Unmanned aerial vehicles use GPS for localization, but there may be situations where GPS is unavailable due to noise or interference and vehicle sensor imprecisions add up over time causing loss of accurate position. One approach to solve this issue is using particle filtering with aerial image similarity metrics based on triplet neural networks. Previous research explored the use of a neural network with VGG-16 base layers for this purpose. This research paper explores creation of similarity metrics based on newer and more efficient neural network architectures such as ResNet, EfficientNet, MobileNet and EfficientNetV2. The base layers of these networks have been trained and used with triplet neural networks. The results of the new similarity metrics are compared with the VGG-16 benchmark to find more suitable architecture configurations for recognizing similar aerial images and localizing unmanned aerial vehicles.


    Access

    Download


    Export, share and cite



    Title :

    Panašių ortografinių nuotraukų atpažinimas, naudojant neuroninius tinklus ; Learning aerial image similarity using neural networks


    Contributors:

    Publication date :

    2022-06-08


    Type of media :

    Theses


    Type of material :

    Electronic Resource


    Language :

    Lithuanian , English


    Classification :

    DDC:    629