MUTANT CLASSIFICATION WITH NEURAL NETWORK BASED ON MODIFIED PATTERN SPECTRUM

摘要

In this paper, a neural network classifier of mutant egg (embryo) using modified morphological pattern spectrum is proposed. The classification of the mutant egg of Medaka is an important process for gene analysis of vertebrate in genetic biology research. A number of mutants were generated by randomly inducing mutations into the Medaka genome. The Medaka is screened for defects in body pattern. First, the color image of microscope is transformed to monotone binary image by reducing background of the individuals using morphology filters and image processing. Next, two modified morphological pattern spectrums are calculated. The pattern spectrum can be a feature of the object that is scale, position and rotation invariant parameters. Further, using the pattern spectrum of features as input signal pattern, the mutant is classified with a three-layer feed forward neural network. The classification ability of two pattern spectrum is compared and verified the effectiveness of the proposed classification system.