| Summary: | Satellite missions which collect hyperspectral data provide detailed spectral information at a lower cost than airborne
missions. The newly launched PRISMA hyperspectral mission provides greater swath coverage than the previous Hyperion
hyperspectral mission. This study aims to assess the potential use of bitemporal PRISMA datasets for change detection
(CD), by means of the clustering of Gaussian mixture models (GMM) with inputs to the magnitude component derived
from change vector analysis (CVA), distance metrics and principal component analysis (PCA) from stacked data, and
image-differenced layers. In addition, a change detection method using a combination of the modified z-score from imagedifferenced
layers and a spectral angle mapper (SAM), SAMZID-TAN, was also assessed. Overall accuracies for CD in our
results varied between 50.90 and 78.83%, with the producer’s and user’s accuracies for the change class ranging from
69.74 to 84.21% and 38.13–66.29%, respectively. SAMZID-TAN was the most accurate method for CD. Moderate CD
accuracy was achieved using PRISMA due to the effects of misregistration and image striping, which contributed to
misclassification. In future research, proper pre-processing should be performed in order to avoid the detection of false
positives when using hyperspectral data.
|
|---|