Assignment of Algorithm (Part-2)

Submitted By: Mansi (21-309)

Topic: Perfectly overlapped merging and sorting
on a two way linear array. (By Yen Chun Lin)
This paper represent perfectly overlapped merging algorithm. It require
almost only input time to perform merge sorting on a linear array of
processors. The optimal merge order that minimizes the merge sorting time
is obtained in constant time. The merging algorithm and a previously
proposed runs-generation algorithm can serve for sorting a very large file;
the time complexity of the resulting algorithm is optimal under the
hardware condition.
Topic: Review of dust storm detection
algorithms for multispectral satellite sensors.

(By Jing Li , Man Sing Wong , Kwon Ho Lee , Janet Nichol , P.W. Chan)

In this research paper satellite remote sensing is used to monitor dust storms in space
and time. To detect dust storm using satellite, it is important to analyze the dust storm
trajectories and sources. This paper reviews the algorithms for dust storm detection
used in multispectral satellite sensors, spanning visible to thermal wavelengths . Four
categories of dust detection algorithms are summarized, namely, dust spectral index algorithms,
temporal anomalous detection algorithms, spatial coherence tested algorithms (physical-based
algorithms) and machine learning-based algorithms. . Future developments for dust storm
detection are focused upon three aspects: detection of dust storms at night time; development
of more efficient machine learning methods for retrieval; and integrating physical and machine
learning methods for satellite images.

1. Satellite and Sensors:

Multispectral sensors are mainly onboard two types of satellites, namely polar orbiting
satellites and geostationary satellites. Polar orbiting satellites provides low temporal
resolution but high spatial and spectral resolution for earth observations. Geostationary
satellites orbit at high altitude and they remain stationary with reference to the Earth
Geostationary satellites allow continuous observations, hence being capable of
monitoring the formation and transportation of dust storms.
2. Dust Detection Algorithm: In medium spatial resolution satellite images, dust
contaminated pixels need to be distinguished from cloud pixels, dark land surface pixels,
bright land surface pixels, and water surface pixels. There are basically two methods for
 dust detection: empirical physical-based, and machine learning-based methods.
Physical-based methods utilize the spectral, spatial and temporal information to
distinguish dust from cloud and clear-day surface.
3. Dust spectral index method: It utilizes the spectral signature of dust storms. Different
features have different spectral signatures that describe their absorption and scattering
ability at different wavelengths. Three major features can be visually distinguished from
medium spatial resolution (e.g. 0.5 km ~ 1 km for MODIS) images: clear-sky surface
(land, ocean), clouds and aerosols (e.g. dust storm, volcanic ash, smoke).
4. Brightness Temprature Difference: . It utilizes the unique spectral signature of dust in
the TIR region. It was later applied for dust detection due to the similarities between
mineral dust aerosol and volcanic aerosol (i.e. the silicate component is contaminated
by both dust and volcanic ash).

5. The anomalous dust detection method:

Disturbances such as cloud, smoke, biomass burning, hurricane and land cover change can be
well-detected by anomalous detection methods. The anomalous detection algorithms use time
series images that capture the scene at the same location and time period. The algorithm
assumes a steady Earth-atmosphere system (i.e. the water vapor effect is not considered).

6. Spatial coherence test method:

The spatial coherence test method is based on this physical difference. Spatial coherence test
also known as spatial variability test or uniformity test is normally represented by standard
deviation. The spatial coherence test method is based on this physical difference. Spatial
coherence test also known as spatial variability test or uniformity test is normally represented
by standard deviation.

7. Machine learning-based algorithms:

Machine learning-based methods are data-driven methods, which can detect dust presence by
self-learning from the given training dataset.