Peer-Reviewed Articles
1225 A Comparison of Activation Functions for Multispectral
Landsat TM Image Classification
Coskun Özkan and Filiz Sunar Erbek
Abstract Download
Full Article
Neural networks, recently applied to a number of image classification problems,
are computational systems consisting of neurons or nodes arranged in layers
with interconnecting links. Although there are a wide range of network types
and possible applications in remote sensing, most attention has focused on
the use of MultiLayer Perceptron (MLP) or FeedForward (FF) networks trained
with a backpropagation-learning algorithm for supervised classification.
One of the main characteristic elements of an artificial neural network (ANN)
is the activation function. Nonlinear logistic (sigmoid and tangent hyperbolic)
and linear activation functions have been used effectively with MLP networks
for various purposes. The main objective of this study is to compare sigmoid,
tangent hyperbolic, and linear activation functions through the one-and two-hidden
layered MLP neural network structures trained with the scaled conjugate gradient
learning algorithm, and to evaluate their performance on the multispectral
Landsat TM imagery classification problem.
Please see these links for the color figures: Figure 4. Figure 6.
1235 Recovery of Systematic Biases in Laser Altimetry
Data Using Natural Surfaces
Sagi Filin
Abstract Download
Full Article
The accuracy of lidar systems and the removal of systematic errors have received
growing attention in recent years. The level of accuracy and the additional
processing that is needed for making the raw data ready to use are affected
directly by the systematic errors in the laser data. It is evident that calibration
of the lidar system, both laboratory and in-flight, are mandatory to alleviate
these deficiencies. This paper presents an error recovery model that is based
on modeling the system errors and on defining adequate control information.
The association of the observations and control information, and configurations
that enhance the reliability of the recovered parameters, are also studied
here in detail. The application of the model is demonstrated on two of the
main error sources in the system, the mounting and the range bias.
1243 Investigating SeaWinds Terrestrial Backscatter:
Equatorial Savannas of South America
Perry J. Hardin and Mark W. Jackson
Abstract Download
Full Article
Because tropical grasslands play an important role in the storage of global
carbon, monitoring them is critical to evaluating global climate change.
The goal of this research is to model seasonal SeaWinds Ku-band backscatter
in five savanna areas of Colombia, Venezuela, and Brazil as a function of
biophysical changes in the savanna landscape. Multiple regression modeling
demonstrates that savanna Ku-band backscatter is a function of (1) savanna
grass biomass/leaf area, (2) soil moisture, and (3) other soil characteristics.
Fit for the regression models is excellent (R = 0.87 and 0.81, respectively,
for the horizontal and vertical polarization case). The horizontal—vertical
polarization difference is also moderately related to precipitation (R =
0.71). The results from this modeling are consistent with theory predicated
on previous C- and X-band research. The possibility of monitoring savanna
vegetation, soil moisture, and rainfall using Ku-band radar and scatterometry
is discussed.
1255 Grasslands Discriminant Analysis Using Landsat
TM Single and Multitemporal Data
Xulin Guo, Kevin P. Price, and James Stiles
Abstract Download
Full Article
Grassland management practices influence many bio- and geophysical processes.
The ability to discriminate among different land-use practices is critical
to an improved understanding of agro-ecosystem dynamics in the tallgrass
prairies of the Central Great Plains. The overall objective of this study
was to assess the spectral separability of three land-use practices on warm-season
(C4 dominated) and cool-season (C3 dominated) grasslands using data obtained
from multitemporal Landsat Thematic Mapper (TM) imagery. Results showed that
cool- and warm-season grasslands could be discriminated with a high level
of accuracy (91.5 percent). When grasslands were categorized by three common
management practices (Conservation Reserve Program [CRP], grazing and haying),
they could be discriminated with a moderately high level of accuracy (70.4
percent). Grassland management practices within warm- and cool-season grasslands
(six types) were spectrally discriminated with a moderate level of accuracy
(67.6 percent overall). The use of a three-date Landsat TM image dataset
spanning the spring-summer-fall seasons significantly improved classification
accuracy over the use of a single-date TM approach.
