Improving parcel level crop classification by integrating a novel red edge maize-cotton mapping index and machine learning: A case study in the Ebinur Lake Basin
Accurate crop type classification remains challenged by dependence on ground-based samples and the presence of ‘salt-and-pepper’ noise. This study presented a hierarchical parcel-level classification framework for multi-crop mapping, integrating the boundary-field interaction network (BFINet), the R...
Saved in:
| Main Authors: | , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-09-01
|
| Series: | International Journal of Applied Earth Observations and Geoinformation |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S1569843225004121 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Accurate crop type classification remains challenged by dependence on ground-based samples and the presence of ‘salt-and-pepper’ noise. This study presented a hierarchical parcel-level classification framework for multi-crop mapping, integrating the boundary-field interaction network (BFINet), the Red Edge Maize-Cotton Index (RMCI), and a random forest (RF) classifier. BFINet enables precise delineation of agricultural field boundaries, reducing the influence of non-cropland areas and minimizing pixel-level noise. RMCI is a new spectral index designing for maize and cotton classification. The RF classifier is used to separate cropland into dominant crops and minor crops, and subsequently to classify the minor crops into different crops. Applied to 2023 Sentinel-2 imagery in the Ebinur Lake Basin (ELB), the framework produced the region’s first detailed crop type map. BFINet delineated agricultural parcels in ELB with IOU of 82.3 % and OA of 87.8 %. RMCI achieved an overall accuracy (OA) of 98.6 % for maize–cotton separation, outperforming RF classifier (98.4 %). For minor crops, the RF model attained an OA of 92.3 %. Compared to directly using standalone RF approach, The hierarchical framework outperformed the standalone RF classifier in classifying all crop types in the ELB with F1 for cotton (99.04 % vs. 87.28 %), maize (97.44 % vs. 96.22 %), wheat–maize (88.2 % vs. 82.0 %), grape (92.7 % vs. 89.0 %), and zucchini (94.4 % vs.75.6 %). This framework offers a scalable and accurate solution for crop mapping in complex agricultural landscapes of arid regions. |
|---|---|
| ISSN: | 1569-8432 |