22-10-2012, 11:14 AM
Machine Based Visual Fish Identification
Machine Based Visual Fish.ppt (Size: 839 KB / Downloads: 17)
The history
Sometimes difficult to obtain accurate catch/bycatch statistics from fishermen.
To obtain more objective data, independent observers were included to do sampling.
The two estimations often do not match.
Observers are in short supply, expensive, take up room, and give a real warm “Big Brother” feel to things.
The idea
Provide a computerized system for counting and classifying fish harvests.
Not in the way of workers.
Cheaper than human observers.
Easier to reproduce than human observers.
Secure from elements and tampering of data.
Records only fishing data and can do that 24 hours a day.
Prior Art
Canadian system from the mast that gave periodic images from mast stamped with GPS and time data.
Another European system was able to identify fish species under heavily controlled lighting and environment conditions.
New Constraints
Needed to be closer to the action to record incoming fish.
Cannot cause excessive interference with harvesting operations
Little lighting control(outside)
Little environment control(maximum control is in the chute).
Segmentation
Trying to separate the image of the fish from the background, line, etc.
The main part of the project that I will be working on.
One of three segmentation approaches being tested.
Current Group Status
A prototype form of each component has already been constructed.
Works at a little more than 80% accuracy for limited in-the-chute test data.
New worries
Variable lighting- Needs to work reliably in variable conditions- cloudy, sunny, even at night.
Over the side data(Loss of all environment control).
Both worries have to do with segmentation.
Over the side data
Improves counts of fish actually caught in the lines instead of those hauled and kept on the boat.
Lose any control of picture elements- no choice of background along with little light control. A whole extra level of difficulty.
Need a breakthrough in segmentation to make this possible(and possibly improve current software).
LEGION
An oscillator based neural network that has shown some amazing results in difficult segmentation problems.
Consists of a smoothing algorithm followed by the oscillator network algorithm.
Must evaluate its feasibility for this particular project, and fully implement/integrate the final segmentation algorithm to meet the product requirements.