Date of Award

8-1-2024

Thesis Type

masters

Document Type

Thesis (Restricted Access)

Divisions

eng

Department

Department of Electrical Engineering

Institution

Universiti Malaya

Abstract

The advancement of unmanned aerial vehicles (UAVs) has encouraged researchers to update object detection algorithms for better accuracy and computational performance. Previous works that apply deep learning models for object detection applications required high graphics processing unit (GPU) computational power. Generally, object detection models suffer trade-off between accuracy and model size where the relationship is not always linear in deep learning models. Various factors such as architectural design, optimization techniques, and dataset characteristics can significantly influence the accuracy, model size and computational cost in adopting object detection models for low-cost embedded devices. Hence, it is crucial to employ lightweight object detection models for real-time object identification for the solution to be sustainable. This work proposes modifications on the head and backbone architecture of YOLOv7-tiny model. Firstly, efficient long-range aggregation network for vehicle detection (ELAN-VD) is incorporated in backbone layer. Secondly, the (UPSAMPLE-VD) on head architecture is improvised resolution to improve the detection accuracy of small vehicles in the aerial image. This study shows that the proposed method yields mean average precision (mAP) of 77.47 %, which is higher than the conventional YOLOv7-tiny of 48.89 %. In addition, the proposed model shown significant performance when compared to previous works, making it viable for application in low-cost embedded devices.

Note

Dissertation (M.A.) – Faculty of Engineering, Universiti Malaya, 2024.

Share

COinS