Titre :	Ensemble Deep Learning-based Semantic Segmentation and Classification of Leaves Images for Agricultural Apple and Wheat Diseases’ Detection
Type de document :	texte imprimé
Auteurs :	Naryméne Kebiche, Auteur ; Faiçal Ammisaid, Auteur ; Abdelouahab Moussaoui, Directeur de thèse
Année de publication :	2023
Importance :	1 vol (165 f .)
Format :	29cm
Langues :	Français (fre)
Catégories :	Thèses & Mémoires:Informatique
Mots-clés :	Plant disease detection Classification
Index. décimale :	004 Informatique
Résumé :	Providing food for the world’s rapidly expanding population is a challenging task with Plant diseases reducing food production due to a demand and supply mismatch, thus,Early detection and appropriate pest control methods are vital to increase crop yield. This master’s thesis proposes reliable approaches for early detection and diagnosis of agricultural plant diseases through the analysis of leaf images. The method uses ensemble deep learning-based semantic segmentation and classification techniques by including transfer learning of CNN-based architectures, a focus on U-Net Attention Mechanisms integration also adapting CNN’s and ViT (Segformer) based encoders, examination of Vision transformers, lastly utilizing ensemble learning to help enhancing models prediction’s confidence. The study also highlights the limited availability of datasets for supervised semantic segmentation in the field of agriculture, resulting in the creation of a new wheat segmentation dataset we conducted Evaluation on public and new datasets demonstrates the method’s efficacy with interesting semantic segmentation scores. Notably, the models achieved up to 77.95% (IoU) and 87.58% (Dice score) for binary semantic segmentation and 98.66% (IoU) and 97.35% (Dice score) for multi-class segmentation on the ATLDSD dataset. Additionally, the models reached 96.30% (IoU) and 98.11% (Dice score) on the newly annotated wheat dataset.
Côte titre :	MAI/0699
En ligne :	https://drive.google.com/file/d/1lxucKFuuP7fc1NLnHviaxI_tE28yAx43/view?usp=drive [...]
Format de la ressource électronique :	pdf

Ensemble Deep Learning-based Semantic Segmentation and Classification of Leaves Images for Agricultural Apple and Wheat Diseases’ Detection [texte imprimé] / Naryméne Kebiche, Auteur ; Faiçal Ammisaid, Auteur ; Abdelouahab Moussaoui, Directeur de thèse . - 2023 . - 1 vol (165 f .) ; 29cm.
Langues : Français (fre)

Catégories :	Thèses & Mémoires:Informatique
Mots-clés :	Plant disease detection Classification
Index. décimale :	004 Informatique
Résumé :	Providing food for the world’s rapidly expanding population is a challenging task with Plant diseases reducing food production due to a demand and supply mismatch, thus,Early detection and appropriate pest control methods are vital to increase crop yield. This master’s thesis proposes reliable approaches for early detection and diagnosis of agricultural plant diseases through the analysis of leaf images. The method uses ensemble deep learning-based semantic segmentation and classification techniques by including transfer learning of CNN-based architectures, a focus on U-Net Attention Mechanisms integration also adapting CNN’s and ViT (Segformer) based encoders, examination of Vision transformers, lastly utilizing ensemble learning to help enhancing models prediction’s confidence. The study also highlights the limited availability of datasets for supervised semantic segmentation in the field of agriculture, resulting in the creation of a new wheat segmentation dataset we conducted Evaluation on public and new datasets demonstrates the method’s efficacy with interesting semantic segmentation scores. Notably, the models achieved up to 77.95% (IoU) and 87.58% (Dice score) for binary semantic segmentation and 98.66% (IoU) and 97.35% (Dice score) for multi-class segmentation on the ATLDSD dataset. Additionally, the models reached 96.30% (IoU) and 98.11% (Dice score) on the newly annotated wheat dataset.
Côte titre :	MAI/0699
En ligne :	https://drive.google.com/file/d/1lxucKFuuP7fc1NLnHviaxI_tE28yAx43/view?usp=drive [...]
Format de la ressource électronique :	pdf