International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B8, 2012 
XXII ISPRS Congress, 25 August - 01 September 2012, Melbourne, Australia 
soil, Cinnamon soil, Dark-brown soil, Brown coniferous forest soils, Fluvo-aquic soil. ) 
4. LS ASSESSMENT AND ANALYSIS BASED ON ANN 
4.1 ANN concepts 
ANNS are generic non-linear functions that have been broadly 
used to solve many problems like confirming weight values and 
classification, with processing unit, network topology, and 
training rules. Multi-Layer Perception (MLP) as a frequently 
used ANN, includes input, output, and one or more hidden 
layers between in-out layers. Meanwhile, the number of 
neurons from the input to output layers is typically fixed by the 
model designed. By trial and error, the neurons and the number 
of hidden layers can be determined (Gong, 1996). There are 
three steps involved in ANN data processing, including: the 
training, the weights confirmation, the classification. Training 
data from input neurons are processed through hidden nodes to 
obtain output values. If the input that a single neuron j with the 
only one hidden layer, may be expressed as: 
1 
net, = > wp, (2) 
izl 
Where w; represents the weights between the node / of input 
layer and the node j of hidden layer; p; is the input data; / 
represents the number of input layers. The relative probabilities 
of factors into LS values may also be regarded as the problem 
of judging weights. Input layers include the above 9 factors: 
Table 3 Occurrence frequencies of different input factors 
Lithology, Convexity, Gradient, Aspect, Elevation, Soil 
property, Vegetation cover, Flow, Fracture; the ANN output 
may be considered as the measurements of the occurrence of 
landslide (Figure 1). 
Hidden layer 
Ü 
    
Elevation 
Figure 1. ANN structure chart of LS analysis 
4.2 Factors acquisition of the samples 
With the previous 1,200 records, and the equation (l), 
occurrence frequencies of different input factors (quantitative 
values) had been obtained (Table 3). 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
Frequencies Classification 
of factors 1 2 3 4 5 6 7 8 
Lithology 0.12 0.09 0.17 0.24 0.15 0.11 0.04 0.08 
Convexity 0.58 0.26 0.16 0 0 0 0 0 
Gradient 0 0.03 0.16 0.29 0.27 0.25 0 0 
Aspect 0.14 0.16 0.18 0.17 0.15 0.09 0.09 0.08 
Elevation 0.01 0.14 0.17 0.22 0.19 0.19 0.08 0 
Soil 0.38 0.36 0.15 0.02 0.05 0.02 0 0.02 
property 
Vegetation 0.34 0.25 0.23 0.11 0.07 0 0 0 
Coverage 
Flow 0.30 0.25 0.24 0.12 0.06 0 0 0 
Fracture 0.36 0.20 0.15 0.13 0.11 0.05 0 0 
  
  
  
  
  
  
The occurrence frequencies reflect the number of landslide 
reports, and output data were set as the summary of different 
factor frequencies (3). 
Z=X+X, + +X, 3) 
Where Xn is quantitative value from every factor; n is the 
48 
number of factors. Z for the ANN trainings would be 
normalized between 0 and 1. 
4.3 LS mapping and analysis 
According to the above methods mentioned, the qualitative 
Int 
values * 
summar 
consider 
layer is 
network 
training 
implem 
Error 
+ 
Some | 
website 
http://l: 
Disaste 
Consor 
Gather 
is abou 
the LS 
more s 
landsli 
then p 
factor | 
fcis 
  
wi 
From 
highei 
be coi 
rainfa 
landsl