Observer

1
Observers Associate

• A consistent, unbiased system using machine
  vision and fish morphometrics to identify species

From Scientific Fishery Systems, Inc. P.O. Box
242065 Anchorage, AK 99524 907.563.3474 Dr. Eric
O. Rogers
2
Observers Associate Team

• Principle Investigator - Pat Simpson - SciFish
• Lead Scientist Eric O. Rogers, PhD (Physics) -
  SciFish
• Luke Jadamec, Fisheries Observer Trainer
• Joe Imlach PE, PhD (ME) Imlach Consulting
• Chris Bublitz, UAF Fisheries Industrial
  Technology Center

3
Issues identified by SciFish

• Increasing pressure on resource
• Increasing complexity of new legislation
• Possible environmental changes affecting fishery
  in unknown ways
• Appropriately harvesting and managing the fishery
  are increasingly difficult tasks
• gt Need the best data possible lt

4
Current Sources of Data

• AFSC Survey Trawls
• Practical limits to time and scope
• Observers Reports
• Most effective means of monitoring CPUE
• Statistically small sample
• Potentially biased by factors outside observers
  and vessel operators control
• Of questionable value in legal action due to
  statistical nature of data

5
SciFishs Proposal

• Using funding form the NSF build and test an
  automated onboard fish cataloging system using
  COTS Hardware and Software that will
• Assist commercial fishery observers with their
  monitoring and assessment tasks at sea
• Provide detailed unbiased species counts to
  manage the Community Development Quota (CDQ)
  program in Western Alaska
• Provide new detailed information on the
  ecological health of each species to assist in
  fisheries management
• Provide detailed information on fish
  morphometrics that will be of value to
  researchers in several academic areas, such as
  fish population studies and fish evolution

6
Key Concepts

• COTS hardware and software
• Candle the fish to separate from background
• Machine Vision and Morphometrics
• Neural Net
• Sample all the fish
• System scales - can add CPUs for faster
  processing and add metrics and/or color for
  greater accuracy

7
Observers Associate Benefits

• More and better data means fewer surprises for
  managers and skippers
• A healthier fishery through management based upon
  more complete knowledge
• Sample entire catch, no extrapolation
• Fair and impartial catch statistics - a level
  playing field
• Easy to identify and reward clean Vs dirty
  boats
• Brings in non-traditional funds for fisheries
  research (NSF )
• Fringe Benefit gt Provides length, width, etc.
  for each fish in addition to species

8
Observers Associate Mechanical Design
9
Observers Associate Logic Flow
Fish Outline
Fish Metrics
Image Capture
Boundary Detection
Measure Fish
Identify Species
Fish Image
Fish Metrics
Fish Species
Fish Image
Data Storage
10
Flatfish Features Used by People
11
Typical Flatfish Features Used by Machine Vision

• Body Width ? Standard Length
• Tail Length ? Standard Length
• Tail Fork Length or Max width to tip for rounded
  tails ?Standard Length
• Body Width ? Standard Length
• (Total Width ? Body Width) / Standard Length
• (Ellipse standard length and body width - body
  perimeter) ? Standard Length
• Fin Perimeter (Total Perimeter Body
  Perimeter) ? Standard Length
• (Ellipse Area Body Area) ? (Standard Length
  Body Depth)
• Fin Area / (Standard Length)2

12
Concept Test

• Scan Pictures from Northeast Pacific Flatfishes
  Book
• Scale to meter stick in picture
• Extract measurements
• Reduce measurements to independent metrics
• Principle component analysis
• Train Neural Net
• Create 100 fish / species by adding various
  percentages of white noise
• Test classifier with white noise fish

13
Normalized Machine Vision Flatfish Metrics
Metrics after reduction to Principle Component
Vectors
14
Neural Net Classification Results
15
Observers Tasks

• Identify Species that Observers Associate does
  not
• Quality Control
• Ensure Appropriate Sampling
• Operate the Observers Associate
• Ensure data integrity and file reports

16
Plan

• Assemble Advisory Panel
• Apply for ASTF Bridge Grant
• Build Proof of Concept Prototype
• Train and Test Prototype
• Apply for NSF Phase II Grant
• Build true prototype
• Test for volume onshore
• Test for suitability at sea
• Initial implementation in the Yellowfin Sole
  fishery

17
Advisors PanelComposition

• Regulators
• Conservationists
• Fisheries Scientists
• CDQ Groups
• Fishermen
• Owners
• Fisheries Consultants

18
Advisory Panel Questions

• Are the issues identified by SciFish of Concern
  to the industry?
• Is the technology presented a viable solution?
• Are the other, more appropriate solutions to the
  problems?
• What is the best way to implement this solution?
• Design Changes?
• Are there other applications to add value to the
  system?
• Number of classes for fish Vs accuracy of
  classification, Vs throughput of fish Vs cost