Terahertz time-of-flight tomography algorithm for different curvature samples

Traditional guide rail-based terahertz reflection tomography works well with planar samples and complements transmission tomography. Most imaging methods perform effectively with plane-like or simple curvature samples. However, these methods are ineffective for complex shapes due to the challenges of dealing with internal corrections and stitching errors. To address the above issues, terahertz time-of-flight tomography is applied to the prepared samples with abnormal curvature using a rotating scanning array and "point cloud splicing" method. This approach enables comprehensive sample information collection and resolves the inner layer point cloud position correction challenge. A radial distance point cloud down-sampling method is proposed to reduce the impact of overlapping point clouds on reconstruction accuracy. After down-sampling, the outer layer reconstruction error reduces to 0.22%. Poisson surface reconstruction further enhances the intuitiveness of imaging results, and the reconstruction accuracy is 0.01 mm. This study broadens the application of terahertz time-of-flight tomography and offers a different approach for reflection tomography.