High Resolution Imagery
based Forest Classification – Woody
has been recognized as an efficient tool for forest inventory. It allows us to observe
accurate and up-to-date data from large forest areas. Imaging spectrometers
allow the acquisition of images in hundreds of contiguous and very narrow
bands. They provide a continuous and exact reflectance spectrum of the observed
object. Usually, the object can be identified from its characteristic spectral
reflectance. The reflectance spectrum of green vegetation in the visible region
is controlled by contribution of chlorophyll. The spectrum in the near infrared
region is controlled by water content and the contribution of other organic
plant species usually have characteristic reflectance spectrum. It is obvious
that water content and contribution of chlorophyll are not constants even in
the same kind of plants and it causes variation in reflectance values. Besides,
the reflectance values of two different plants are sometimes very similar. This
leads to the problem that certain vegetation classes mix together. When looking
at remote sensing images with high spatial resolution, it is possible to
distinguish different age forest types. Even if reflectance of different age
trees are very similar. This is mainly based on the texture of the forest.
Taking texture into consideration we get more information about the forest. It
was required to carry out satellite based study using High Resolution satellite
Image to find out the different LU/LC (including detailed forest
classification) classes in this case study.
The LU/LC was to be
created for more than 60 sq miles area of US.
Current case study
required to carry out satellite based study using High Resolution Satellite
Image to find out the different LU/LC classes. Hybrid classification techniques
were used to extract the required information in raster format especially to
distinguish the woody vegetation into coniferous and deciduous classes. Some of
the classes classified were:
The inputs that were
used for this case study were:
·Image type- High Resolution Satellite Images
·Images Resolution- 2.0 Ft
·No. of bands used- 03
The following software was used:
methodology followed in this project is
·Image preparation (Sub-setting image as
per area of interest)
·Image classification and Compilation
·Assessment (Quality Analysis/Error
The images were
given by the client as a high-resolution satellite images.ERDAS Imagine 9.1 software was used for
extracting area of interest as well as for the entire classification.
Classification and Compilation
The commonly used classification
methods are Supervised and Unsupervised classification. Normally, multispectral
data are used to perform the classification. The objective of image
classification is to identify and portray, as a unique gray level (or color),
the features occurring in an image in terms of the object or type of land cover
these features actually represent on the ground.
is more computer-automated. It enables to specify some parameters that the
computer uses to uncover statistical patterns that are inherent in the data.
These patterns are simply clusters of pixels with similar spectral
characteristics. In some cases, it may be more important to identify groups of
pixels with similar spectral characteristics than it is to sort pixels into recognizable
categories. This method is usually used when less is known about the data
before classification. It is then the analyst’s responsibility, after
classification, to attach meaning to the resulting classes. Unsupervised
classification is useful only if the classes can be appropriately interpreted. Thus,
with the help of ERDAS IMAGINE and its assistance in this process, various
classes were classified. However, Luminous ETS has adopted an automated
process of classification followed by manual cleaning. These automated
2.1 Hybrid Classification
Hybrid classification is the use of both supervised and unsupervised
techniques to classify an image. Both methods, when taken singularly, have
their drawbacks. Supervised classification requires knowledge of the area
and/or detailed field data. Unsupervised classification requires minimum
initial input from the analyst, but the output takes a significant amount of
time to assign the computer-generated clusters to a known land cover.Hybrid combines the benefits of both
2.2 Land Cover Mapping
The term land cover relates to the type of features present on the
surface of the earth whereas the term land use relates to the human activity or
economic function associated with the specific piece of land. Depending upon
the level of mapping details, its land use can be described as in urban use,
residential use or single-family residential use.
which shed their leaves in the fall on an annual basis.
·Coniferous: Cone shaped trees
often found at high altitudes.
·Mixed: Trees which may
comprise of the above mentioned two classes.
buildings, driveways, sidewalks, gravel and hard-packed dirt roads
·Water: Swimming pools, open
sites, mines, beaches, etc.
2.3 Generation of Land Cover
Geometrically and radiometrically corrected satellite images were
used for classification. Classification was carried out using the unsupervised
classification in which the number of classes, maximum iterations and
convergence threshold were given first. Further, the classification was carried
on by analyzing the spectral values of the pixels as well as its tone. Finally,
the classes were finalized as mentioned by the client.
At many times it becomes difficult to
classify certain pixels especially the ones which fall in
water and shadow areas. Therefore, to avoid such confusion neighborhood
function was run in order to remove water and shadows.For it kernel filter of 3*3 (function: majority) was used. This was saved
in a particular file for further use.Masking is an
automated process which is performed to overall enhance the image. During the
image classification process its obvious that many pixels can be left without
having a proper class. Therefore, masking was done for such pixels. This
process enables us to extract a particular class in which pixels are left
After which very small
pixels were clumped and eliminated. Clump performs a contiguity analysis on a
single layer RASTER. Each separate raster region, or clump, is recoded to a
separate class. The output is a single layer raster in which the contiguous
areas are numbered sequentially. Then mosaicing of the extracted classes was done
with the earlier saved file on which neighborhood function was run. Further,
these classes were masked again so that more information could be extracted
about the woody and impervious classes respectively. Once again the above
mentioned steps were carried out.Finally, both the images were mosaiced again for which neighborhood
function was run. This time it was done in order to remove line feature of the
class. Finally, it was mosaiced with water body, which was separately
·Miss classification: The main problem that was faced during
classification was of miss-classification by the automatic classification
process. In this many of the pixels were not being correctly classified.
·Shape: Another problem
that was faced was in the shape of the feature which was not being properly
identified. Especially the impervious features did not pick the exact shape.
·Shadow: In automated
classification the shadow of the trees were classified in Water due to the
·Other problems: Many times the
pixels were not being identified either by its tone or by its texture. Thus,
making it difficult to categorize them as per their respective class. Due to
which some of the pixels belonging to one class were found in some other class.
·Some of the classes were masked and extracted separately to pay
attention on that particular class. It was further classified and picked
accurately. When automated process and picking of the pixels completed then it
was manually recoded with the help of AOI creation to fulfill the quality
requirement of the client.
·Whereas, at certain times the pixels of the particular class were
clumped and then recoded in the defined class. Further viewing the quality of
the images it was thought that the image should be masked and then recoded.
·The general classification and processing techniques were not
feasible to extract the features exactly in their proper shape. To overcome
with these problems a semi automated classification technique was used which
showed good results in a very less manual efforts and time.
·Neighborhood function allows performing one of several analyses on
class values in an .img file using a process similar to convolution filtering. They
are specialized filtering functions that are designed for use on thematic
layers. Each pixel was analyzed with the pixels in its neighborhood. The number
and location of the pixels in the neighborhood were determined by the size and
shape of the filter, which was defined. Further, each
filtering function that resulted in the center pixel value was replaced by the
result of the filtering function.
Analysis/ Error Removing)
the classification of the image was done, the final assessment was carried out
which comprised of the manual editing along with the Quality Check/Quality
Analysis, so as to remove the errors.
The following final
outputs were made-
i)Classified LU/LC Map
ii)Thematic Maps in Img/TIFF format
For more information education or Business queries contact me at email@example.com