This model is a SegFormer-B5 architecture fine-tuned for semantic land use segmentation and bi-temporal change detection on Sentinel-2 satellite imagery covering 5 districts of Andhra Pradesh, India. Developed by Yantrikaran Innovations Pvt. Ltd. for smart property identification and land analytics for the Andhra Pradesh government. CAPABILITIES: 1. Land Use Segmentation — classifies each pixel into 5 classes from a single satellite tile. 2. Change Detection — detects land use changes between two temporal satellite images (new construction, encroachment, deforestation). CLASSES: - Background (IoU: 0.880) - Buildings (IoU: 0.430) - Road (IoU: 0.431) - Water Body (IoU: 0.635) - Open Plot (IoU: 0.281) TRAINING DETAILS: - Architecture: SegFormer-B5 - Dataset: 20,000+ chips, 5 districts of Andhra Pradesh - Image Source: Sentinel-2 MSI (Multispectral Instrument) - Chip Size: 512x512 px - Total Epochs: 60 | Best Epoch: 56 - Best mIoU: 0.5474 - Training Loss: 0.8196 | Validation Loss: 0.8482 - Early Stopping Patience: 10 epochs - Also hosted at: https://huggingface.co/yantrikaran-innovations/segformer-b5-andhra-landuse
Apache 2.0
Housing, Urban Planning and Infrastructure
yantrikaran-innovations
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
Open
30/04/26 10:14:41
N.A.
0
© 2026 - Copyright AIKosh. All rights reserved.