Crop classification of multitemporal PolSAR based on 3-D attention module with ViT

Multitemporal polarimertic SAR is considered to be very effective in crop classification and cultivated land detection, which has received much attention from researchers. Currently, for most multitemporal polarimetric SAR data classification methods, the simultaneous temporal–polarimetric–spatial f...

Descripción completa

Detalles Bibliográficos
Autores: Yin, Qiang, Lin, Zhiyuan, Hu, Wei, López Martínez, Carlos|||0000-0002-1366-9446, Ni, Jun, Zhang, Fan
Tipo de recurso: artículo
Fecha de publicación:2023
País:España
Institución:Universitat Politècnica de Catalunya (UPC)
Repositorio:UPCommons. Portal del coneixement obert de la UPC
Idioma:inglés
OAI Identifier:oai:upcommons.upc.edu:2117/390748
Acceso en línea:https://hdl.handle.net/2117/390748
https://dx.doi.org/10.1109/LGRS.2023.3270488
Access Level:acceso abierto
Palabra clave:Synthetic aperture radar
Polarimetric remote sensing
Imaging systems in geophysics
Crop classification
Multitemporal PolSAR
Temporal–polarimetric–spatial characteristics
Vision transformer (ViT)
Radar d'obertura sintètica
Imatgeria en geofísica
Àrees temàtiques de la UPC::Enginyeria de la telecomunicació::Radiocomunicació i exploració electromagnètica::Radar
Descripción
Sumario:Multitemporal polarimertic SAR is considered to be very effective in crop classification and cultivated land detection, which has received much attention from researchers. Currently, for most multitemporal polarimetric SAR data classification methods, the simultaneous temporal–polarimetric–spatial feature extraction capability has not been exploited sufficiently. Also, the diversity of different time and different polarimetric features has not been taken into account sufficiently. In this letter, we propose a classification model that combines a dual-stream network as a temporal–polarimetric–spatial feature extraction module with vision transformer (ViT) called temporal–polarimetric–spatial transformer (TPST) to address the above problems. Second, a 3-D convolutional attention module that enables the network to weight the temporal dimension, polarimetric feature dimension and spatial dimension is developed, according to their importance. Experimental results on both the UAVSAR and RADARSAT-2 datasets show that the proposed method outperforms ResNet.