Thermal drone detection research reserved 600 images for an independent test
A July 2025 study separated 600 thermal images from previously unseen video footage for testing. That separation changes how a reader should interpret the reported detection results.
The UXV Searcher Research Desk produced this analysis with AI assistance. The cited publishers are responsible for their original reporting.
How the researchers divided the evidence
Tijeni Delleji, Feten Slimeni and Ahmed Siala curated 6,000 images for training and validation. Of those, 4,800 went into training and 1,200 into validation. They drew the additional 600 test images from videos excluded from both earlier groups.[S1]
What the quantities mean
UXV Searcher calculates that the independent images comprise about 9.1% of the 6,600 images across all three groups. That fraction describes the design of the evaluation. It says nothing about the frequency of drone encounters at an airport or the probability of detecting one.[S1]
UXV Searcher analysis
The distinction matters when comparing suppliers that advertise a single accuracy number. A result from new video sequences tests a different proposition from a result on familiar footage. Device, operating environment and test-set provenance are essential context for any cross-study comparison.[S1]
Sources
Original reporting is credited to its publisher. The byline covers UXV Searcher's analysis only.
Related in UXV Searcher
Explore the anti-drone market
Compare counter-UAS announcements with the underlying contract, market and supplier records.
Share this analysis
Need the underlying records or a custom brief? Contact UXV Searcher or see access options.