Automatic Optical Identification and Classification for the Industrial Processing of Fish

This thesis deals with the optical measurement and classification of fish, which is required as a prerequisite to enable automatic sorting. Two main issues are investigated in this thesis: Part I deals with the geometric measurement based on light sectioning. With this method the size and volume of the fish should be determined, whereby assumptions are made with respect to the symmetry of the fish; Part II deals with profile description with the aim of identifying different species of fish. Elliptical Fourier Descriptors, Cyclic Polynomial Descriptors as well as the radius of curvature of this contours are analyzed. It is shown, that a classification based solely on this data is not possible, since even forms and appearances of fish within one species differ widely. However, the method applied to register two curvatures can be used e.g. to compare the contours of a rolling profile with its corresponding CAD-data. All the proposed and investigated methods have been implemented in MATLAB® and tested with data measured on real fish.