform them when leveraging the learnt representation using K-Means clustering.

Mots clés: Autoencoder; Deep Learning; dual polarization SAR; Time Series; Clustering

Domaines: Sciences de l'ingénieur [physics]; Physique [physics]; Mathématiques [math]; Informatique [cs]

Thomas Di Martino, Regis Guinvarc'H, Laetitia Thirion-Lefevre, Elise Koeniguer. Deep Temporal Convolutional Autoencoder for Unsupervised Representation Learning of Incoherent Polsar Time-Series. IGARSS 2021, Jul 2021, Bruxelles (virtual), Belgium. ⟨hal-03345533⟩