Initial Situation
Existing solutions for tree ring analysis often require specialized equipment or are only suitable for professional users. An automated, user-friendly, and hardware-independent method has not yet been available, which this work aims to address.
Motivation and Research Questions
Tree ring analysis is central in fields such as climate research and archaeology. Modern image processing and machine learning offer new possibilities to automate age determination. The study investigates questions regarding a modular pipeline, algorithm suitability, validation, and integration into mobile applications.
Objectives
The goal is to develop a system that segments tree trunk cross-sections, determines the trunk’s center, and analyzes tree rings for age estimation – all within a user-friendly mobile and web application.
Structure of the Work
The work is divided into theoretical fundamentals, system concept, development of AI models, implementation of analysis modules, evaluation, and finally, a conclusion.
The system is based on a modular architecture with separate components for segmentation, center detection, ring analysis, API, and mobile application.
A YOLO-v11 is used to segment the tree trunk cross-section from an input image. The training dataset comprises over 1127 images and 25,256 cross-section annotations.
The modules for segmentation, center detection, and ring analysis are developed as standalone Python packages. CI/CD pipelines are also set up for automatic deployment (e.g., to PyPI).
The tree disk analyzer system exposes the following RESTful API endpoints to interact with the application:
/segmentation/image Segment a tree trunk image
/pith/detectDetect the center of a tree trunk
/rings/detectDetect tree rings in a segmented image
/healthRetrieve health status of the system
The frontend of the tree disk analyzer system is developed using Expo.
The performance of the models is evaluated using classic regression metrics (MAE, RMSE, MPE) as well as a tolerance-based accuracy measure.
| Metric | σ=1 | σ=2 | σ=3 | σ=4 |
|---|---|---|---|---|
| Exact Match (%) | 1.56 | 12.5 | 14.06 | 26.56 |
| Tolerance ±1 Year (%) | 9.38 | 23.44 | 53.13 | 45.31 |
| MAE (years) | 7.02 | 3.63 | 2.22 | 2.73 |
| RMSE (years) | 8.56 | 4.87 | 3.65 | 4.61 |
| MPE (%) | 38.22 | 19.48 | 11.79 | 14.07 |
The results indicate that a moderate smoothing parameter (σ=3) achieves the best prediction accuracy – for example, a MAE of only 2.22 years.
The developed tree ring analysis enables automated age determination with high accuracy. Despite some challenges in edge detection, the system delivers practically relevant results and paves the way for further improvements, such as integrating additional deep-learning approaches.