Extraction of gully erosion using multi-level random forest model based on object-based image analysis
Gully erosion cause soil organic matter loss, which poses a grave threat to food security and regional ecological sustainability. Remote sensing monitoring and information extraction of gully erosion are of great significance to protect cultivated land resources and agricultural production. To impro...
Saved in:
| Main Authors: | , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-03-01
|
| Series: | International Journal of Applied Earth Observations and Geoinformation |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S1569843225000810 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Gully erosion cause soil organic matter loss, which poses a grave threat to food security and regional ecological sustainability. Remote sensing monitoring and information extraction of gully erosion are of great significance to protect cultivated land resources and agricultural production. To improve the extraction accuracy of gully erosion, multi-level random forest (RF) extraction model based on object-based image analysis (OBIA) is proposed to extract gully erosion information. The Gaofen-2 (GF-2) image was selected as the main data source, supplemented by topographic data, to segment the features in Dehui City based on multi-scale segmentation method. Fusing spectral, textural and geometric feature information, the RF Gini index (GI) was used for feature optimization. Gully erosion extraction based on feature classes was performed using multi-level RF model based on OBIA in the southwestern part of Dehui City, with an overall accuracy (OA) of 96.71% and a Kappa coefficient (Kappa) of 0.865. Compared with the single-level extraction results, the OA and Kappa were improved by 8.4% and 0.102, which indicated that this model has better performance and has certain application value for the research of gully erosion information extraction and dynamic monitoring. |
|---|---|
| ISSN: | 1569-8432 |