Releases · AnwarDebes/drone-id

A source-cited reference catalog and matching API for drone models. It identifies what a drone is, never whose.

Highlights

56 cited drone models, from sub-250g consumer quadcopters to military ISR and strike platforms, across 9 manufacturer countries. DJI lines first, since they dominate the market.
Match an observation to a known model two ways: semantic search over descriptions and spec text (POST /search/describe), and a structured signature match on radio bands, size, weight, Remote ID, and visual attributes (POST /match/signature).
Every populated spec carries a citation; unknowns stay null. 50 of 56 entries are verified against primary sources, with per-candidate confidence and an ambiguity flag surfaced in results.
FastAPI + PostgreSQL (Alembic) + ChromaDB, runnable offline on SQLite with no external services. 46 tests.

Model identification, not operator attribution. manufacturer_country is the country that makes a model, not who flies it. There is no operator field anywhere.
A catalog, not a sensor. It stores what a sensing layer would match against; it does not capture RF, radar, acoustic, or image signals.

6 of 56 entries rest on open-source estimates (CH-4, Wing Loong II, Lancet-3, Orlan-10, Shahed-136, Freefly Alta X) and are labeled partial, never inflated to look more certain than they are.
Live RF, radar, and acoustic sensing, camera ML classification, Remote ID decoding, and operator attribution are out of scope (Phase 2, needing hardware and/or labeled datasets).
The FAA, EASA, and manufacturer importers are documented scaffolds; the curated YAML path is the source of truth today.
The admin API is gated by a simple shared-secret header, not a full identity system.