Download

Download Full Text (1.4 MB)

Description

Rail maintenance is a core necessity of upkeep, maintenance, deterioration, and safety implications of large rail systems. Doing so in an orderly and uniform way, however, has proven difficult with the thousands of miles that railways span. Use of computer vision on locomotives during normal operation can allow for a more streamlined process that would eliminate the need for costly specialized equipment and disruptions of normal operations. This project aims to efficiently detect vegetation overgrowth, mud-pumping, and standing water on railways. With the use of machine learning algorithms and dataset curation, models can be trained, allowing them to gain a better ability to detect when maintenance is due. Through training of these models, improvements in rail anomaly detection have been made, detectable through a comparison of other datasets and models. This work is the first to explore the ability of modern segmentation algorithms to accurately detect anomalies through the creation of a new Rail Anomalies Dataset.

Publisher Location

Las Vegas (Nev.)

Publication Date

Fall 11-21-2025

Publisher

University of Nevada, Las Vegas

Language

English

Keywords

Machine Learning; Semantic Segmentation; Computer Vision; Dataset Curation; Algorithm Analysis

Disciplines

Electrical and Computer Engineering | Engineering

File Format

PDF

File Size

1400 KB

Permissions

Google Drive\Institutional Repository\OUR_OfficeOfUGResearch\Symposia\2025 Fall Symposium

Comments

Mentor: Brendan Morris

Rights

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

Rail Anomalies Dataset for Semantic Segmentation Analysis


Share

COinS