This thesis explores the application of Deep Learning models for the automated identification of tree species from anatomical image features. The primary focus is on developing and evaluating the performance of GELAN-C and YOLOv9 series models, specifically GELANC, YOLOv9-e, GELAN-C-seg, and YOLOv9-C-seg. These models leverage advanced architectures and techniques to accurately classify and segment anatomical features in wood images.
The research methodology involved the meticulous collection and annotation of a comprehensive dataset using the Computer Vision Annotation Tool (CVAT). Key anatomical features such as vessels, growth rings, axial parenchyma, and woody rays were labeled to facilitate detailed analysis. The dataset was subjected to extensive preprocessing and data augmentation to enhance model robustness and generalization. It was validated by professors of Escuela Técnica Superior de Ingeniería de Montes, Forestal y del Medio Natural.
Through rigorous training and validation processes, the models demonstrated significant capabilities in both object detection and instance segmentation tasks. GELAN-C and YOLOv9-e models exhibited strong performance in rapid training and inference, with notable precision and recall metrics. The GELAN-C-seg and YOLOv9-C-seg models further improved upon these results by accurately segmenting and classifying anatomical features, achieving high mean Average Precision (mAP) values.
The analysis of confusion matrices provided insights into the models’ strengths and areas for improvement, particularly in handling class imbalances. The comparison with existing methodologies underscored the effectiveness of our approach, which combines advanced deep learning architectures with explainable AI (XAI) techniques to provide transparent and interpretable results.
Key findings highlight the models’ potential applications in the timber industry and environmental conservation, aiding in the regulation of timber trade, enforcement of legal frameworks, and preservation of biodiversity. Despite challenges such as class imbalance, the study’s results indicate a promising future for automated tree species identification using deep learning. This works is related to the GoIMAI project.
This thesis explores the application of Deep Learning models for the automated identification of tree species from anatomical image features. The primary focus is on developing and evaluating the performance of GELAN-C and YOLOv9 series models, specifically GELANC, YOLOv9-e, GELAN-C-seg, and YOLOv9-C-seg. These models leverage advanced architectures and techniques to accurately classify and segment anatomical features in wood images.
The research methodology involved the meticulous collection and annotation of a comprehensive dataset using the Computer Vision Annotation Tool (CVAT). Key anatomical features such as vessels, growth rings, axial parenchyma, and woody rays were labeled to facilitate detailed analysis. The dataset was subjected to extensive preprocessing and data augmentation to enhance model robustness and generalization. It was validated by professors of Escuela Técnica Superior de Ingeniería de Montes, Forestal y del Medio Natural.
Through rigorous training and validation processes, the models demonstrated significant capabilities in both object detection and instance segmentation tasks. GELAN-C and YOLOv9-e models exhibited strong performance in rapid training and inference, with notable precision and recall metrics. The GELAN-C-seg and YOLOv9-C-seg models further improved upon these results by accurately segmenting and classifying anatomical features, achieving high mean Average Precision (mAP) values.
The analysis of confusion matrices provided insights into the models’ strengths and areas for improvement, particularly in handling class imbalances. The comparison with existing methodologies underscored the effectiveness of our approach, which combines advanced deep learning architectures with explainable AI (XAI) techniques to provide transparent and interpretable results.
Key findings highlight the models’ potential applications in the timber industry and environmental conservation, aiding in the regulation of timber trade, enforcement of legal frameworks, and preservation of biodiversity. Despite challenges such as class imbalance, the study’s results indicate a promising future for automated tree species identification using deep learning. This works is related to the GoIMAI project. Read More
Related posts:
Network anomaly detection using Machine Learning techniques
Automated extraction of polyphenols contents in food from the scientific literature using deep learning techniques
Exploring the use of contrastive language-image pre-training for human posture classification: insights from yoga pose analysis
A Novel Cross-Attention-Based Pedestrian Visual–Inertial Odometry With Analyses Demonstrating Challenges in Dense Optical Flow
Estudio de la atención en las arquitecturas transformer
Optimización de Vision Transformer