Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.445154
Title: The CatchMeter : application of computer vision for fish species recognition
Author: White, D. J.
Awarding Body: University of Aberdeen
Current Institution: University of Aberdeen
Date of Award: 2007
Availability of Full Text:
Access from EThOS:
Abstract:
This thesis describes trials of a computer vision machine (The CatchMeter) for identifying and measuring different species of fish. The fish are transported along a conveyor underneath a digital camera.  Image processing algorithms determine the orientation of the fish utilising a moment-invariant method, identify whether the fish is a flatfish or roundfish with 100% accuracy, and measure the length with a standard deviation of 1.2mm and species with up to 99.8% sorting reliability for sixteen species of fish.  The machine can theoretically process up to 30 000 fish per hour using a single conveyor based system. The length measurement algorithms are then further developed so that fish may move along an opaque conveyor belt, through the system and be presented in any position or orientation, against a relatively complex background.  By this method the minimum length of fish that can be measured is 50mm and since images can be stitched together the upper limit is >1.5m.  The length of fish is measured with an average error of ± 3%. Two methods of object recognition by colour are compared and are applied to fish species identification.  The colour histogram method and generates variables for subsequent analysis.  The grid method generates a grid on the object and uses the average RGB values in the grid elements as a set of variables for the object.  It was found that increasing the number of grid elements and the number of colour cubes (bins) increased sorting accuracy.  A classification accuracy of 82.9% for nine species of fish was achieved using colour histograms and 98% using average colours.  Furthermore, simple shape descriptors were added to the analysis and this improved the sorting accuracy to 98.5% for the colour histogram method and 99.8% for the grid with average colours method for seven species of fish. Fish species determination using black and white images and by feature extraction using edge detection methods are described with sorting accuracies of up to 95.3% and 97.7% respectively.  A machine that was constructed based on the methods in this thesis is currently installed on one of the most advanced marine research vessels in the world, the Norwegian G.O. Sars.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.445154  DOI: Not available
Share: