Skip to Main content Skip to Navigation
New interface
Theses

Éléments pour l'analyse et le traitement d'images : application à l'estimation de la qualité du bois

Rémi Decelle 1 
1 ADAGIO - Applying Discrete Algorithms to Genomics and Imagery
LORIA - ALGO - Department of Algorithms, Computation, Image and Geometry
Abstract : In the context of climate change and its mitigation, the development of the circular bioeconomy, and the resulting increased pressure on the wood resource, one of the priority research areas is to optimize the transformation of this resource, which can be done at different levels. The objective here is to extract information on wood quality from the analysis of cross-sections of logs or logs in the forest or in the sawmill. To estimate this quality, several visible characteristics can be extracted: sapwood and heartwood areas, position of the pith and geometric center, number of tree rings and their width. First, we are interested in the segmentation of the log in the image. This segmentation makes the analysis of other features easier and allows us to locate the geometric center. For this, we propose several approaches. First, classical methods from image processing are discussed, such as the K-Means method or active contours. We also use convolutional neural networks. We show the advantage of neural networks over the other methods. The second estimated feature is the duramen area (more colored central area). We propose a new attention layer for neural networks using the less commonly used mathematical morphology. Attention layers have allowed networks to outperform by focusing on the most relevant information. In our context, this layer has two goals: reduceing the amount of parameters and increasing performance. Our attention layer shows better performances compared to other attention layers. In the third step, we propose to analyze the tree rings. Our method is in three main steps. First, a directional smoothing to enhance the rings ( keeping the edges as well as possible) and reduce both the intra tree rings texture and the saw marks. Then, an adaptive thresholding to segment the areas of potential tree rings. Finally, a second thresholding in order to have the ring limits. From the final segmentation, the analysis of the rings (number, average width, etc.) is made possible. Finally, the estimation of the pith position is addressed. We propose a new original approach based on the ant colony algorithm to estimate the location of the pith. Using this approach allows to skip a usual step, namely the way to accumulate normals at the tree ring tangents. Our method shows many advantages compared to state-of-the-art approaches, including neural networks. In a last part, we will present a work in discrete geometry: a directional filter. It estimates the maximal digital straight segments at any point of a connected set. This work is intended to be applied to estimate some tree ring features.
Complete list of metadata

https://hal.univ-lorraine.fr/tel-03794911
Contributor : Thèses UL Connect in order to contact the contributor
Submitted on : Monday, October 3, 2022 - 3:59:38 PM
Last modification on : Tuesday, October 4, 2022 - 3:33:02 AM

File

DDOC_T_2022_0105_DECELLE.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : tel-03794911, version 1

Citation

Rémi Decelle. Éléments pour l'analyse et le traitement d'images : application à l'estimation de la qualité du bois. Informatique [cs]. Université de Lorraine, 2022. Français. ⟨NNT : 2022LORR0105⟩. ⟨tel-03794911⟩

Share

Metrics

Record views

44

Files downloads

2