Drone Detection Using Modified Deep Learning Methods

Document Type



Drone detection is an important domain as the accessibility of drones has increased dramatically in the last decade. Many object detection techniques are available, however, drone detection is not investigated widely. Furthermore, these object detection techniques utilize convolution operation extensively which uses multiplication mainly. Most techniques found in literature investigating drone detection utilize small datasets. In this study, two large scale datasets are used to investigate drone detection. Moreover, two methods are proposed to decrease the complexity of drone detection for real time applications. These methods include modified SSD with AdderNet filters instead of normal convolution and a path that adds low frequency features with high frequency features. In addition to that, SWIN transformer was integrated into SSD model to investigate the possibility of utilizing transformers as backbones for drone detection tasks. The aim of this research work is to decrease the complexity of drone object detection while preserving the performance of the model. The experimental results were compared with the state-of-the-art methods including Faster-RCNN, YOLOv3, SSD, and DETR. The proposed model SSD with AdderNet was able to perform similar to one-stage detectors with 90.4G less multiplication or around 90% less multiplications in the convolutions operations.

Publication Date



Thesis submitted to the Deanship of Graduate and Postdoctoral Studies

In partial fulfillment of the requirements for the M.Sc degree in Machine Learning

Advisors: Dr. Min Xu, Dr. Shijian Lu

with 2 year embargo period

This document is currently not available here.