Large-area automatic detection of shoreline stranded marine debris using deep learning
Marine debris is a global crisis impacting human health, wildlife, and coastal economies. Remote sensing and geospatial technologies are an efficient way to document marine debris across large areas, but the high costs of fieldwork and manual interpretation of debris imagery are major barriers to br...
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2023-11-01
|
| Series: | International Journal of Applied Earth Observations and Geoinformation |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S1569843223003394 |
| _version_ | 1797633293538033664 |
|---|---|
| author | W. Ross Winans Qi Chen Yi Qiang Erik C. Franklin |
| author_facet | W. Ross Winans Qi Chen Yi Qiang Erik C. Franklin |
| author_sort | W. Ross Winans |
| collection | DOAJ |
| description | Marine debris is a global crisis impacting human health, wildlife, and coastal economies. Remote sensing and geospatial technologies are an efficient way to document marine debris across large areas, but the high costs of fieldwork and manual interpretation of debris imagery are major barriers to broader use of these methods. Recent advances in machine learning (ML) have brought rapid automation to many remote sensing domains, including the detection and classification of marine debris. This study evaluates the ability to detect and classify coastline marine debris objects in a real-world setting across a large geographic extent. Three leading ML object detection models were trained to detect and classify large shoreline stranded marine debris from a set of aerial images collected over 1,900 km of Hawaiian coastline. The three models evaluated 1,587 image chips containing 10,703 individual debris labels from 8 debris classes. The SS-MN was both the fastest model and provided the best percentage of accurate predictions, achieving an average precision of 72 %. Nonetheless, this performance was achieved at the expense of missing numerous debris objects, with an average recall of 40 %. The other models provided distinct advantages for certain object classes and use cases. The results show that ML techniques show strong potential to automatically detect and classify certain types of marine debris from aerial surveys. However, there are existing methodological and technical challenges left to overcome before ML methods can outperform human observers in manual interpretation of marine debris from imagery. |
| first_indexed | 2024-03-11T11:52:03Z |
| format | Article |
| id | doaj.art-93813533ab4443f59ee968ab043af1c0 |
| institution | Directory Open Access Journal |
| issn | 1569-8432 |
| language | English |
| last_indexed | 2024-03-11T11:52:03Z |
| publishDate | 2023-11-01 |
| publisher | Elsevier |
| record_format | Article |
| series | International Journal of Applied Earth Observations and Geoinformation |
| spelling | doaj.art-93813533ab4443f59ee968ab043af1c02023-11-09T04:11:40ZengElsevierInternational Journal of Applied Earth Observations and Geoinformation1569-84322023-11-01124103515Large-area automatic detection of shoreline stranded marine debris using deep learningW. Ross Winans0Qi Chen1Yi Qiang2Erik C. Franklin3Department of Geography and Environment, University of Hawai‘i at Mānoa, 422 Saunders Hall, 2424 Maile Way, Honolulu, HI 96822, USA; Corresponding author.Department of Geography and Environment, University of Hawai‘i at Mānoa, 422 Saunders Hall, 2424 Maile Way, Honolulu, HI 96822, USASchool of Geosciences, University of South Florida, Tampa, FL 33620, USADepartment of Geography and Environment, University of Hawai‘i at Mānoa, 422 Saunders Hall, 2424 Maile Way, Honolulu, HI 96822, USA; Hawai‘i Institute of Marine Biology, School of Ocean and Earth Science and Technology, University of Hawai‘i at Mānoa, Kāne‘ohe, HI 96744, USAMarine debris is a global crisis impacting human health, wildlife, and coastal economies. Remote sensing and geospatial technologies are an efficient way to document marine debris across large areas, but the high costs of fieldwork and manual interpretation of debris imagery are major barriers to broader use of these methods. Recent advances in machine learning (ML) have brought rapid automation to many remote sensing domains, including the detection and classification of marine debris. This study evaluates the ability to detect and classify coastline marine debris objects in a real-world setting across a large geographic extent. Three leading ML object detection models were trained to detect and classify large shoreline stranded marine debris from a set of aerial images collected over 1,900 km of Hawaiian coastline. The three models evaluated 1,587 image chips containing 10,703 individual debris labels from 8 debris classes. The SS-MN was both the fastest model and provided the best percentage of accurate predictions, achieving an average precision of 72 %. Nonetheless, this performance was achieved at the expense of missing numerous debris objects, with an average recall of 40 %. The other models provided distinct advantages for certain object classes and use cases. The results show that ML techniques show strong potential to automatically detect and classify certain types of marine debris from aerial surveys. However, there are existing methodological and technical challenges left to overcome before ML methods can outperform human observers in manual interpretation of marine debris from imagery.http://www.sciencedirect.com/science/article/pii/S1569843223003394Marine debrisDeep learningMachine learningObject detectionConvolutional neural networkCoastal management |
| spellingShingle | W. Ross Winans Qi Chen Yi Qiang Erik C. Franklin Large-area automatic detection of shoreline stranded marine debris using deep learning International Journal of Applied Earth Observations and Geoinformation Marine debris Deep learning Machine learning Object detection Convolutional neural network Coastal management |
| title | Large-area automatic detection of shoreline stranded marine debris using deep learning |
| title_full | Large-area automatic detection of shoreline stranded marine debris using deep learning |
| title_fullStr | Large-area automatic detection of shoreline stranded marine debris using deep learning |
| title_full_unstemmed | Large-area automatic detection of shoreline stranded marine debris using deep learning |
| title_short | Large-area automatic detection of shoreline stranded marine debris using deep learning |
| title_sort | large area automatic detection of shoreline stranded marine debris using deep learning |
| topic | Marine debris Deep learning Machine learning Object detection Convolutional neural network Coastal management |
| url | http://www.sciencedirect.com/science/article/pii/S1569843223003394 |
| work_keys_str_mv | AT wrosswinans largeareaautomaticdetectionofshorelinestrandedmarinedebrisusingdeeplearning AT qichen largeareaautomaticdetectionofshorelinestrandedmarinedebrisusingdeeplearning AT yiqiang largeareaautomaticdetectionofshorelinestrandedmarinedebrisusingdeeplearning AT erikcfranklin largeareaautomaticdetectionofshorelinestrandedmarinedebrisusingdeeplearning |