Research Proposal

Using Deep Learning Techniques On Satellite Images For
Identification And Health Analysis Of Crops

Due to increase in population, there is a need for higher agricultural production and reliable
crop status information throughout the world. So, we need a better management of world’s
agricultural resources, especially in developing countries. To make this possible it is required
to obtain crop status information on types, health, quality, quantity and location of these
resources. Satellite Imagery is very useful for large area with no requirements for permitting,
and avoid any security issues. It also helps in precise weather forecast of different farm location
based on field coordinates. With the developments in remote sensing applications, it has
become easier in monitoring large-scale crop condition that is focusing on vegetation index
analysis. It is typically done by using satellites, which collect data reflected from the Earth's
surface. The information is obtained with the help of electromagnetic radiation, force fields or
acoustic energy using various types of remote sensors.
1. Our primary goal is to make the crop searchable from satellite imagery and identify the
geo location of the crops with crop based deep learning models. We explain the e-
farmerce Platform Implementation of Deep Learning in Satellite Imagery for Coconut
Crop Identification.
2. Deep learning and satellite imagery are changing the way we look at agriculture.
3. Every crop exhibits a unique image in satellite imagery and deep learning techniques
are found to be useful to identify those signatures in satellite imagery.
In the area where our Deep Learning codes for coconut model has been tested, we show
the Crop Health from the Multispectral Satellite Imagery which represents relative
chlorophyll content, which correlates with vegetation vigour and productivity.
This shows our Implentation can be used to identify a Specific Crop and assess productivity of
crops in the same geolocation to identify site-specific issues that can be addressed by corrective
management decisions.
We have Used NASA World Wind for Data Visualisation and Interpretation.
The satellite imagery is retrieved using Google Static Maps API using Ruby and GeoJSON.
Extract Features
For each image tile, use a neural network to extract features (eg. shapes, colours, textures)
can include both the visible and non-visible spectrum.
Analysis the pixels that make use satellite imagery and establish what’s unique about that
image for each crop.
Intelligent Deep Learning Crop Models

Continuously train the deep learning model, to establish closer and tighter link between crop
and satellite imagery tiles.
STEPS
First, we have downloaded the satellite imagery and extracted the satellite imagery tiles.
• Define region of interest
• Extract the tiles
• Label them –which crop is this.
• Train a convolution neural network on crop.
• Test the deep learned features model on the data set and do predication on satellite
imagery tiles in the specified geographical location.
Generally, the vegetation images would show crop growth right from planting to harvest. As
the season changes the variation in soil compaction, dry areas or water problems, and weed
patches can be identified. The size of the area that is affected can be determined as well. The
benefit of satellite crop monitoring and the vegetation control is beyond the human naked eyes.
Geographic Information Systems (GIS) tools and information obtained from satellites can help
farmers to conduct crop forecasting and manage their agriculture production by utilizing
multispectral imagery collected by Satellites. We apply several deep learning algorithms on
multi-spectral and multi-temporal satellite images and derive crop classification models.
Images obtained from satellites is in various spectral bands defined by its wavelength and
bandwidth. These bands are used in calculating certain indices, such as NDWI (Normalized
Difference Water Index) and NDVI (Normalized Difference Vegetation Index), which are used
in detecting water and vegetation, respectively. However, use of predefined indices has shown
lacking due to periodic change in crop behaviour, variation of crops health and quality during
time. The use of multi-spectral and multi-temporal It also helps in precise weather forecast of
different farm location based on field coordinates. With the developments in remote sensing
applications, it has become easier in monitoring large-scale crop condition that is focusing on
vegetation index analysis. . It is typically done by using satellites, which collect data reflected
from the Earth's surface. The information is obtained with the help of electromagnetic radiation,
force fields or acoustic energy using various types of remote sensor satellite images and
analysing them pixel by pixel has proven beneficial in crop classification. we utilize multi-
temporal properties of satellite images by collecting a sequence of images taken throughout the
year and apply deep learning algorithms on different indices of the same area to classify
different crops. Satellites images not only helps in highlighting problematic areas but also helps
in checking of applied preventive actions that a farmer can choose to improve the crop
condition. Other importance of farm Satellite Monitoring due to satellite images are Wage costs
reduces a smaller number of people is required to monitor large piece of farm, accuracy
increases and increases data collection for future reference and research work. We use several
supervised machine learning algorithms including linear, non-linear and tree-based. This way
the satellite images can be used for crop classification and improving crops health with
satisfying accuracy, minimum image pre-processing with less agricultural domain knowledge.
Farmers in those stressful regions will be advised and helped to manage the crops effectively
to increase productivity.
Related Work
There is a great amount of work done on land usage detection from satellite images, where R.
Khatami gives some general guidelines, and C. Gomez provides challenges and opportunities.
J. Fan et use a single high-resolution image and apply pixel-based analysis to detect vegetation.
N. Laban. achieve F-score of 0.89 using deep learning convolutional neural networks. Analysis
is performed in two stages, firstly with supervised and secondly with unsupervised clustering.
G. Alp take it further by classifying land usage based on CORINE nomenclature.

Research Proposal

Uploaded by

Copyright:

Available Formats

Research Proposal

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Research Proposal

Uploaded by

Copyright:

Available Formats

Using Deep Learning Techniques On Satellite Images For

Identification And Health Analysis Of Crops

Intelligent Deep Learning Crop Models

You might also like