The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008 
Network Outputs 
(Crop Heavy Metal 
Stress Level) 
Figure 1. A schematic diagram of dynamic fuzzy neural-network model for crop stress level assessment 
Each layer in a FNN model contains sufficient numbers of 
neurons which depend on the specific application. The neurons 
in a layer are connected to the neurons in the next successive 
layer and each connection carries a weight (Atkinson P.M. et 
al., 1997). In this model, the input layer receives the data from 
three hyperspectral vegetation indices. Hence, there are three 
neurons in this layer, corresponding to three influencing factors 
in crop stress level assessment. The hidden and output layers 
process the data actively. The number of hidden layers and 
their neurons are determined by trial and error (Gong, 1996). 
By varying the number of neurons in hidden layers, the neural 
network is run for several times to identify the most 
appropriate neural network architecture based on training and 
testing accuracies. The number of neurons in output layers is 
determined in reference to the National Standard (GB15618- 
1995) and national food standards. Table 1 shows the soil 
heavy metal pollution classification standard. There are four 
values in output layer, corresponding to four levels of crop 
heavy metal stress, including pollution-free, light pollution 
stress, moderate pollution stress and severe pollution stress. 
They are presented by number 0, 1, 2 and 3. 
Classification 
As 
Hg 
Cd 
Pb 
Cr 
Cu 
Ni 
Zn 
wl(mg.Kg') 
pH<6.5 
40 
0.3 
0.3 
250 
150 
50 
40 
200 
GB II 
6.5<pH<7.5 
30 
0.5 
0.3 
300 
200 
100 
50 
250 
pH>7.5 
25 
1.0 
0.6 
350 
250 
100 
60 
300 
Table 1. National Standard (GB 15618-1995): Soil Heavy Metal Pollution Classification