PhD Thesis Review on
Pattern Analysis of Chlorophyll Fluorescence Signals
by Marius Cosmin CODREA
"Pattern Analysis of Chlorophyll Fluorescence Signals" was elaborated by Marius Cosmin CODREA in the Turku Centre of Computer Science, and presented for public criticism in Auditorium DataCity,
on the 24th of May 2006, with the permission of the Faculty of Mathematics and Natural Science of the University of Turku.
The thesis's main objectives are to plant species identification, and the fruit quality assessment relies on the chlorophyll fluorescence phenomenon that occurs in all chlorophyll containing material.
The author uses quantitative methods in order to optimize the features to be used for classification by designing and testing a particular form in a selection method. The technique is a non-destructive one, based on the reflectance imaging, observation of pre-symptomatic disorders before defect on the peal of fruit become visible for human eye.
An algorithm is proposed for filling region in digital images. The developed researches indicate the potential of the method as a practical tool for automatic fruit sorting, based on the visual inspection tasks like defect detection.
The authors' researches are based on the results of the original experiments published in the following papers:
- Feature Learning With A Genetic Algorithm For Fluorescence Fingerprinting Of Plant Species
- Genetic Feature Learning Algorithm For Fluorescence Fingerprinting Of Plants
- On The Robustness Of Fluorescence Fingerprinting Of Plants
- Classifying Apples By The Means Of Fluorescence Imaging
- Classification Of Apples according To Physiological Status Measured By Fluorescence Imaging
- An Algorithm For contour-based region filling.
The structure of the thesis includes a first part SYNOPSIS and a second part PUBLICATION REPRINTS.
The introduction presents details about plant species identification and fruit quality control. A special chapter is dedicated to the chlorophyll a fluorescence, and the author presents chlorophyll fluorescence induction and fluorescence imaging problems.
Marius Cosmin CODREA reserves a large area to the methodological background, and to the special details on basic concept in pattern recognition and feature selection.
Outline of the thesis includes the main author research results on the plant fluorescence fingerprinting, fruit quality control and plant species identification.
The bibliography contains 117 titles, books, papers and presentation in conferences proceedings.
The author has oriented his researches to new techniques and methods based on the neural networks, genetics algorithms, and fuzzy classification methods.
The experimental results use the representative data set, and the tests validate the entire hypothesis and the models built by author.
There are taken into consideration four quality levels: very good, good, bad, very bad, and are defined modalities to establish the affiliation to one of these classes.
Marius Cosmin CODREA makes a comparative analysis of two classification techniques - neural network and k-NN - using the same lots of products according to fluorescence. Using a performance criteria, accepted to be representative, it is established when each of the two methods can be used. The algorithms used for processing images, in order to obtain a classification, give the possibility to calculate a regularity coefficient using data that describes size, perimeter, circularity, elongation and localization.
The thesis includes the original and valorous results of a young researcher and it is a mark for his future as a scientist.