FIELD-BASED CROP MAPPING THROUGH SEQUENTIAL MASKING
CLASSIFICATION OF MULTI-TEMPORAL LANDSAT-7 ETM+ IMAGES IN
KARACABEY, TURKEY
M. Turker ', M. Arikan
Middle East Technical University, Graduate School of Natural and Applied Sciences
Geodetic and Geographic Information Technologies, 06531 Ankara, Turkey - mturker@metu.edu.tr
Commission VII, WG VII/2
KEYWORDS: Landsat, Agriculture, Crop, Classification, Multitemporal, Integration, Object
ABSTRACT:
This study presents a field-based crop mapping through sequential masking classification of multi-temporal Landsat-7 ETM+ images
acquired in May, July, and August 2000 in Karacabey, Turkey. First, the classification of each image date was carried out on a
standard per pixel basis. The results of the per pixel classification were integrated with digital agricultural field boundaries and for
each field, a crop type was determined based on the modal crop class calculated within the field. The classification accuracy was
computed by comparing the reference data, field-by-field, to each classified image. The individual crop accuracies were examined on
each classified data to determine those crops whose accuracy exceeds a preset threshold level. Then, the multi-temporal masking
classification of the crops was carried out in sequential steps using the three image dates, excluding after each classification the crop
properly classified. The masking technique was applied to overcome the problems caused by the spectral overlaps between some
classes. The final classified data was analyzed in a field specific manner to assign each field a crop label. An immediate update of the
database was provided by directly entering the results of the analysis into the database. The use of sequential masking procedure for
field-based crop mapping improved the overall accuracies of the classifications of the July and August images alone by more than
10%.
1. INTRODUCTION
The availability of remotely sensed images and the advances
in digital processing and analysis techniques have enabled
research scientists to have information about the type,
condition, area, and the growth of agricultural crops. Image
classification is one of the crucial techniqes in detecting the
crops from remotely sensed data. Most current automatic
classification techniques to obtain land covér maps from
digital imagery operate on a per-pixel basis in isolation from
other pertinent information. Therefore, per-pixel techniques
often yield results with limited reliability. The reliability of
image classification can be improved by including apriori
knowledge about the contextual relationships of the pixels in
the classification process. Agricultural field boundaries
integrated with remotely sensed data divide the image into
homogeneous units each of which can be analyzed seperately.
In each field, the geometry of field boundaries defines the
spatial relationships between the pixels contained within, and
enables those pixels to be processed in coherence. The
decision by the analysis is taken, for each field, based on the
coherent processing of the pixels falling within the field.
Therefore, the standard per-pixel image classification can be
replaced by a classification which operates in a field specific
manner.
Field-based approaches to the classification have been
adopted by several researchers (Catlow er a/. 1984; Mason ef
al. 1988; Janssen et al. 1990; Janssen et al. 1992; Aplin ef al.
1999; Turker and Derenyi 2000; Aplin and Atkinson 2001).
To perform field-based classification, the vector field
boundaries must be integrated with the imagery. The
integration between the two data sets can be achived at three
* us .
Corresponding author
192
stages: (i) before classification, (ii) during classification, and
(iii) after classification. Usually, field-based classification
employs the integration between raster imagery and vector
data after classification (Brisco er al. 1989; Janssen et dl.
1990; Janssen ert al. 1992; Aplin er al. 1999; Turker and
Derenyi 2000; Aplin and Atkinson 2001). The imagery is
classified on a per-pixel basis before integrating the classified
output with digital vector data. A per-field analysis is then
carried out to assign each field a class label based on the
analysis of the classified pixels contained within the field.
The success of a field-based approach that incorporates
vector data after a per-pixel classification depends mainly on
the success of the classification. Several studies have shown
that multi-temporal images improve the classification
accuracy by utilizing different spectral responses of the land
cover classes over a period of time according to phenological
evolution (Maracci and Aifadopoulou 1990; Conesa and
Maselli 1991; Kurosu ef al. 1997; Panigrahy and Sharma
1997; Beltran et al. 2001; Lanjeri et al. 2001; Murakami ef
al. 2001).
The objective of this study was field-based mapping of
summer (August) crops in Karacabey, Turkey through
sequential masking classification of Landsat? ETM+ satellite
data. The sequential masking classification technique was
applied to improve discrimination between the crop classes.
We made an assumption that each field grows one type of
crop. Field-based classification was performed by computing
the percentages of classified pixels within each field and
assigning a class label to the field based on the majority class.
The fields were selected through a database query and the
results were directly inserted into the database.
Internationa
VI EE
2. THE
The study a
town of Kar
The geogra|
N40?13'43"
grown in th
and sugar b
include A
Kucukkaraa;
region, a lai
1988 and 19
shapes that
significant n
of the field
difference a
The Landsat
classification
and August
quality. A €
covering th
proposed sec
+
LE
L LÁ
z>
ÿ
The field bo
sheets. The
scanning the
Kruger (Zon
intersections
polynomial :
registration 2
Next, the fie
on screen di
database. E.
number and
those fields :
during site vi
The multispe
band were r
retains the s
provides the
Furthermore,
resolution in
since the nu
increase. The
Universal Ti
and the Eu
