questions/notes for 23/03 meeting

[pedrorohde]

• benchmark
  • performance
  • model and features

• 2

  • patches: how to extract them
• which period to take when splitting across years/seasons (may reduce total number of examples)
  • currently december-november because of data
  • ideally split at lowest level for deciduous (~winter)
• features
  • validate current ones
  • more ideas
• time series smoothing
  • fit them to a function
  • which function(s)?
  • see scipy.optimize.curve_fit
  • other methods
• barycenters
  • make sense?
  • see tslearn.barycenters
• extract features from the images (possibly too much work)

[pedrorohde]

our approach so far:

• extract time series for each index for each parcel (averaging over parcel, but also std and median) ignoring clouds ndvi, gndvi, ndmi, evi, avi parcels have an irregular shape => average over something else? e.g. patches

• split the time series by season => needed? does inter-season information help?

• extract features from time series: winter/summer difference, average derivative, day of max, amplitude (max-min) => more features? e.g. papers simply use index values at specific dates

• results: over hold out test set over 90% accuracy for deciduous x evergreen ~80-85% accuracy for species (but imbalance in classes) example confusion matrix:

  array([[ 17,   0,   4,   0,   0,   3,   0,   0,   0,   4,   0,   0,   0],
     [  0,   2,   0,   1,   0,   2,   0,   0,   0,   7,   0,   0,   0],
     [  3,   0,   3,   0,   0,   5,   0,   0,   3,   1,   0,   0,   0],
     [  0,   0,   1,   4,   0,  18,   0,   0,   0,   1,   0,   0,   0],
     [  0,   0,   0,   0,   0,  13,   0,   0,   0,   0,   0,   0,   0],
     [  0,   0,   1,   1,   0, 505,   0,   0,   0,   0,   0,   0,   0],
     [  0,   0,   0,   0,   0,   1,   0,   0,   0,   0,   0,   0,   0],
     [  0,   0,   0,   0,   0,   0,   0,   1,   0,   0,   0,   0,   0],
     [  0,   0,   1,   0,   0,   2,   0,   0,   0,   1,   0,   0,   0],
     [  4,   2,   1,   3,   0,   3,   0,   0,   0,  65,   7,   0,   0],
     [  0,   0,   0,   0,   0,   1,   0,   0,   0,  14,   1,   0,   0],
     [  0,   0,   0,   0,   0,   2,   0,   0,   0,   5,   1,   0,   0],
     [  1,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0]])

  => what metrics to use => how to deal with imbalance

[pedrorohde]

split problem in two parts: deciduous vs evergreen then species just index values is generally not enough