Table 6-1. Detection Matrix for Objects at Various GSDS

EM 1110-2-2907

1 October 2003

roofs. A preliminary analysis suggests that this is possible because of the strong covariance

displayed by roofs shingled with monochromatic materials. Any automated process devel-

oped would need to address the limitations posed by non-monochromatic shingles (which

would appear spectrally mixed and indistinguishable from damaged roofs).

(4) A vegetation analysis was also explored to test the resolution required to accu-

rately describe tree type and condition. At the 1-ft (30.5 cm) resolution, researchers were

able to determine leaf on/off conditions (data were collected in February). However, at this

resolution it was not possible to delineate any details regarding leaf morphology. At the 8-in

(20.3 cm). resolution, palms were distinguishable, although it was not possible to differenti-

ate broad versus narrow leaves.

f. Conclusions. Evaluation of the Emerge sensor led to the development of a detection

matrix. This matrix reviews the capabilities of the sensor at various spatial resolutions for all

objects studied (see Table 6-1). This study determined that Emerge could adequately meet

the requirements of emergency management systems. High-resolution data can be acquired

within 4 hours of the plane's landing. This includes the time needed for pre-processing

(orthorectification and the production of geo-TIFF files for CD-ROM and ftp). Shingles and

tarpaper are not resolvable, though rafters and plywood are at the 2-ft (~61 cm) resolution.

For high-resolution images, a medium sun angle increased roof detail. Palm trees and leaf

on/off conditions can be visually identified at the 8-in (20.3 cm). resolution; however,

broad-leafed trees cannot be distinguished from narrow-leafed trees. The only limitations

placed on these data centered on over-saturation and sensor inability to distinguish tree

types. The covariance displayed by band 1 relative to band 2 indicates the potential success

for developing an automated algorithm to locate and count damaged roofs.

Table 6-1

Detection Matrix for Objects at Various GSDS

Objects/GSD

3-ft (91.4)

2-ft (61 cm)

1-ft (30.5 cm)

8-in. (20.3 cm)

Roof rafters

Not visible

Barely visible

Often visible

Visible

Shingles/tarpaper

Can

Can often

Can determine

(other) vs. plywood

sometimes

separate

wood vs. other

separate

cover

Rafters in 3-band

Causes rafter

saturation

detail loss

Broad-leaf vs. narrow-

Cannot

Can determine

Palms are always

leaf

separate

leaf on/off

visible

All in cloud shadow

Degrades

Some info

image

recoverable

Roofs as a function of

Best detail,

sun zenith angle

near zero

medium angle,

angle,

shadow casting

overhead sun

All in 1, 2, 3 RGB, 2∑

Enhances

Enhances imagery

stretch

imagery

Point of Contact: Robert Bolus, Phone: (603) 646-4307

6-7