Field-Validated Reference Data
SEVESCAN is grounded on a curated reference dataset SEVIMAG-DB built from digital field photographs whose fire severity was assessed using an adapted Composite Burn Index (CBI).
Technology
Technology and Methods
SEVESCAN integrates field-validated reference data, deep learning models, and similarity-based retrieval to support consistent fire severity classification from digital field photographs.
SEVESCAN leverages field-validated reference imagery, deep learning models, and multimodal embeddings to classify fire severity from digital field photographs. The system combines visual evidence, ecosystem context, and geospatial information to generate consistent predictions.
MongoDB supports scalable storage for embeddings and metadata, enabling efficient similarity-based analysis and supporting long-term validation workflows across ecosystems and fire conditions.
Core Technologies
Deep Learning Models
Advanced visual models trained on field-validated wildfire imagery enable robust classification of fire severity across diverse post-fire conditions, improving consistency and reducing observer-dependent variability.
MongoDB Vector Search
SEVESCAN uses multimodal embeddings and MongoDB Vector Search to relate new observations to previously characterized severity patterns. This enables scalable similarity-based retrieval, efficient metadata storage, and flexible classification workflows across ecosystems and fire conditions.
Operational Geospatial Integration
SEVESCAN is designed for operational integration within geospatial validation workflows supported by Vexiza, allowing satellite-derived fire severity products to be validated using field images uploaded through the platform.
