Evaluating Adaptive Classification Methods for Mangrove Mapping with Multi-Resolution Remote Sensing Imagery
View Full Paper
Ask AI
Bookmark
Share
Discussion
Loading...
Member takes
Implication
Randomized trial compares mangrove mapping accuracy using adaptive classification and remote sensing imagery, indicating deep learning's superiority.
Key Points
This study aims to evaluate the effectiveness of various adaptive classification methods for determining mangrove areas using multi-resolution remote sensing imagery.
Utilized multi-source satellite imagery including Landsat-8, Sentinel-2, ZiYuan-3, and GaoFen-1 for mangrove mapping in the Beibu Gulf.
Compared pixel-based Random Forest, object-oriented RF, and a deep learning model (U-Net + ResNet-34) for accuracy and area estimates.
Determined optimal feature combinations based on resolution, integrating spectral, texture, and topographic features.
Deep learning achieved the highest accuracy using Sentinel-2 and terrain features, with an F1-score of 96.45%.
Medium-resolution images showed improved accuracy with terrain features (F1-score increase of 1.91%), while high-resolution images benefitted more from texture features (F1-score increase of 2.99%).
Object-oriented RF with Sentinel-2 had a recall of 94.38%, but area discrepancies of 9.7% were noted, likely due to tidal dynamics.