(a1)
(a2)
(a3)
(b1)
(b2)
(b3)
(c1)
(c2)
(c3)
(d1)
(d2)
(d3)
(e1)
(e2)
(e3)
Figure 8. Comparison of landslide detection between two methods, together with ground truth images: (a1) through (n1) land-
slide detection by our method; (a2) through (n2) landslide detection by the semi-automatic method in (Li
et al
., 2016); (a3)
through (n3) ground truth landslide regions by visual interpretation.
Continued on next page.
Accuracy of study regions I and J are upmost 18 percent lower
than the semi-automatic method. That accords with the sev-
eral false alarms in Figure 8, which is acceptable for hazard
detection. For most cases, our method is reliable, effective,
and robust in landslide detection from the performance analy-
sis in Table 1.
Furthermore, we calculated the computing time (s) of our
method on a laptop with a processor of Intel (R) Core (TM)
i5-4300M CPU @ 2.60GHz for each sub region to evaluate its
speed and compared with that of semi-automatic method (Li
et al
., 2016), as shown in Table 2. The computation time of
our method is almost half of that by semi-automatic method
which added further possibility of our method being practi-
cally applicable for emergency response where time is of
prime importance.
360
May 2017
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
