Skip to Main content Skip to Navigation

Définition et mise en place d'un outil temps réel d'analyse des caractéristiques physiques des semences sèches

Abstract : The goal of this PhD Thesis is to study and design an automated artificial vision system to discriminate the nature and the variety of seeds from their numerical image. A first study has been done on a mono-camera acquisition system to determine the ability of such a system to certify seed lots. This study has shown that a mono-camera is not able to discriminate damaged or multicolor seeds in one pass. A second study has been done on the study and the design of a new acquisition system, with three cameras, able to acquire three different seed faces images simultanously. A prototype has been designed based on a new electronic and optical arrangement. A collection of seed images has been collected thanks to this system. On this collection, image processing and pattern classifier algorithms have been tested. The seeds are numerically characterised by their morphology, their color and their texture and stored in a dataset. These features are used as input of pattern classifiers. Some pattern classifiers have been tested (Linear Discriminant Analysis, Factorial Discriminant Analysis, PLS Discriminant Analysis, Probabilistic Neural Network) on a reduced dataset on which features were selected by their discriminative power with a forward LDA algorithm. The three cameras used offer the possibility of multiple classifications, either by combining pattern classifiers and use majority or unanimity vote, or by concatenating features to design single classifiers. The comparison test shows that in all conditions the PNN classifier is more accurate than the others. However, the LDA results are close to the PNN. For sake of simplicity, the LDA classifiers seem to be the best compromise for a real time automated system able to certify seed lots.
Document type :
Complete list of metadatas

Cited literature [70 references]  Display  Hide  Download
Contributor : Anne-Marie Plé <>
Submitted on : Tuesday, March 23, 2010 - 4:37:04 PM
Last modification on : Monday, October 19, 2020 - 11:13:01 AM
Long-term archiving on: : Monday, June 28, 2010 - 4:28:17 PM


  • HAL Id : tel-00466401, version 1
  • PRODINRA : 248288



Vincent Muracciole. Définition et mise en place d'un outil temps réel d'analyse des caractéristiques physiques des semences sèches. Sciences de l'ingénieur [physics]. Université d'Angers, 2009. Français. ⟨tel-00466401⟩



Record views


Files downloads