Skip to content

E-TRAINEE: Discrimination of selected grass species from time series of RPAS hyperspectral imagery - report

Example results

Random forest classification in R

Image Classification
June Figure 1
August Figure 2
June + August Figure 3
August MNF Figure 4

Figure 5

Classification legend

Accuracy assessment

June OA=85.91% August OA=88.03%
Class Recall Precision F1-score Recall Precision F1-score
afs 29.13 48.05 36.27 29.13 56.92 38.54
cv 96.89 86.16 91.21 99.42 99.13 99.27
cxbig 0.00 0.00 0.00 0.00 0.00 0.00
desch 87.30 87.30 87.30 80.95 74.45 77.56
mol 88.12 95.10 91.48 96.62 94.07 95.33
nard 79.28 71.56 75.22 85.62 75.70 80.35
smrk 90.00 97.83 93.75 26.00 92.86 40.63
June + August OA=92.60% August MNF OA=91.89%
Class Recall Precision F1-score Recall Precision F1-score
afs 41.73 47.75 44.54 51.18 54.62 52.84
cv 99.22 98.08 98.65 99.42 99.22 99.32
cxbig 10.00 62.50 17.24 18.00 81.82 29.51
desch 90.00 86.04 87.98 90.63 84.72 87.58
mol 97.58 99.02 98.29 98.07 97.60 97.83
nard 93.23 86.81 89.91 84.57 82.82 83.69
smrk 92.00 97.87 94.84 72.00 100.00 83.72

Q&A

  • Is it possible to classify individual grass species from a mono-temporal UAS dataset with very high spatial resolution (9 cm) and spectral resolution (54 bands) with an overall accuracy higher than 85%?
    • yes, both of the mono-temporal datasets (June and August) achieved an OA above 85%
    • however, the precision and recall varied for different species
  • What is the classification accuracy of the dominant and sparse growth species?
    • dominant grass species (cv, mol, nard, desch) reach excellent accuracy; sparse species not so much
    • abundance/density/homogeneity of the species and its cover are essential for classification accuracy
  • Can we reach higher accuracy using time series of intra-seasonal data? How significant are the differences?
    • accuracy increased with spatial resolution; multitemporal composite from June and August 2020 reached the highest OA of 92.6%
  • Optional: Can image data transformation that reduces noise and data dimensionality (MNF transformation) produce better results than the original hyperspectral dataset?
    • yes, the MNF transformed image OA is 91.89%, which corresponds to a 3.96 % increase when compared to the classification of the original August dataset

Back to case study

Proceed by returning to Discrimination of selected grass species from time series of RPAS hyperspectral imagery