173
v.3.0 (Pix4D 2016). The Sony and Mobius output mosaics were generated without
spatial referencing, using the automatic batch-process mosaic tool in Adobe
Photoshop CC 2015. All of the mosaics were transferred to ERDAS
®
Imagine
v2016(64bit) (ERDAS IMAGINE 2016) mapping software to be spatially aligned
with one another and overlaid.
Mosaic Alignment
Non-georeferenced mosaics were manually tied to the spatially accurate DJI mosaic
using visually derived Manual Tie Points (MTPs). The Sony and Mobius mosaics
were transformed using 3rd order polynomial spline transformation to conserve the
exact overlay of MTPs and effectively mimic a georeferenced Ground Control Point
(GCP). Examples of MTPs used in this study are Recognisable Visual Points
(RVPs), which stand out in an image. Common RVPs are - manhole covers, forked
branches, or obvious tree apexes etc. and were easily seen in the three mosaics to tie
images together at these locations. A greater number of MTPs will result in more
areas accurately overlapped and thus a more precise overlay. MTPs were used for
two reasons; there are unlikely to be many feasible locations, which are clearly visible from the air, for accurate GCPs in a large catchment of woodland, and entering
the site poses a risk of spreading the infection.
Multispectral Image Generation
We used the manufacturer’s spectral response curves to identify the best wavelengths to extract into bands (Sony 2016; Mobius 2016). A 10 nm bandwidth was
extracted around the point of highest quantum efficiency % on the y-axis (or the
wavelength at which the most light is absorbed by the sensor/ graphical peak – see
Fig. 2). The NIR extracted bandwidth was expanded to 50 nm to include a greater
portion of NIR, the most important spectral region for plant identification
(Buschmann et al. 2012).
We created a spatial model in ERDAS
®
Imagine v2016 (64bit) mapping software
to extract the bands from the three mosaics and recombine these into a 7 band
multispectral mosaic (Fig. 3). Bands were ordered into ascending wavelength in
order to appropriately retrieve spectral signatures from the features in the image.
The final mosaic had the following discrete bands centred at 440, 450, 540, 550,
600, 610 (all 6 with 10 nm width), and 750 nm (50 nm width).
Mapping the Distribution of Understorey Rhododendron Ponticum Using…
v.3.0 (Pix4D 2016). The Sony and Mobius output mosaics were generated without
spatial referencing, using the automatic batch-process mosaic tool in Adobe
Photoshop CC 2015. All of the mosaics were transferred to ERDAS
®
Imagine
v2016(64bit) (ERDAS IMAGINE 2016) mapping software to be spatially aligned
with one another and overlaid.
Mosaic Alignment
Non-georeferenced mosaics were manually tied to the spatially accurate DJI mosaic
using visually derived Manual Tie Points (MTPs). The Sony and Mobius mosaics
were transformed using 3rd order polynomial spline transformation to conserve the
exact overlay of MTPs and effectively mimic a georeferenced Ground Control Point
(GCP). Examples of MTPs used in this study are Recognisable Visual Points
(RVPs), which stand out in an image. Common RVPs are - manhole covers, forked
branches, or obvious tree apexes etc. and were easily seen in the three mosaics to tie
images together at these locations. A greater number of MTPs will result in more
areas accurately overlapped and thus a more precise overlay. MTPs were used for
two reasons; there are unlikely to be many feasible locations, which are clearly visible from the air, for accurate GCPs in a large catchment of woodland, and entering
the site poses a risk of spreading the infection.
Multispectral Image Generation
We used the manufacturer’s spectral response curves to identify the best wavelengths to extract into bands (Sony 2016; Mobius 2016). A 10 nm bandwidth was
extracted around the point of highest quantum efficiency % on the y-axis (or the
wavelength at which the most light is absorbed by the sensor/ graphical peak – see
Fig. 2). The NIR extracted bandwidth was expanded to 50 nm to include a greater
portion of NIR, the most important spectral region for plant identification
(Buschmann et al. 2012).
We created a spatial model in ERDAS
®
Imagine v2016 (64bit) mapping software
to extract the bands from the three mosaics and recombine these into a 7 band
multispectral mosaic (Fig. 3). Bands were ordered into ascending wavelength in
order to appropriately retrieve spectral signatures from the features in the image.
The final mosaic had the following discrete bands centred at 440, 450, 540, 550,
600, 610 (all 6 with 10 nm width), and 750 nm (50 nm width).
Mapping the Distribution of Understorey Rhododendron Ponticum Using…