AUTOMATIC CLASSIFICATION OF NANOPLANKTON USING A NEURAL NETWORK ON COLOR FLOURESCENCE MICROSCOPE IMAGE

WHILE SIGNIFICANT PROGRESS HAS BEEN MADE RECENTLY IN COLOR IMAGE-ANALYZED EPIFLUORESCENCE MICROSCOPY FOR THE PURPOSE OFCOUNTING AND MEASURING MARINE PICO- AND NANOPLANKTON CELLS, CONSIDERABLE INTERVENTION BY A SKILLED MICROSCOPIST IS STILLREQUIRED FOR CLASSIFICATION OF CELL TYPES. THESE CLASSIFICATIONS ARE CURRENTLY PERFORMED VISUALLY USING COLORINFORMATION DERIVED EITHER FROM ADDED FLUORSCENT STAINS (FLUOROCHROMES) OR AUTOFLUORESCENCE OF PHOTOPIGMENTS. AN ACCURATE, RAPID CLASSIFICATION METHOD COULD SIGNIFICANTLY IMPROVE THE AUTOMATION OF COLOR IMAGE-ANALYZED FLUORESCENCE MICROSCOPY FOR ESTIMATING PLANKTON CELL SIZE AND BIOMASS ANDCOULD BENEFIT ENVIRONMENTAL MANAGERS AND POLICY MAKERS CONCERNED WITH THE QUALITY OF COASTAL AND ESTUARINE WATERS. OUR OBJECTIVE IS TO DEVELOP AN AUTOMATIC CLASSIFICATION TECHNIQUE WHICH MIMICS HUMAN VISUAL CLASSIFICATION. THE SPECIFIC TASK ADDRESSED HERE IS THE AUTOMATIC CLASSIFICATIONOF NANOPLANKTON-SIZED PARTICLES AS DETRITUS, HETEROTROPHIC, CHLOROPHYLL-DOMINANT PHOTOTROPHS, OR PHYCOERYTHRIN-DOMINANT PHOTOTROPHS. WE WILL TRAIN A MULTILAYERED PERCEPTRON NEURALNETWORK ON COLOR IMAGE FEATURES WHICH ARE DERIVED FROM THE HUE-SATURATION-VALUE COLOR SPACE TO PERFORM THIS CLASSIFICATION TASK IN REAL TIME. THE IMAGE DATA WILL BE OBTAINED FROM WHOLE WATER PLANKTON SAMPLES FROM ESTUARINE, COASTAL, AND OCEANIC ENVIRONMENTS WHICH HAVE BEEN PREPARED USING STANDARD TECHNIQUES IN IMAGE ANALYZED FLUORESCENCE MICROSCOPY.