Potential Application of Unmanned Aerial Vehicle for Foreign Object Detection in an Airport Movement Area

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F60162694%3AG43__%2F26%3A00563822" target="_blank" >RIV/60162694:G43__/26:00563822 - isvavai.cz</a>

Výsledek na webu

<a href="http://en.ktu.lt/" target="_blank" >http://en.ktu.lt/</a>

DOI - Digital Object Identifier

<a href="http://dx.doi.org/10.5755/e01.2351-7034.2024.P377-382" target="_blank" >10.5755/e01.2351-7034.2024.P377-382</a>

Jazyk výsledku

angličtina

Název v původním jazyce

Potential Application of Unmanned Aerial Vehicle for Foreign Object Detection in an Airport Movement Area

Popis výsledku v původním jazyce

Detection of foreign objects in the airport movement area is a critical aspect of ensuring air traffic safety at airports. The presence of even the smallest object in the movement area is a prerequisite for aircraft incidents or accidents. The aim of this paper was to investigate the capacities of using unmanned aerial vehicles, technical, and software means for image recording and processing in the automatization process of periodic visual inspection of the movement area at airports. A test section of a real airfield was used to acquire imagery from several flight heights using a DJI MAVIC II Enterprise unmanned aerial vehicle. Image matching, simple colour image segmentation, and convolutional neural network was used for image classification. It was found that the use of image matching is possible, nevertheless, it brings complications in the augmentation of individual images, and satisfactory results were not achieved. In contrast, simple colour image segmentation is much more manageable, although it has many limitations. Exploring the image classification using the convolutional neural network, the SqueezeNet network was augmented and retrained for foreign object classification at an airport movement area. Despite using limited data for training and validation, the performance displayed a great deal of accuracy. To test the trained network not only the common 15 % data proportion but also an additional larger dataset acquired at a different airport was applied. After evaluation of the results, there was only a minor decrease in accuracy. Considering the high accuracy of test results, transfer learning using the SueezeNet convolutional network can be proposed for practical application.

Název v anglickém jazyce

Potential Application of Unmanned Aerial Vehicle for Foreign Object Detection in an Airport Movement Area

Popis výsledku anglicky

Detection of foreign objects in the airport movement area is a critical aspect of ensuring air traffic safety at airports. The presence of even the smallest object in the movement area is a prerequisite for aircraft incidents or accidents. The aim of this paper was to investigate the capacities of using unmanned aerial vehicles, technical, and software means for image recording and processing in the automatization process of periodic visual inspection of the movement area at airports. A test section of a real airfield was used to acquire imagery from several flight heights using a DJI MAVIC II Enterprise unmanned aerial vehicle. Image matching, simple colour image segmentation, and convolutional neural network was used for image classification. It was found that the use of image matching is possible, nevertheless, it brings complications in the augmentation of individual images, and satisfactory results were not achieved. In contrast, simple colour image segmentation is much more manageable, although it has many limitations. Exploring the image classification using the convolutional neural network, the SqueezeNet network was augmented and retrained for foreign object classification at an airport movement area. Despite using limited data for training and validation, the performance displayed a great deal of accuracy. To test the trained network not only the common 15 % data proportion but also an additional larger dataset acquired at a different airport was applied. After evaluation of the results, there was only a minor decrease in accuracy. Considering the high accuracy of test results, transfer learning using the SueezeNet convolutional network can be proposed for practical application.

Druh

D - Stať ve sborníku

CEP obor

—

OECD FORD obor

20304 - Aerospace engineering

Projekt

—

Návaznosti

I - Institucionalni podpora na dlouhodoby koncepcni rozvoj vyzkumne organizace

Rok uplatnění

2024

Kód důvěrnosti údajů

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Název statě ve sborníku

Transport Means - Proceedings of the International Conference

ISBN

—

ISSN

1822-296X

e-ISSN

—

Počet stran výsledku

6

Strana od-do

377-382

Název nakladatele

Kauno Technologijos Universitetas

Místo vydání

Kaunas

Místo konání akce

Hybrid, Kaunas, Republic of Lithuania

Datum konání akce

2. 10. 2024

Typ akce podle státní příslušnosti

WRD - Celosvětová akce

Kód UT WoS článku

—