Biology Reference
In-Depth Information
3.4.2 The Maximum likelihood estimator (Mle) Method
The maximum likelihood estimator (MLE) method, one of the most com-
mon classification methods, is a procedure based on finding the values of the
model parameters that maximizes the likelihood function or distribution.
The likelihood function is calculated based on statistics, and data can take
the form of a normal, Poisson, Bayesian, or Bernoulli distribution, among oth-
ers. In remote sensing applications, normal distributions are often assumed.
After establishing class type from training sites, a pixel/entity from applica-
tion sites is assigned to the classification that has the highest likelihood value
(Duda et al. 2000).
3.4.3 The Neural Network Method
Unlike the maximum likelihood estimator and parallelpiped methods, the
neural network (NN) method does not require the knowledge of the model's
distribution function and, therefore, belongs to the nonparametric class of
statistical methods. The NN method is a powerful universal approximator.
Quite a few types of neural networks can be used for the unsupervised and
supervised classification of remote sensing data (Principe et al. 2000). Some
examples where neural networks have been used in ground cover, vegeta-
tion, and land use classification can be found in the literature (Bocco et al.
2007; Debeir et al. 2001).
3.4.4 Spatial Techniques
In most ground cover classifications, only spectral information associated
with a pixel is used. That is, a pixel is assumed an isolated entity unaware
and independent of all the neighboring pixels. This assumption may be suf-
ficiently valid when the size of a pixel is large. However, for remote sensing
data with high spatial resolution or when pixel size is on the order of 1 m,
per-pixel spectral classification is no longer adequate and may result in low
classification accuracy. The information of the neighboring pixels or neigh-
borhood can be included in the classification by using textural or contextual
information of the neighborhood (Petrou and Sevilla 2006). Only by includ-
ing spatial information can accuracy in classification be restored.
3.5 Modeling Techniques with Utilization
of Remote Sensing Data
Quantitative study of infectious disease transmission dates back as early
as 1760, when Daniel Bernoulli investigated the effect of inoculation on the
 
Search WWH ::




Custom Search