Abstract:

Objective: In response of the problems of low detection accuracy and inaccurate defect contour prediction in existing wood stress wave tomography algorithms, a wood defect detection framework integrating the SVAM and the LVCC algorithm was constructed to systematically improve the accuracy of stress wave tomography, providing technical support for the efficient detection of wood defects. Method: The imaging area was divided into grids to form the basic units for imaging. Stress wave propagation paths were established between each sensor and its neighboring sensors to form the basic sectors required for interpolation in the SVAM algorithm. The velocity on each sector edge represented the stress wave propagation speed along that path. The grid velocity was determined by all sectors covering that grid. The weighted sum of the influence velocities from these sectors was calculated to obtain the velocity values for all grids. The intersection points of all paths were computed to form a set of low-velocity points. A convex hull was constructed for this set, and the region within the convex hull is used as a constraint area to improve the accuracy of defect prediction. Result: Comparative experiments were conducted on four real wood samples and three simulated samples using the Fakopp algorithm, the RSIA algorithm, and the SVAM algorithm. Additionally, comparative experiments were performed on two Salix babylonica samples using the Fakopp algorithm and the SVAM algorithm. In the experiments on the four real log samples and three simulated samples, the SVAM algorithm achieved accuracy, precision, and recall rates of 85.5%, 94.1%, and 89.3%, respectively, outperforming the Fakopp algorithm (79.9%, 97.8%, and 78.8%) and the RSIA algorithm (80.9%, 92.8%, and 84.8%). In the experiments on the two living S. babylonica samples, both the SVAM and Fakopp algorithms demonstrated good prediction performance. Cross-validation was conducted on two paths of the defective sample using a micro-drilling resistance instrument. The prediction errors of the SVAM algorithm were 2.09% and 9.99% lower than those of the Fakopp algorithm, respectively. The experimental results validated the effectiveness of the proposed method in defect prediction across different types of samples. Conclusion: The proposed SVAM algorithm can effectively detect wood defects in three different types of data samples. The detection performance is less affected by the size, shape, and location of defects, demonstrating high robustness. The algorithm provides a reliable solution for accurate defect detection in trees.

Key words: stress wave tomography imaging, sector velocity attenuation model (SVAM), low velocity contour constraint (LVCC), convex hull