Award Date

May 2025

Degree Type

Thesis

Degree Name

Master of Science in Engineering (MSE)

Department

Electrical and Computer Engineering

First Committee Member

Mei Yang

Second Committee Member

Brendan Morris

Third Committee Member

Shahram Latifi

Fourth Committee Member

Shengjie Zhai

Fifth Committee Member

Fatma Nasoz

Number of Pages

95

Abstract

The integration of autonomous unmanned aerial vehicles (UAVs) with edge computing technology and deep learning (DL)-based object detection offers a groundbreaking solution for real-time wildfire detection, enabling rapid data processing directly on devices and minimizing response delays in critical scenarios. However, although showing early promise, performance is often constrained by limited training data and edge computing devices that lack graphics processing unit (GPU) acceleration. This thesis seeks to address these limitations in two stages.First, this work explores the transformative potential of Transfer Learning (TL) to enhance wildfire object detection model accuracy while also investigating TL’s impact, for DL-based object detection models, on edge computing performance metrics including inference speed, power consumption, and energy efficiency. Towards this end, we introduce the Aerial Fire and Smoke Essential (AFSE) dataset as a target dataset while utilizing the Flame and Smoke Detection Dataset (FASDD) and the general Microsoft Common Objects in Context (COCO) dataset as source datasets. By leveraging the AFSE, FASDD, COCO, and D-FIRE datasets, we also developed and tested a two-stage cascaded TL approach. The application of TL in a single stage significantly enhanced the detection accuracy of the You Only Look Once version 5 nano (YOLOv5n) model, achieving up to 79.2% mean Average Precision (mAP@0.5), while also reducing training time and increasing model generalizability across the AFSE dataset. Notably, cascaded TL showed no further improvement and TL alone did not enhance edge computing performance metrics. Secondly, this research develops a novel one-stage object detection algorithm based on the YOLOv5n architecture, optimized specifically for central processing unit (CPU)-based edge computing devices. YOLOv5n was selected for modification after demonstrating its superiority in edge computing device applications resulting from its speed and accuracy. Without hardware acceleration, an unmodified YOLOv5n model is shown to be able to inference images at nearly two-times the speed of YOLO11n, the latest in the YOLO family of object detectors. Architecture modifications include the use of MobileNetV3-Small as a backbone, Ghost Convolution modules, half the number of output channels in the neck, and the use of 3x3 kernels in the first convolution of all Bottleneck modules. After training, PyTorch weights are exported to two deployment optimized frameworks, Open Neural Network Exchange (ONNX) and Open Visual Inference and Neural Network Optimization (OpenVINO), to accelerate CPU-based inference. Compared to the original YOLOv5n, the modified model converted to OpenVINO demonstrates a 423% increase in inference speed - up to 31.9 frames per second (FPS) - along with an 11.4% reduction in power consumption on a CPU-based edge computing device. The experimental results confirm TL's role in augmenting the accuracy of early-wildfire object detectors while also illustrating that the optimized architecture developed can significantly improve detection speed, power consumption, and overall energy efficiency for CPU-based edge computing devices.

Keywords

computer vision; deep learning; edge computing; ONNX; Open VINO; You Only Look Once

Disciplines

Computer Sciences | Electrical and Computer Engineering

Degree Grantor

University of Nevada, Las Vegas

Language

English

Rights

IN COPYRIGHT. For more information about this rights statement, please visit http://rightsstatements.org/vocab/InC/1.0/

Download


Share

COinS