Assessing the Utility of Pixel-Based and Object-Based Classfication Methods for Surveying Wetland Vegetation with an Unmanned Aerial System

Transcript

1. Nicholas Villarreal, Jennifer Jensen, Thom Hardy, Kristina Tolman Texas State University – San Marcos
2. Ecological Importance of wetlands v  Biofiltration • Phosphorus • Nitrogen • Heavy Metals & Other Nutrients v Habitat • Amphibians • Invertebrates • Migratory Birds • Water Fowl
3. Richland Creek WMA v  Palestine, TX v  WMA: 5,762 Hectares v  Wetland: 899 Hectares v  5 Sediment Basins v  20 Inundated cells v  Nutrient-rich Trinity water v  95% effluent during low flow v  6.5 Days for water to cycle v  Released into Richland- Chambers Reservoir
4. Ø  Create vegetation classification with area per vegetation type Ø  Compare results of pixel-based and object-based classification methods to estimate wetland vegetation cover Results: v  Establish plant species distribution • Annual growth and productivity estimate v  Management tool • Existing • Eradication • Restoration Objectives
5. Classification Methods Pixel-based Object-based
6. v Pixel-based • Can be unsuitable for high-resolution (Lechner et al. 2012)) • Classifies each pixel independently • “Salt & Pepper Effect” • Efficient classification Red Green Blue NIR NDVI NIR – Red ÷ NIR + Red Image Analysis
7. v Object-based • Vector compatibility • Feature space optimization • Implementation of user background knowledge • “Salt and Pepper Effect” less common (Dronova et al. 2012) Image Analysis Color/Shape 0.5/0.5 Sm./com. 0.3/0.7 Color/Shape 0.9/.01 Sm./com. 0.3/0.7 (Benz et. al 2004) Color/Shape 0.5/.05 Sm./com. 0.5/0.5 Color/Shape 0.9/.01 Sm./com. 0.5/0.5
8. Data Processing Classification ①  Cattail ②  Bulrush ③  Spikerush ④  Algae, Pondweed, Duckweed ⑤  Smartweed ⑥  Millet, Barnyard Grass ⑦  Submerged, Open Water ⑧  Sedge v Acquisition • Imagery • Ground truth data (GTD) v Mosaic (Agisoft) •  Geo-reference RGB & NIR v Calculate NDVI v 5 Layer Stack v Classification • Pixel-based (ERDAS) • Object-based (eCognition – Trimble) v Accuracy Assessment
9. MCWE UAS AgiSoft’s Photoscan Professional
10. Flight Coverage Flight 1: - 437 Images collected - 376 Images used Flight 2: - 333 Images collected - 283 Images used Flight 3: - 479 Images collected - 411 Images used Flight 4: - 450 Images collected - 398 Images used
11. Ground Truth Data
12. UAV Imagery with Ground Truth DATA UAV 2014 – 16 to 20 cm resolution NAIP 2012 - 1m resolution imagery
13. Pixel-based Results
14. Object-based Results
15. Areal Coverage Results Algae Bulrush Cattail Millet,   Barnyard 60,892 70,504 118,227 18,906 Sedge Smartweed Spikerush Submerged 22,136 54,908 67,149 21,627 Object-­‐  based  Classified  Areal  Coverage  (m²) Algae Bulrush Cattail Millet,   Barnyard 74,932 53,329 122,455 2,042 Sedge Smartweed Spikerush Submerged 57,008 50,425 61,136 8,398 Pixel-­‐  based  Classified  Areal  Coverage  (m²)
16. Accuracy Assessment
17. ACCURACY Results Class  Name Reference   Totals Classified   Totals Number   Correct Producers   Accuracy Users   Accuracy Kappa  (K^) Cattail 53 50 10 19% 20% 0.0778 Bulrush 85 50 21 25% 42% 0.2635 Spikerush 44 50 15 34% 30% 0.2135 Algae 54 50 22 41% 44% 0.3526 Smartweed 55 50 20 36% 40% 0.3043 Millet 32 50 8 25% 16% 0.087 Submerged 50 50 37 74% 74% 0.7029 Sedge 27 50 8 30% 16% 0.0992 Overall   Accuracy Overall 400 400 141 35% 0.26 ERDAS  Classification  -­‐  Accuracy  Totals Class  Name Reference   Totals Classified   Totals Number   Correct Producers   Accuracy Users   Accuracy Kappa  (K^) Cattail 58 50 40 69% 80% 0.7661 Bulrush 77 50 41 53% 82% 0.7771 Spikerush 51 50 31 61% 62% 0.5645 Algae 55 50 45 82% 90% 0.8841 Smartweed 58 50 36 62% 72% 0.6726 Millet 37 50 20 54% 40% 0.3388 Submerged 38 50 36 95% 72% 0.6906 Sedge 26 50 20 77% 40% 0.3583 Overall     Accuracy Overall 400 400 269 67% 0.6257 eCognition  Classification  -­‐  Accuracy  Totals
18. Future work
19. acknowledgements Tarrant Regional Water District Austin Jensen & Dan Robinson Kristy Kollaus, Tom Heard, John Fletcher Mike Frisbie & Matt Symmank
20. References Benz, Ursula C., Peter Hofmann, Gregor Willhauck, Iris Lingenfelder, and Markus Heynen. “Multi-Resolution, Object-Oriented Fuzzy Analysis of Remote Sensing Data for GIS-Ready Information.” ISPRS Journal of Photogrammetry and Remote Sensing 58 (2004): 239–58. Blaschke, T. “Object Based Image Analysis for Remote Sensing.” ISPRS Journal of Photogrammetry and Remote Sensing 65, no. 1 (2010): 2–16. Blaschke, Thomas, Stefan Lang, Eric Lorup, Josef Strobl, and Peter Zeil. “Object-Oriented Image Processing in an Integrated GIS / Remote Sensing Environment and Perspectives for Environmental Applications.” Umweltinformation Für Planung, Politik Und Öffentlichkeit / Environmental Information for Planning, Politics and the Public, no. 1995 (2000): 555–70. Dronova, Iryna, Peng Gong, Nicholas E. Clinton, Lin Wang, Wei Fu, Shuhua Qi, and Ying Liu. “Landscape Analysis of Wetland Plant Functional Types: The Effects of Image Segmentation Scale, Vegetation Classes and Classification Methods.” Remote Sensing of Environment 127 (2012): 357–69. doi:10.1016/j.rse.2012.09.018. Lechner, a. M., a. Fletcher, K. Johansen, and P. Erskine. “Characterising Upland Swamps Using Object-Based Classification Methods and Hyper-Spatial Resolution Imagery Derived From an Unmanned Aerial Vehicle.” ISPRS Annals of Photogrammetry, Remote Sensing and Spatial Information Sciences I– 4, no. September (2012): 101–6. Walter, Volker. “Object-Based Classification of Remote Sensing Data for Change Detection.” ISPRS Journal of Photogrammetry and Remote Sensing 58 (2004): 225–38.
21. Questions?