A Migration-Assisted Deep Learning Scheme for Imaging Defects Inside Cylindrical Structures via GPR: A Case Study for Tree Trunks

Ground-penetrating radar (GPR) has emerged as a prominent tool for imaging internal defects in cylindrical structures, such as columns, utility poles, and tree trunks. However, accurately reconstructing both the shape and permittivity of the defects inside cylindrical structures remains challenging due to complex wave scattering phenomena and the limited accuracy of the existing signal processing and deep learning techniques. To address these issues, this study proposes a migration-assisted deep learning scheme for reconstructing the shape and permittivity of defects within cylindrical structures. The proposed scheme involves three stages of GPR data processing. First, a dual-permittivity estimation network extracts the permittivity values of the defect and the cylindrical structure, the latter of which is estimated with the help of a novel structural similarity index measure-based autofocusing technique. Second, a modified Kirchhoff migration incorporating the extracted permittivity of the cylindrical structure maps the signals reflected from the defect to the imaging domain. Third, a shape reconstruction network processes the migrated image to recover the precise shape of the defect. The image of the interior defect is finally obtained by combining the reconstructed shape and extracted permittivity of the defect. The proposed scheme is validated using both synthetic and experimental data from a laboratory trunk model and real tree trunk samples. Comparative results show superior performance over existing deep learning methods, while generalization tests on live trees confirm its feasibility for in-field deployment. The underlying principle can further be applied to other circumferential GPR imaging scenarios. The code and database are available at: https://github.com/jwqian54/Migration-Assisted-DL.

💡 Research Summary

This paper presents a novel migration‑assisted deep learning framework for reconstructing both the shape and permittivity of internal defects in cylindrical structures, with a focus on tree trunks. Conventional GPR imaging techniques and existing deep‑learning models are largely designed for flat‑surface scans, where hyperbolic reflections directly map to target locations. In circumferential scans of cylinders, reflections become sinusoidal, multiple internal reflections generate clutter, and the half‑space assumption breaks down, leading to poor accuracy in both geometry and material property estimation.

The proposed scheme addresses these challenges through three tightly coupled stages. First, a Dual‑Permittivity Estimation Network (DPE‑Net) processes raw B‑scan data and simultaneously predicts (i) the equivalent permittivity of the host cylindrical medium and (ii) the permittivity of the defect itself. DPE‑Net incorporates a feature extraction backbone of eight residual blocks, a feature aggregation module that aligns multi‑scale representations, and a Convolutional Block Attention Module (CBAM) to emphasize informative channels and spatial locations. Crucially, the network is trained with supervisory labels for the host permittivity generated by a novel Structural Similarity Index Measure (SSIM)‑based Autofocusing Technique (AFT). Unlike traditional entropy‑based AFT, the SSIM‑AFT evaluates the geometric fidelity of migrated images, ensuring that the selected host permittivity yields a migration that best preserves the defect’s shape.

Second, the estimated host permittivity is fed into a modified Kirchhoff migration algorithm. Because the correct wave velocity is known a priori, the migration can be performed in a single pass (“one‑shot”), dramatically reducing the computational burden that plagues conventional migration (which typically requires a sweep over many candidate velocities). The migration maps the sinusoidal reflections from the B‑scan onto the spatial domain of the cylinder, producing an image where the defect appears as a bright, albeit cluttered, region.

Third, a Shape Reconstruction Network (SR‑Net) refines this migrated image. SR‑Net follows an encoder‑decoder architecture with residual blocks, feature aggregation, and CBAM‑driven attention. It learns to suppress background clutter, resolve multiple reflections, and output a clean binary mask of the defect’s geometry. By combining the mask with the defect permittivity predicted by DPE‑Net, the method yields a full permittivity map of the internal anomaly.

The authors validate the framework on three data tiers: (1) synthetic datasets covering a wide range of cylinder radii, layer thicknesses, defect sizes, and shapes; (2) laboratory measurements using a fabricated tree‑trunk model with known defects; and (3) real‑world tree trunk samples, including live‑tree field tests. Quantitative metrics (Intersection‑over‑Union for shape, RMSE for permittivity) show consistent improvements over state‑of‑the‑art flat‑scan networks such as MRF‑UNet, PiNet, and hybrid MultiPath‑Net. Notably, the one‑shot migration reduces processing time to under one second per scan, making near‑real‑time inspection feasible.

The paper’s contributions are threefold: (i) a physics‑consistent, migration‑assisted deep learning pipeline that bridges the domain gap between B‑scan and image domains for cylindrical scanning; (ii) an SSIM‑based autofocusing strategy that eliminates iterative velocity searches and yields accurate host permittivity estimates; (iii) two specialized neural networks (DPE‑Net and SR‑Net) that jointly maximize performance on permittivity estimation and shape recovery.

Limitations include reliance on simulated SSIM‑AFT labels during offline training, potential degradation when faced with highly irregular or multiple defects, and the need to retrain if GPR hardware parameters (frequency, antenna geometry) change substantially. Future work is suggested on label‑free or weakly supervised training, extension to multi‑defect scenarios, and application to other cylindrical infrastructures such as utility poles, pipelines, or cultural‑heritage columns.

Overall, the study demonstrates that integrating a physics‑driven migration step with carefully designed deep networks can achieve accurate, efficient, and deployable GPR‑based defect imaging in cylindrical objects, moving the technology from laboratory proof‑of‑concept toward practical field deployment.