O'Reilly's notes:

Attendees: Jac Fought, Harry Nelson, Francisco Chavez, Denis Klimov, Tom O'Reilly

Action items
1. Chavez - compare Harry's FlowCam result to microscope analysis, sample 14213CO3
2. Nelson - work to get a FlowCam, training to MBARI for CANON (cruises on Sept 16-19, Sept 30, October 6)

Harry has done some analysis of samples provided by Francisco. Used 100x magnification - 25-30 micron image resolution. 100 micron pre-filter. Analyzed with Visual Spread Sheet.
Pseudo-nitzschia predominates.
Some image processing problems, due to "pilot error" by new technician.

Chavez processes a 25 ml sample thusly:
Fix, freeze sample
Operator counts cells under microscope - takes several hours
Operator enters data - takes about 2 hours
FlowCam processed Chavez 25 ml sample as follows:
Acquire images at 25 frames per second - actually images 3.189 ml of the total sample
About 1 hour to process 20 ml sample - automated aspect ratio computation
Can process large size fractions faster, about 1 minute per ml

How could analysis be done autonomously, e.g. on AUV?
Trigger imaging on fluorescence signal (chlorphyll or phycoerythrin), which is useful in oligotrophic water (too much signal in coastal water)
Will get one image frame per trigger
Note that laser excites entire camera fov

FC: We are thinking about a good MBARI FlowCam application
Replace microscope counts
Integrate with LRAUV for in situ analysis
Attach FlowCam to ship water intake, or Pennington's pumped profiler - autonomously id/count Pseudo N
Trigger ESP sample with FlowCam

Realtime image filtering
HN: Difficult to automatically ID all organisms; filtering helps. Alexandrium is difficult; small, spherical. Pseudo N is easier.
HN: Submersible FlowCam is designed for remote operation. Have run on cable for 56 days, ~10 minutes/hour duty cycle during dayl
JF: Submersible FlowCam memory is expandable. Expect to use 150 MByte per hour

Notes added by Klimov:

- 100um Nitex Mesh was used for pre-filtering
- Pseudo-nitzschia chains may stuck on filter and it explains why they were not visible in FlowCam data
- "Segmentation Threshold" (to discriminate between "dark" and "bright" pixels) was apparently set too high, this chops up long cells into separate images
- Fluorescence trigger mode, every frame is unique, 20 FPS
- Stats for 1H run:
    150 Mb file   
    93k frames total, 3.6k frames used.
    1.64 parts/ image.
    19.97 FPS
    Intensity Mean: 179.90
    Intensity Min: 161.77
    Intensity Max: 183.05