Abstract:
In recent times overhead imagery is readily available in the form of satellite or drone imagery.
Such imagery, now a days, has a wide range of applications for example detection and
identification of desired targets such as convoys, trains, and trucks. It can also be used for the
identification of infrastructures such as runways, sheds, and airports. Overhead imagery has been
proved useful in the field of agriculture where identification of crops is required. Extending this
range of applications, our project identifies another very important use of overhead imagery that
is the identification of aircrafts in high resolution satellite imagery. This can be especially useful
in cases where it is required to search and identify aircrafts in a desired area. In this project, we
have trained a convolutional neural network-based machine learning algorithm YOLO V5 on a
data set containing a large number of multiresolution satellite images. A prototype is built first to
serve as a source to identify the shortcomings and required improvements. In the next phase, the
actual project is developed incorporating the appropriate modifications to enhance the results.
After the intensive model training period, once a handsome value of accuracy is achieved the
development of a Laravel-based user-friendly interface takes place. With the help of this interface,
users can input a high-resolution image file where the detection of aircrafts is required. As a result,
they can obtain the same image file with aircrafts identified and marked with the facility to
download this output image. With minor modifications, the process is also able to precisely
provide the pixel values where an aircraft has been found and if the user inputs a geo-referenced
image, the pixel values can further be mapped to actual geographic co-ordinates.
Key Words: High resolution satellite imagery, YOLO V5, Convolutional Neural Network,
Multiresolution satellite images.
