Imaging and Classification System for Harmful Algal Bloom Detection

1
Imaging and Classification System for Harmful
Algal Bloom Detection

• Lisa Campbell
• Texas AM University
• Robert J. Olson, Heidi M. Sosik
• Woods Hole Oceanographic Institution

The Cooperative Institute for Coastal and
Estuarine Environmental Technology (CICEET)
2
OBJECTIVES
Test deployment of Imaging FlowCytoBot (IFCB) at
U. Texas Marine Sciences Institute pier at
entrance to Mission-Aransas NERR Optimize
automated classification with images of natural
cells and co-occurring species Establish a
continuous monitoring program and disseminate
results Long-range goal Evaluate modifications
for smaller, simpler, non-cabled instrument.
3
Imaging FlowCytobot
Nano- and microplankton Long term, high
resolution
4
(No Transcript)
5
Long Term - Antifouling

• Filtered sheath fluid, sample in core
• Recirculation with continuous biocide
• Bleach-treatment of sample tubing
• Nylon mesh excludes large particles
• Copper prevents overgrowth

Bleach
6
Automated image analysis and classification
22 categories (16 phytoplankton genera) 88
overall accuracy
Image processing Feature extraction Feature
selection Trained support vector
machine Statistical error correction
7
Example data from Woods Hole Harbor
-Fluorescence / scattering
signature -Associated images
Chl fluoresence
Light scattering
8
IFCB cell counts are comparable to those from
manual microscopy
9
IFCB images of cultured Karenia brevis
Automated classification of Karenia (albeit
against Woods Hole plankton) was encouraging
(gt95 accuracy).
10
K. brevis
11
K. papilionacea
12
Galveston, May 2007
13
Deployment on UT-MSI pier
14
(No Transcript)
15
(No Transcript)