Page 702 - NGTU_paper_withoutVideo

Modern Geomatics Technologies and Applications

operation of a single-pixel as a homogenous region, repeatedly connecting pairs of these homogeneous regions together to
form larger segments. In examining the condition of local homogeneity, similarities between adjacent image regions are used
to connect them.
After performing segmentation on each of the images, knowledge-based classification should be performed for
classifying the image segments into the predefined object classes. In this regard, a suitable and reliable knowledge base should
be gathered by the expert knowledge and the types of spectral interactions between the pixels of each image segments.
According to the research objective and the characteristics of the study area, vegetation, soil and built up areas are
consider as pre-defined object classes in object based classification. For performing object classification, some spectral
features are measured based on the ratios between various spectral bands. Table 1 shows the mathematical basis of the utilized
spectral features in this research.

Table 1 mathematical basis of the spectral features
Spectral Features Mathematical Formula Description
(NIR − RED)
OSAVI (NIR + RED) + 0.16 Optimized Soil-Adjusted Vegetation Index
NDVI NIR − R Normalized Difference Vegetation Index
NIR + R
DBI (Blue−TIR) -(NDVI) Difference Builtup Index
(Blue+TIR)
MBSI (Red − Green) ∗ 2 Modified Bare Soil Index
(Red + Green) − 2
DBSI (SWIR−GREEN) Difference Bare Soil Index
(SWIR+GREEN) -(NDVI)
NDSI2 (Red − Green) Normalized Difference Soil Index
(Red + Green)
BSI (SWIR1 + RED) − (Red + Blue) Bare Soil Index
(SWIR1 + RED) + (Red + Blue)
In the proposed object based method in this research, using the defined spectral features, appropriate rules are formed
for the classification of image segments. The general hierarchical structure of the applying classification rules on image
segments is presented in Figure 2.

Fig. 2. Proposed hierarchical classification rules

3.3. Change Detection
In the final step of the proposed method, comparing the produced classification maps of 2002 and 2019, change map is
obtained. The results of land use/cover change detection are presented in two ways: 1) Determining the changed and
unchanged object classes, 2) Determining the classes that changed and the type of their changes.
3