Acoustic Tomography of Wood Internal Defects with Convolutional Neural Network

Acoustic tomography is a viable approach to assess internal defects within tree trunks. The Simultaneous Iterative Reconstruction Technique (SIRT) is commonly used to reconstruct the internal wood structure, but it suffers from reduced precision. To address this issue, this paper presents a method that employs a Convolutional Neural Network (CNN) to conduct acoustic tomography on live trees with exceptional accuracy. The primary objective is to generate a tomographic image that accurately indicates the presence, locations, and dimensions of any cavities within the tree trunk. Non-destructive devices have been developed to facilitate easier data acquisition from the trees. Furthermore, the use of batch normalization and the Adam optimizer on the CNN enables fast convergence of the training process. The proposed method was evaluated using both simulation and experimental data under various settings. The area and shape of the imaging results were analyzed, and the method was compared with existing techniques, such as SIRT, in terms of accuracy and overall image quality. The comparison results show that the proposed CNN-based method can produce much better tomographic reconstructions than SIRT.