Quaternary sediments range from Upper Pleistocene to Present
day. But only the Holocene sediments are exposed on the surface.
This Quaternary basin was formed due to the rejuvenation of four
groups of lineaments. The basin floor is uneven and the existing
northward drainage pattern of the area has evolved during the
Neogene-Quaternary period. From the tectonic evolution, it is
evident that intra-continental Quaternary basins were formed due
to neo-tectonic movement and there is no possibility of obtaining
Neogene sediments in any of these areas. The presence of
Neogene sediments in the grabens within the basin, including the
shelf area of the Bengal basin and in the delta of Godavari and
Krishna suggests that these areas bordering the Peninsular Upland
of India would be the probable areas for demarcating the
Neogene-Quaternary boundary (Rabindra K. Roy and Bidyut K.
Ghosh, 1981).
Methodology
Multi-date IRS Wide Field Sensor (WiFS) data over Varanasi
(Fig.1), India is employed in this study. The study area is marked
by 82°11’ - 83°35" E latitude and 24°40’ - 25°38’ N longitude.
This area is considered for hierarchical landuse/landcover
classification using neural networks trained by the back
propagation algorithm. The focus of the present work is the
design of a hierarchy for classification of images. Initially, a
single stage network is configured The neural network is then
configured in a hierarchical topology and a series of small
networks will be employed in successive stages to obtain the final
class. In the initial classification, the study area has been broadly
classified into water, agricultural land, forest and settlements. In
the second stage the focus is on separating the agricultural crop
types by limiting the pixels of interest to those that were marked
agricultural land earlier. These pixels are further classified into
subclasses such as early wheat, late wheat, rice, etc. This
approach is helpful if the area of interest is limited to a portion of
the overall image. The major crop pattern in the study area, as
stated above, includes different crops of wheat, masoor (pulses),
and rice with other classes being forest, urban and water bodies.
Results and Conclusion
The present study is being carried out to investigate the use of the
multi-layered feed forward neural network to classify multi-
temporal images in a hierarchical order. Principal Component
Analysis was done with the ten dates of Varanasi multi-temporal
data, in which six components were chosen for further
classification. All the six components were classified using
conventional MLC Technique (Fig 2) & single stage neural
networks. Initial classification using single stage network is
highly encouraging, even with sub-classification of broader
classes, like agricultural crops which has been divided into, early
wheat, late wheat, rice, masoor, in the present stage itself, as
shown in the output-classified image (Fig 3) & also with the
confusion matrix generated for the classification (Table 1). A
comparative study also was carried with the classified output
using conventional MLC and its matrix (Table 2). The results
reveal the usefulness of the neural network approach in obtaining
accurate image classification. Further, refinement is being carried
out in this regard to generate a hierarchical chain of networks to
have a better classification with higher accuracy as well as a back
propagation algorithm with time varying inputs.
IAPRS & SIS, Vol.34, Part 7, "Resource and Environmental Monitoring", Hyderabad, India,2002
REFERENCES
Ban, Y., and Howarth, P.J., 1999, Multi-temporal ERS-1 SAR
data for crop classification: A sequential mastering approach,
Canadian Journal of Remote Sensing Vol.25 No.5 pp. 438 — 447
Benie, G.B and Thomson, K.P.B., 1992, Hierarchical image
segmentation using local and adaptive similarity rules,
International Journal of Remote sensing, Vol.13 No.8, pp.1559 —
1570
Byeungwoo Jeon and David A. Landgrebe., 1999, Decision
Fusion approach for Multi-temporal classification, IEEE
Transactions on Geoscience and Remote Sensing, Vol.37 No.3
pp. 1227 — 1243 Dec pp. 188 — 194
Derek. R. Peddle, Criles M. Foddy, Airon G. Zhang, Steven E.
Franklin, Elseworth F. Ledrew., 1994, Multi-source image
classification II: An empirical comparison of Evidential
Reasoning and Neural Network approach, Canadian Journal of
Remote Sensing Vol.20 No.4, pp. 396 — 404
Gerylo, G., Hall,R.J., Franklin,S.E., Roberts,A., Milton,E.J.,
1998, Hierarchical Image classification and extraction of forest
species composition and crown closure from Airborne multi-
spectral images, Canadian Journal of Remote Sensing, Vol.24
No.3 pp. 219-232
Gonzalez Alonso, F., and Lopez Soria, S., 1991, Using
Contextual information to improve land use classification of
satellite images in Central Spain, International Journal of Remote
sensing, Vol.12 No.11, pp. 2221 - 2235
Guindon, B.,1988, Multi-temporal scene analysis — A tool to aid
in the identification of cartographically significant edge features
on satellite images, Canadian Journal of Remote Sensing Vol.14
No.1 May pp. 38 — 45
Kohei Arai, 1991, Multi-temporal texture analysis in TM
classification, Canadian Journal of Remote Sensing Vol.17 No.3
July pp. 263 — 270
Liu, Z.K., and Xiao, J.Y.,1991, Classification of remotely sensed
image data using artificial neural network, International Journal of
Remote sensing, Vol.12, No.11, pp. 2433 - 2438
Lorenzo Bruzzone, diego Fernandez Prieto and Sebastiano B.
Serpico., 1999, A Neural-Statistical approach to Multitemporal
and Multi-source remote sensing image classification, IEEE
Transactions on Geoscience and Remote Sensing, Vol.37 No.3
pp.1350 — 1359
Philip. D. Hermann and Nahid Khazenie., 1992, Classification of
Multispectral Remote sensing data using a Back-Propagation
Neural Network, IEEE Transactions on Geoscience and Remote
Sensing, Vol.30 No.1 pp. 81 — 87
Ping An, Chang-Jo F., Chung, 1994, Neural Network approach
for geological mapping: Technical background and case study,
Canadian Journal of Remote Sensing Vol.20 No.3 pp. 293 - 301
Qiu. Z.C
based up
Remote :
Rabindre
morpho-
Ganga V
Conferer
F
