Automatic detection of Hawai'i's shoreline stranded mega-debris using deep learning-based object detection

Date
2021
Authors
Winans, Willard Ross
Contributor
Advisor
Chen, Qi
Department
Geography
Instructor
Depositor
Speaker
Researcher
Consultant
Interviewer
Annotator
Journal Title
Journal ISSN
Volume Title
Publisher
Volume
Number/Issue
Starting Page
Ending Page
Alternative Title
Abstract
Marine debris is a global crisis that negatively affects human health and safety, wildlife, and coastal economies. Geospatial technology has been deployed to map, measure, and model the sources, pathways, and eventual sinks of marine debris, however data are still scarce due to the high cost of conducting fieldwork or manually labeling debris in remote sensing images. Recent advances in artificial intelligence and deep learning have brought rapid automation of tedious tasks to new domains, such as marine debris identification in aerial imagery. This study evaluates deep learning-based object detection to automatically detect, localize, and classify stranded marine debris along 1,200 km of the diverse Hawaiian coastline. Two leading object detection models were tested: Faster RCNN with Inception-ResNet-v2 (FR-IR) and SSD with MobileNetV2 (SS-MN). These two models offer a tradeoff between detection speed and identification precision, with one model performing 5x’s as quickly. The results show that the model with higher computational cost, FR-IR, can detect and classify mega-debris (marine debris larger than 10cm) with a mean average precision (mAP) of 57.4%. The speedier model, SS-MN, detects and classifies mega-debris with an mAP of 36.1%. When viewed in detail the results of our object detection and deep learning classifiers show clear best practices for the advancement of deep learning for real-world management of shoreline stranded marine debris at the large, regional scale.
Description
Keywords
Remote sensing, Computer science, Geography, aerial imagery, coastal management, convolutional neural network, deep learning, marine debris, remote sensing
Citation
Extent
116 pages
Format
Geographic Location
Time Period
Related To
Table of Contents
Rights
All UHM dissertations and theses are protected by copyright. They may be viewed from this source for any purpose, but reproduction or distribution in any format is prohibited without written permission from the copyright owner.
Rights Holder
Local Contexts
Collections
Email libraryada-l@lists.hawaii.edu if you need this content in ADA-compliant format.