پديد آورندگان :
علوي، علي دانشگاه تبريز , روستايي، شهرام دانشگاه تبريز - گروه ژئومورفولوژي , يوسفي، مريم دانشگاه بيرجند , كيا، روح الله دانشگاه شهيد بهشتي
كليدواژه :
آشكار سازي تغييرات , شبكه عصبي , الگوريتم پس از انتشار خطا , تبريز
چكيده لاتين :
Introduction
Investigating land use changes in urban areas over time, provides information on how
urban parameters are expanded and also how to manage and plan micro and macro
planning. Detection and recognition of changes in time intervals for urban areas as
the most important place that is changing throughout the world is of paramount
importance. Because in these areas, land with different landuses has located next to
each other and they are more rapidly growing than other area, over time, and are
converted to other landuses. In this regard, satellite images of different times can be
used in precise, efficient and economical monitoring. Change detection is one of the
major applications of remote sensing. Using the repeatability feature of remote
sensing data of different times, makes it possible to identify and investigate the
variable and dynamic phenomena in the environment. The city of Tabriz, as the
largest metropolitan area in the northwestern Iran, has Witnessed a lot of changes in
the amount and type of urban usage over the last three decades. Therefore, by using
satellite images and processing them, this issue was studied. For classification of
satellite images, the neural network classifier method was used as one of the most
significant methods in land use classification BY remote sensing data.
Materails and Methods
The satellite images used in this study are presented in (Table 1).
Table 1: The satellite images used
Date sattelite Frame
20 july 1990 TM 168-34
12 june 2000
ETM+ 168-34
24 july 2010
12 may 2005 IRS 64-43-lC
Number1.2.3
ove the accuracy of geometric corrections. Urban use maps prepared from the city's
comprehensive plan to assess the changes occurring in the study area, The ground
control points collected by GPS from the area were used for comparison and
evaluation with the results of the analysis of the images, In order to perform the
necessary steps in geometric corrections and the classification and presentation of
the changes ENVI 4. 7 and PCI GeomaticalO software. The GPS device for the
acquisition of ground control points and the GIS software, including ArcView 2.3 and
ArcGIS 10, were used to assess the changes and implement the PCC technique.
Discussion
After classification of the images in the neural network, the training error, which
indicates the amount of training and reduction of network error, was also extracted.
The error matrix derived from satellite imagery for all periods. As the percentage of
extraction of landusess showed, the minimum error numbers dedicated for three
land uses such as access roads, barren lands and built up area, and then in the other
two landuses such as vegetation and water bodies for all years. The main reason for
the reduction of error in the main uses, which has a high percentage of images, is the
training of the neural network in the classification. The Overall accuracy shows
accuracy above 90 percent. The highest overall accuracy of the 2010 classification
was 94.86% and the lowest overall accuracy of 2005 was 90.01%. The Kappa
coefficients in the 2000 and 1990 images were 0.9, the 2005 figure was 0.85 and
0.09, respectively. The maps for evaluating images obtained from the
implementation of the neural network algorithm with the PCC technique.
Conclusion
The purpose of this study was to investigate variations of Tabriz city over the 20-year
period. In this research, five main landuses were used to extract the changes.
Classification accuracy checking by calculating general accuracy and Kappa
coefficients for classified images shows the efficiency of the neural network in the
classification of the above images. The results Show that, baren land decreased 61%
in 1990 to 55% in 2000. Built changes in 1990 for the whole region were 18% and in
2000 the land use of this site in the region has increased by 18%. In 2005, the built
up increase was 29.6% and the increase was 1.3%. In 2010, changes in this land use
were about 31%. Access roads are the complications that are difficult to extract in
satellite images with moderate spatial resolution and in most cases, due to the same
reflection value it was mixed with other land uses, such as land, and baren land areas.
In 1990, access roads in the region comprise 6% of the total area. The increase in this
land use in 2000 was about 7%. In 2005, this land use increased to 1.5% to 8.5% of the
region. In 2010, this increase was about 11.7%. In 1990 the amount of vegetation
extracted was 15%. This figure was 12% in 2000. In 2005, there was also a slight
decline, down from around 10%. and finally, in 2010, we see an increase, which is up
to 12.5%.
Neural network method showed better capability for extraction of urban landuses
according to the error matrix in the extraction of land uses such as built up area baren
land and access roads. Therefore, in order to extract changes in urban land uses, it is
necessary to use a combination of techniques and classification algorithms such a -
Support Vector Machine and Object-Oriented Classification.
