Sensing and Accounting of Aquatic Floatables · IIDA
SAAF turns drone imagery of beaches and water surfaces into a verified count of the litter present — what it is, how much of it there is, and what it weighs.
Request a demoSAAF is a measurement instrument, not a campaign. Nobody is asked to donate. The output is a per-class count, a footprint area and an estimated mass for a stretch of coastline, repeatable across seasons, so you can prove whether a cleanup worked.
00The problem
Cleanup budgets are set on impressions: a photo, a volunteer's estimate, a transect walked with a clipboard and multiplied up. Two people produce two answers.
river mouth · floating raft
01Fly the survey
No special sensor. A consumer 4K camera is enough, because altitude and focal length give the ground scale: every pixel is a known number of centimetres.
10–40 m · 3840×2160
02Detect and classify
Six classes today: glass, plastic bottles, cans, takeaway cups, retort pouches, food containers. About a second per 4K frame, so a morning's flying is done by lunch.
6 classes · ~1 s per frame
03Measure and report
Box dimensions pass through the ground sample distance, then sourced reference weights. The result is an estimate, not a weighing, and it is reported as one.
count · footprint · mass
What the model sees
A real frame from our test split. 3840×2160, shot at altitude. Twelve pieces of litter, none of them obvious. Boxes are deliberately tight: a loose box inflates the area, and area is what the mass estimate is built on.
Why it is hard
This rectangle is a 3840×2160 frame scaled to your screen. The small box inside it is a 60-pixel object at the same scale. That is what the model has to find, name and measure.
Who it is for
Direct crews to the stretches that need them most, and prove what changed once they have been.
Learn more
Repeatable, gridded counts across seasons. Same method every time, every frame retained.
Learn more
Quantify a cleanup's effect in kilograms, from imagery flown before and after the work.
Learn more
Send us a survey. We return the annotated frames and the totals before you commit to anything.
Request a demoQuestions
Where there is a limit, it is stated. Mass is an estimate from reference weights, not a weighing, and the site says so everywhere the number appears.
99.2% mAP@50 and 68.2% mAP@50-95 on a held-out test split never used in training. mAP@50 says whether litter is found; mAP@50-95 says how tightly it is boxed, which is what the size and mass estimates depend on.
Each detection's box is converted to real dimensions using the ground sample distance, derived from sensor width, focal length and flight altitude. Those dimensions are matched against published reference weights for each class of item. The result is an estimate, not a weighing, and it is reported as one.
Six classes today: glass, plastic bottles, tin and aluminium cans, takeaway cups, retort pouches and food containers. New classes are added by annotating examples and retraining.
Yes. The model is trained on beach and water-surface imagery, so it handles shorelines, riverbanks and floating litter on open water.
A consumer drone with a 4K camera, and the altitude and camera model used for the flight. Nothing else. Imagery is uploaded and processed after the flight.
One survey. Annotated frames and totals back. Then decide.
Request a demo