Blog
- 2026-11-16
Clustering vs nearest-neighbor retrieval for satellite imagery similarity searchClustering groups imagery into categories. Nearest-neighbor retrieval ranks everything by distance to one query site. Here's where each works best.
- 2026-11-16
Field notes: the week the archive refresh reshuffled every match we'd calledThe archive changed underneath a standing query, and the ranked list moved. Notes on sitting with that for a week.
- 2026-11-09
Scoping an AOI before a broad area similarity sweepHow to size and shape a search area before running an imagery similarity sweep, so the sweep returns sites worth reviewing instead of noise.
- 2026-11-09
Tile-by-tile review vs. embedding search for archive-wide site huntingWhy scrolling a collection area tile by tile breaks down at scale, and how similarity search on a reference site changes the math.
- 2026-11-02
Field notes: the day the ranked list put the wrong site at the topA field note on trusting rank order in similarity search, and why the top hit on a lookalike list still needs a human eye on the tell.
- 2026-11-02
Optical vs SAR for site-matching searches under persistent cloudWhy optical imagery still drives site-matching similarity search in persistent cloud cover, and where SAR fits into that workflow.
- 2026-10-26
How to rule out a look-alike site before you spend a tasking requestA screening pass for candidate matches from similarity search, so you spend tasking on the sites worth a closer look, not the near-misses.
- 2026-10-26
What is a diagnostic feature in site-to-site imagery comparison?A plain look at what makes an imagery feature diagnostic rather than circumstantial, and why it matters once a site comparison spans a wide search area.
- 2026-10-19
Field notes: the afternoon one query tile turned up sites on three continentsA field-notes piece on what happens when one known site gets run as a similarity query across a continent-spanning search area.
- 2026-10-19
Handling resolution mismatch between a query tile and archive imageryA practical look at GSD mismatch in satellite search, why 0.5m vs 2m imagery breaks pixel matching, and what holds up across resolutions.
- 2026-10-14
How to set a similarity score threshold for a site searchA practical way to pick a similarity cutoff for imagery-based site search, and why ranking by match confidence beats a fixed threshold.
- 2026-10-14
What is a hard negative in satellite image similarity search?A hard negative is the tile that looks right but isn't. Here's what causes near-miss results in embedding search, and how to triage them fast.
- 2026-10-13
Embedding search vs keyword tagging for satellite imagery archivesHow image embeddings and metadata tags differ for finding look-alike sites in a satellite archive, and when each approach holds up.
- 2026-10-13
How to pick a good query tile for satellite image similarity searchWhat makes a reference tile work for satellite image similarity search, and the crop and resolution mistakes that quietly wreck your matches.
- 2026-10-07
One reference image or twenty? What similarity search needsWhy embedding-based similarity search needs one clean reference tile, not a training set, and when a second example helps.
- 2026-10-07
Query-by-example vs trained classifier: which one finds your site faster?A look at when a trained classifier earns its keep for satellite imagery search, and when marking one example site gets you there faster.
- 2026-10-07
What counts as a false positive in satellite imagery pattern matchingA field guide to false positives in satellite pattern matching: what causes look-alike hits and how to tell them from a real near-duplicate site.
- 2026-10-07
What is a target deck in imagery intelligence?A plain explanation of target decks, nomination lists, and target folders in imagery intelligence work, and where the slow part sits.