Satellite is particularly reflected in the introduction

Satellite images are characterized by a large number
of features in spectral and spatial domain. However, due to Signal to Noise
Ratio constraints, development of a high spatial resolution and spectral
resolution sensor is problematic. Hence, fusion techniques were developed for
integrating the spatial characteristics from a high spatial resolution
multispectral or panchromatic sensor and the spectral information from the
hyperspectral sensor of low spatial depth. Here, the fusion was achieved by
utilizing the Gram-Schmidt orthogonalization procedure and classification was
performed on the obtained high spatial and spectral resolution image. We have
also compared the improvements over the classification using the red edge
subsets of the Hyperion and fused data sets where a number of vegetation
classes have been selected. All comparisons were performed using advanced
supervised classifiers such as maximum likelihood classifier, spectral angle
mapper and support vector machines with the polynomial and radial basis
function kernel.

1. INTRODUCTION

The economy of India is primarily agrarian with 56.3%
of its population dependent upon agriculture as a source of income and 17.32%
contribution to the GDP (Ministry Of Statistics And Programme Implementation). Modern
agriculture has seen rapid improvements starting with mechanization during the
industrial revolution to manipulation of the genetic structure of the crop.
Technological advancements in the form of accurate navigation systems and
sophisticated satellite and airborne sensors have permitted large scale
monitoring of agricultural structures. Over the last decade, space based
instruments such as RapidEye, QuickBird, LandSat, CartoSat, Hyperion, CHRIS/PROBA
etc have gained much importance in precision agriculture. This is particularly
reflected in the introduction of the red edge band in the sensors. Red Edge
refers to the sharp increase in reflectance as observed in the vegetation
spectrum beyond the red wavelengths, generally from 680-750 nm (Vogelmann, 1993,
Penuelas et al, 1995, Cho et al, 2008). Horler et al. 1983, Dawson and Curran,
1998 and Schulster et al, 2012 document two reasons for this prominent red edge
in vegetation: (1) strong chlorophyll absorption in the red region of the
electromagnetic spectrum and (2) high internal leaf scattering causing a strong
reflectance of the near infrared region of the electromagnetic spectrum.

1.1
Classical Techniques and Challenges

Early
studies in crop stage discrimination, plant health monitoring or even species
identification  using the red edge region
were established (Gupta et al. 2003, Smith et al. 2004). However, the large
weight of the instruments and costs associated with such techniques led to the
necessity of using broad band, multispectral sensors for the same. Plenty of
studies have performed classification of multispectral images utilizing
spectral indices generated out of manipulating the features of the spectrum (Jackson
and Huete, 1991, Schuster et al, 2012, Ustuner et al 2014). However, these methods
face a significant drawback as a result of the broad band nature of
multispectral imaging being unable to map the finer aspects of the spectrum (Ferrato
2012). To cater to this specific advantage, hyperspectral imaging sensors for
space borne and airborne applications were developed and improved significantly
over the last two decades (Pignatti et al., 2009; Purkis and Kemis, 2011;
Heiden et al., 2012). Through hyperspectral imagery, we are now able to monitor
crop health utilizing the chlorophyll and photosynthesis related absorptions at
437 (chlorophyll a), 460 (chlorophyll b), 642 (chlorophyll a) and 659 nm
(chlorophyll b) (Porra et al, 1989, Penuelas et al, 1995, Wu et al, 2008), 705
nm (chlorophyll absorption Wu et al, 2008), 530, 735 nm (indicating
photosynthesis Lang et al, 1991). Soil moisture studies can be performed to
optimize growth parameters possible in VNIR as well as SWIR due to soil-water
interactions from 350 nm to 2500 nm (Gao, 1996, Ceccato et al, 2001, Champagne
et al, 2003, Stimson et al, 2005). Vegetation parameters such as Leaf Area
Index, biomass are estimated using the red edge specifically from 680 to 750 nm
(Vogelmann, 1993, Penuelas et al, 1995, Cho et al, 2008). Plant physiological
parameters including nitrogen content using 1510, 1680 nm bands (Penuelas et al
1995, Serrano et al, 2002), lignin using 1680 and 1754 nm bands and cellulose
quantification using bands from 2000 to 2200 nm (Serrano et al, 2002, Daughtry
et al 2006) can be achieved. Moisture related plant stress can also be studied
due to the water absorptions at 1400 and 1900 nm (Gao 1996, Ceccato et al
2001).

1.2
Motivation and Contributions

Hyperspectral
imagery permit a precise measurement of the radiant flux (Jensen, 2007) and the
differentiation of a variety of agricultural species and crop health stages.
They however fail spatially due to the trade off between spectral resolution,
spatial resolution and SNR (Yokoya et al, 2017). This inherent trade-off
between spatial resolution, spectral resolution and SNR prompted the necessity
to use image processing techniques to use the spatial information of low
resolution panchromatic imagery and spectral information from hyperspectral
imagery thereby metaphorically being able live in the best of both worlds. This
process is called fusion. The fused data sets along with hyperspectral and
multispectral data sets were compared for their classification accuracy using
standard  classifiers such as support
vector machines, spectral angle mapper and maximum likelihood classifier. The
major contributions of this study are summarized as follows:

a. Performing fusion of Hyperion data set with the red edge
band of the RapidEye data set via the technique of Gram Schmidt spectral
sharpening to yield a high spatial and spectral resolution image.

b. Classification of the multispectral image, hyperspectral
image and fused image using support vector machine (using radial basis function
kernel and polynomial kernel of order 3), spectral angle mapper and maximum
likelihood classifier.

x

Hi!
I'm Mary!

Would you like to get a custom essay? How about receiving a customized one?

Check it out