You are using an outdated browser that does not fully support the intranda viewer.
As a result, some pages may not be displayed correctly.

We recommend you use one of the following browsers:

Full text

Mapping without the sun
Zhang, Jixian

the condition probability p^X/w^ be called as
multi-variable probability density function, and its expression
P&l w k) = n .„J, c , /2 expf|(X-prfSÎiX-ti)]
(2ft) I *->/t I ^
1 n k
"k H
s k =- L tZk x j -Mk)( x j -Mk) T ]
n k 1 7=1
In this formula, m is the band number; ¡Ll k is the mean value
vector of training samples; S k is the covariance matrix; Yl k
is the pixel number in category W k .
3.1 Detection Principle
Basic principle: At First this paper carries the MLC on the sole
period RS image, and obtains the classification results. Then
makes the pretreatment of space transformation separately
between the present land use map and the classified RS image,
and makes the two have the very good geometry space
uniformity. Then overlays the present land use map and RS
image in space, makes the land codes be related to spots, and
ascertains pixels contained in each spot by Vertical Line Law.
According to the Maximum Likelihood Classification results,
this paper carries out the category attribute judgment in the
land use spot unit utilizing the Spot Class Statistic and Judge
Function, and makes the attribute matching between the
classified and judged spot units and the object codes, then
detects the changed spots, marks them in high-light, and
calculates those areas.
3.2 Detection Flow and Key Technique
Detection procedure has two key questions to need to solve in
practical realization, which can be divided explaining as
(1) Determining pixel sets contained in each plot
There are several methods to examine whether the given point
locates in the given region or not, such as the Radial Law, the
Comer Law, the Area Law, and the Vertical Line Law. This
paper uses the Vertical Line Law to processing, and its
principle is:
Obtains all pixel sets in the irregular spot. It can be solved by
judging whether the point is in the polygon or not.
The practical algorithm is the Vertical Line Law: makes the
perpendicular line to under from point needed to judge, and
calculates the number of intersection points between the
perpendicular line and the spot. If the number is odd, this point
will be in the spot; otherwise, this point will be outside the
In order to enhance the operating efficiency of program, firstly
obtains the all pixels involved in the smallest outside rectangle
of one polygon, which will reduce the searching scope.
Secondly do the judgment using the Vertical Line Law.
Before to calculate intersection points, it should filtrate
voluminous segments through judging whether the each
segment of spot contains the given pixel point in X axis,
because only the segments having given pixel point have the
possibility to intersect with perpendicular line. By this way, it
can enhance the operating efficiency of program.
(2) Judge the category attribute of spot
Considering there are various objects or mixed pixels involved
in spots, it cannot obtain a unique classified object utilizing the
Maximum Likelihood Classification. Therefore, the attribute
judgment of spot cannot simply depend on one kind of object
classification result. According to this problem, this paper
proposes the Spot Class Statistic and Judge Function, it means
that calculates the homogeneous category pixels in the same
spot using statistical methods, and divides this spot to the
category having the most pixels.
Fig. I the flow chat of land use changing detection program
4.1 Experiment Results
In the classification experiments, the Maximum Likelihood
Classification is used to carry on the classification based on the
spectral feature, and obtains the classification results. Fig 2 is
the map overlaying the MLC results and the present land use
map. In this classified result, yellow expresses the vegetable
plot, the aqua expresses the residential area, the bottle green
expresses the paddy field, the purple expresses the dry
farmland, and the black expresses water surface. This paper
randomly selects 500 samples in the sample region covered by
RS image, and constructs the corresponding confusion matrix
to assess the accuracy of classification results. Table 1 is the
confusion matrix of MLC results. And the calculated total