Machine learning based tomographic image reconstruction technique to detect hollows in wood

• Published in: Wood science and technology Vol. 58; no. 4; pp. 1491 - 1516
• Main Authors: Yıldızcan, Ecem Nur, Arı, Mehmet Erdi, Tunga, Burcu, Gelir, Ali, Kurul, Fatih, As, Nusret, Dündar, Türker
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.07.2024; Springer Nature B.V
• Subjects: Algorithms; Biomedical and Life Sciences; Ceramics; Classification; Clustering; Composites; Defects; Glass; Image processing; Image reconstruction; Image segmentation; Learning algorithms; Life Sciences; Machine learning; Machines; Manufacturing; Natural Materials; Original; Processes; Propagation velocity; Segmentation; State-of-the-art reviews; Stress propagation; Stress waves; Velocity; Wave propagation; Wave velocity; Wood Science & Technology
• ISSN: 0043-7719; 1432-5225
• DOI: 10.1007/s00226-024-01580-z

A new technique based on machine learning algorithms was introduced to detect internal wood defects. This technique relies on analyzing segmented propagation rays of stress waves and successfully generates the tomographic images of the defects by using the stress wave velocity. Utilizing a dual-stage methodology, the initial phase involves ray segmentation for the precise delineation of stress wave propagation, while the subsequent stage integrates advanced classification and clustering algorithms to facilitate the generation of tomographic images. This approach effectively tackles the inherent challenges associated with accurate segmentation and classification of stress wave velocity rays. The effectiveness of the proposed method was evaluated using both synthetic and experimental data. The results showed that the proposed method, when compared with some state-of-the-art methods, has a superior ability to accurately detect defective regions in the wood. The success of the proposed method is evaluated with four different evaluation metrics. It determined that over 90% success is achieved for all metrics. In comparison with related studies, it determined that the results are improved by 7–22% compared to the literature.