The objective of this thesis is to develop a way to detect weeds in a field using multispectral images, in order to determine which weeds should be eliminated during the current crop cycle and more particularly at the early stages. The multi-criteria approach focuses on the spatial arrangement, the spectral signature, the morphology and the texture of the plants located in the plots. This work proposes a method for selecting the best criteria for optimal discrimination for a given setup. Prior to the extraction of these criteria, a set of methods was developed in order to correct the errors of the acquisition device, to precisely detect the vegetation and then to identify within the vegetation the individuals on which the different criteria can be computed. For the individual detection step, it appears that leaf scale is more suitable than plant scale. Vegetation detection and leaf identification are based on deep learning methods capable of processing dense foliage. The introduction of these methods in a usual processing chain constitutes the originality of this manuscript where each part was the subject of an article. Concerning the acquisition device, a method of spectral band registration was developed. Then, new vegetation indices based on artificial intelligence constitute one of the scientific advances of this thesis. As an indication, these indices offer a mIoU of 82.19% when standard indices ceil at 63.93%-73.71%. By extension, a leaf detection method was defined and is based on the detection of their contours, this method seems advantageous on our multispectral data. Finally, the best property pairs were defined for crop/weed discrimination at leaf level, whith classification performances up to 91%.
Authors
- Authors Organism
- Institut National de Recherche pour l’Agriculture, l’Alimentation et l’Environnement
- Bibliographic Reference
- Jehan-Antoine Vayssade. Approche multi-critère pour la caractérisation des adventices. Intelligence artificielle [cs.AI]. Université Bourgogne Franche-Comté, 2022. Français. ⟨NNT : ⟩. ⟨tel-03688127⟩
- HAL Collection
- ['GIP Bretagne Environnement', 'Institut National de Recherche en Agriculture, Alimentation et Environnement', 'ANR', 'Agroécologie, génétique et systèmes d’élevage tropicaux', 'Réseau "Systèmes Agricoles et Eau"']
- HAL Identifier
- 3688127
- Institution
- Institut National de Recherche pour l’Agriculture, l’Alimentation et l’Environnement
- Laboratory
- Agroécologie, génétique et systèmes d’élevage tropicaux
- Published in
- France
Table of Contents
- Introduction 12
- Agriculture : un avènement 12
- Dynamique des systèmes 13
- Limites de l'agriculture intensive 14
- Modes de production 17
- Transition agroécologique 18
- Agriculture de précision 19
- La télédétection au service de l'agriculture de précision 20
- Vers une agriculture numérique 22
- Le désherbage 23
- Place des adventices dans les systèmes de culture 23
- La biologie des adventices 25
- Fenêtres de traitement 27
- Les technologies de désherbage actuelles 28
- Les technologies de désherbage en développement 30
- Conclusion de chapitre 32
- Vision par ordinateur 34
- Systèmes d'imagerie 35
- Percevoir la couleur 37
- Des capteurs multispectraux 38
- Prétraitement 39
- Correction géométrique 39
- Rehaussement d'image 40
- Indices de télédétection 41
- Segmentation 45
- Segmentation sémantique 45
- Segmentation en région 47
- Segmentation des instances 50
- Extraction de propriétés 52
- Propriétés géométriques 52
- Propriétés spatiales 53
- Propriétés de texture 53
- Robustesse des propriétés 54
- Apprentissage profond 54
- Classification 55
- Classification supervisée 55
- Classification non-supervisée 58
- Classification semi-supervisée 59
- Métriques d'évaluation 60
- Classification 60
- Régression et calibration 61
- Applications en agriculture 62
- Échelle du pixel et voisinage proche 62
- Échelle de la fenêtre 63
- Composantes connexes et super-pixels 64
- Échelle de la plante 66
- Conclusion de chapitre 67
- Matériel et données 68
- Les sites expérimentaux 68
- Capteur optique 69
- Jeux de données 69
- Campagnes d'acquisition 70
- Annotations 71
- Calibration 73
- Résolution spatiale 73
- Correction géométrique 74
- Recalage des bandes spectrales 75
- Introduction 78
- Material and Method 79
- Results and discussion 82
- Conclusion 87
- Conclusion de chapitre 88
- Indices de végétation 90
- Introduction 92
- Material and Data 94
- Instrument Details 94
- Image Dataset 94
- Data Pre-Processing 95
- Training and Validation Datasets 96
- Methodology 96
- Existing Spectral Indices 96
- Deepindices: Baseline Models 96
- Enhancing Baseline Models 99
- Last Activation Function 100
- Loss Function 101
- Performance Evaluation 101
- Comparison with Standard Indices 101
- Training Setup 102
- Results and Discussion 103
- Fixed Models 103
- Deepindices 104
- Initial Image Processing 106
- Discussion 106
- Conclusions 110
- Conclusion de chapitre 113
- Détection des individus 114
- Introduction 117
- Related Works 117
- Objectives 118
- Material and data 119
- Specific multispectral dataset 119
- Online image databases 120
- Methodology 121
- Proposed CNN architecture 121
- Loss function 125
- Refinement with vegetation mask and watershed 127
- Training setup 127
- Evaluation metrics 128
- Results 128
- Komatsuna dataset 128
- Leaf Segmentation Challenge dataset 130
- Airphen dataset 134
- Discussion 136
- Conclusion 137
- Further research 138
- Conclusion de chapitre 139
- Extraction de propriétés 140
- Introduction 142
- Material and data 143
- Experimental plot 143
- Multispectral camera 144
- Image acquisition and annotation 144
- Methodology 144
- Literature review 145
- Optimization 150
- Feature Selection 152
- Results and Discussion 152
- Feature optimisation 152
- Feature selection by type 153
- Best of all features 156
- Visual results 157
- Conclusion 159
- Further research 160
- Conclusion de chapitre 163
- Conclusion Générale 164
- Contributions 165
- Perspectives 166