1263 Estimating Tree Crown Size from Multiresolution
Remotely Sensed Imagery
Conghe Song and Curtis E. Woodcock
Abstract Download
Full Article
The spatial arrangement of the brightness values of remotely sensed imagery
bears important information for forest canopy structures. This paper presents
the theory and a simple analytical model to estimate tree crown size using
sills of semivariograms from images at two spatial resolutions. The theoretical
basis for the model is that the spatial patterns of multiresolution imagery
are diagnostic of tree size. The sills of variograms from images containing
larger trees decrease more slowly than those for images containing smaller
trees as the image spatial resolution decreases. Tests with generated images,
integrated simulation of stand development and its spatial patterns, and
Ikonos images show that the model can provide realistic estimates of tree
crown size. Errors of the estimated crown size strongly depend on the spatial
resolutions of the images. The best combination of spatial resolutions is
at the ratio of pixel size to the object size around unity.
1271 Evaluating the Performance of Spatially Explicit
Models
Robert Walker
Abstract Download
Full Article
Spatially explicit models are now widely used for conducting ecological research
and for managing natural resources, due in part to the difficulty of undertaking
empirical work at landscape scale. Unfortunately, error assessment and analysis
of the predictive ability of such models is not well-developed, and has relied
on the Kappa statistic and information-based measures. As has been pointed
out, however, such approaches are limited by virtue of their global nature
and weak hypotheses. As it turns out, the literature on map accuracy does
provide a way of assessing model performance, and the goal of this paper
is to adapt this literature to the need for evaluating the predictive ability
of spatially explicit models. To this end, the paper first considers inference
using the Kappa statistic. This is followed by a commentary on information
theory, and a critique of both the Kappa statistic and information-based
approaches given their global structure and underlying null hypotheses. A
probabilistic treatment of alternative measures recently suggested follows,
as does a direct adaptation of map inference to the modeling case. Examples
of the proposed measures are given, using an application of logistic regression
applied to land-cover changes that have recently occurred in the Muskegon
River watershed of the State of Michigan.
1279 Strategies for Integrating Information from
Multiple Spatial Resolutions into Land-Use/ Land-Cover Classification
Routines
DongMei Chen and Douglas Stow
Abstract Download
Full Article
With the development of new remote sensing systems, very high spatial and spectral
resolution images now provide a source for detailed and continuous sampling
of the Earth’s surface from local to regional scales. This paper presents
three strategies for selecting and integrating information from different
spatial resolutions into classification routines. One strategy is to combine
layers of images of varying resolution. A second strategy involves comparing
the a posteriori probabilities of each class at different resolutions. Another
strategy is based on a top-down approach starting with the coarsest resolution.
The multiresolution strategies are tested using simulated multiresolution images for a portion of the rural-urban fringe of the San Diego Metropolitan Area. The classification accuracy obtained from using three multiple strategies was greater when compared with that from a conventional single-resolution approach. Among the three strategies, the top-down approach resulted in the highest classification accuracy with a Kappa value of 0.648, compared to a Kappa of 0.566 for the conventional classifier.
1289 Automated Change Detection for Updates of
Digital Map Databases
Thomas Knudsen and Brian P. Olsen
Abstract Download
Full Article
Updating digital map databases to reflect continuous changes is a major task,
which could be significantly reduced by introducing an automated change-detection
process. We suggest an algorithm inspired by the specific problem of updating
the Danish National Topographic Map Database (TOP10DK). The emphasis is on
buildings, which are very important mapping objects. The algorithm uses vector
and spectral data as input to an unsupervised spectral classification method
which controls a subsequent Mahalanobis classification step. For evaluation,
we present four test cases based on the TOP10DK database in combination with
RGB and color-infrared (CIR) aerial photos. A main problem is roofs covered
with roofing felt, which are hard to discriminate spectrally from roads.
Apart from that, RGB photos are barely sufficient for change detection, while
CIR data give more satisfactory results. Refinements are, however, still
needed to reduce the number of false alarms.
| Top | Home |