Re-using field reference data in space and time for vegetation mapping: the potential of semisupervised and active LEARNing techniques (RE-LEARN)

Start-End 01/02/2012 -  30/06/2014
Programme STEREO 2
Contract SR/00/154
Objective The main objective of this project is to explore and further develop the current semi-supervised and active learning techniques for the specific application of vegetation mapping. In particular, our aim is to tackle the problem of limited ground reference data by investigating the re-use of vegetation reference data. As a prototype problem, we envisage the classification of vegetation from hyperspectral images acquired at the same location on different occasions or at different locations containing similar vegetation types. The goal is then to design strategies for the re-use of reference samples obtained from one occasion or location to improve the classification at the other occasions or locations.
Method
  • WP 1: Project management and dissemination
  • WP 2: Remote sensing data acquisition and pre-processing
  • WP 3: Development of semi-supervised and active learning strategies
  • WP 3.1: Study and implementation of the relevant state-of-the-art semi-supervised and active learning techniques for hyperspectral image classification.
  • WP 3.2: Development of specific strategies for re-using labeled data on similar multispatial and multitemporal data.
  • WP 4: Validation of semi-supervised and active learning strategies
Result The researchers will start from the state-of-the-art kernel- and graph-based semi-supervised and active learning techniques. In a first step, the most relevant techniques are studied and implemented. In particular, the researchers will pay specific attention to the  adaptability to spatially/temporally related imagery.

When sufficient expertise is built up with respect to the domains of semi-supervised and active learning, strategies will be developed for the specific goal of re-using labeled data obtained from one occasion or location to improve the classification at the other occasions or locations. In particular, the researchers will pay specific attention to the available strategies of domain adaptation and the problem of spectral shift.

The implemented state-of-the-art techniques and the specific  strategies for the re-use of labeled data on different occasions and locations will be applied to the problem of vegetation monitoring, and more in particular heathland habitat monitoring. The goal is to apply the framework using only very limited new field reference data, to a newly acquired hyperspectral image of the 'Kalmthouthse heide' heathland area that was previously studied and a new heathland area which has restricted access.
Website link
Team
Team Member: HAEST Birgen VITO - Remote Sensing
Project Leader: SCHEUNDERS Paul UA - Visielab
Team Member: CAMPS-VALLS Gustavo University of Valencia
Location
Sensors used
Applications
Related presentations RE-LEARN - RE-using field reference data in space and time for vegetation mapping: the potential of semisupervised and active LEARNing techniques
APEX for heathland and coastal vegetation monitoring applications: first results and ongoing activities
17 small project: RELEARN
Related publications
Datasets used