Fathi Kazerouni, Masoud: Fully-automated plant recognition systems in challenging controlled and uncontrolled environments using classical and Deep Learning methods. 2019
Inhalt
- Abstract
- Zusammenfassung
- Acknowledgments
- Contents
- List of Figures
- List of Tables
- Introduction
- Motivation
- Problem Description
- Classification of Plants
- Real-time Natural Plant Recognition System
- Problems and Needs of Plant Recognition System
- Goals of the Dissertation
- Novel Contributions of the Dissertation
- Combined Feature Detection and Description in Plant Classification
- Classifying a Large Number of Plant Species
- Significant Improvement of Systems for Classification of a Large Number of Plants
- General Natural Plant Recognition based on Modern Combined Algorithms
- Novel Natural Plant Recognition System Based on a Deep Learning Algorithm
- Novelty of Dataset and Systems Implemented for Natural Plant Recognition
- Real-time, Mobile, Natural, Plant Recognition System
- Fully Automated Plant Recognition System
- Document Structure
- Publications
- Literature Review and Fundamentals
- Types of Plant Species
- State-of-the-Art in Plant Recognition
- Short Literature Review for Plant Recognition Systems Based on Neural Networks and Plant Robots
- Summary
- Datasets and Availability
- Classic Datasets
- One-hundred Species Plants Leaf Data Set
- Leaf Shapes Database
- Flavia Dataset
- Swedish Leaf Dataset
- Smithsonian Leaf Dataset
- Semi-Modern Datasets
- Modern Dataset
- Image Analysis
- Investigation of Image Histograms
- Investigation of Histogram Equalization
- Channels of Image and Image Reconstruction
- Conclusion
- Keypoint Detection, Feature Description and Matching
- Related Work
- Keypoint Detection and Feature Extraction
- Local Features: Detection and Description
- Matching
- Combined Detection and Description Methods
- Conclusion
- Implementation and Comparison of Efficient Modern Description Methods for Recognition of Classic Plant Species
- Introduction
- General Overview
- Image Pre-processing
- Bag of Words
- Classifier Training
- Experiment, Discussion, Results and Performance Analysis
- Some Important Metrics for Measuring the Quality of Classifier Systems
- Experiment and Discussion of the Systems by the SIFT Component
- Experiment and Discussion of the Systems by the SURF Component
- Applications of the Proposed Systems
- Acknowledgment
- Conclusions and Future Scope
- Automatic Plant Recognition Systems for Challenging Natural Plant Species using Modern Detection and Description Methods
- Introduction
- Pre-processing Examination
- General Overview
- Approach for Natural Plant Recognition Systems
- Image Pre-processing
- Feature Detection and Extraction
- Modeling and Training
- SVM Classification and Testing
- Experiment, Discussion, Results and Performance Analysis
- Short Description of the Dataset and Setups
- Details of Equipment
- Visual Analysis of Natural Images
- Experiments and Measurements
- A Short Talk on the Experiments, Results and Performances of the Natural Recognition Systems
- Systems Potential for Future Use
- Acknowledgment
- Conclusions and the Future Scope
- Novel System: Deep Learning System for Recognition of Natural Plant Species
- Introduction
- Deep Learning and Neural Networks' Fundamentals
- Deep Learning Timeline
- ANN
- Deep Learning Definitions and Classes
- Deep Learning and Traditional Machine Learning in Classification Tasks
- Proposed Approach
- CNN History and State-of-the-art
- Linked Concepts
- Building Blocks of Deep CNNs and Relevant Definitions
- Convolutional Layer
- Activation Layer
- Pooling Layer
- Fully Connected Layer
- Loss Layer
- Local Response Normalization
- Blob
- Topology of the Proposed Deep CNN Model
- Materials and Equipment
- Experiments and Results
- Classification Accuracy
- Runtime by using GPU and CPU
- Confusion Matrix, Precision and Recall
- Visualization of Proposed Deep Model and Scoring
- Deep CNN, Drawbacks and the Most Recent and Potential Upcoming Breakthrough
- DNPRS, Applications and Future Work
- Conclusion
- Mobile Plant Recognition Robot (Real-time Application of the DNPRS in Challenging Outdoor Environments)
- Introduction
- Related Works
- System Set-Up & Schematic
- Experimental Evaluation and Results
- Feature Work
- Conclusion
- Conclusions
- Appendices
- Implementation of Several Pre-processing Algorithms
- Canny Algorithm (Edge Detector)
- K-means Color Clustering
- Implementation of Grabcut Algorithm
- Superpixel-based Segmentation Algorithm
- Human Nervous System
- Human Learning vs Machine Learning
- Feedforward Neural Network
- Common Deep Learning Frameworks
- Constructed Confusion Matrix for each Proposed System
- References
- Acronyms