ABSTRACT In this work, a novel migration method is applied to Ground-Penetrating Radar (GPR) data to detect the internal flaws of ornamental stone blocks. To detect and classify fractures in accordance with their spatial orientation, a Shift-Invariant Probabilistic Latent Component Analysis (SI-PLCA) is proposed. GPR simulations are conducted using modeling software to test several types of fractures (with different positions, thicknesses, and lengths) in rock blocks and to train several patterns as inputs for the SI-PLCA method. An 800 MHz antenna is used to assess both simulated and real data. The accuracy rate of the proposed approach is evaluated and compared with that of classical migration methods for detection and is compared to a Template Matching approach for classification; promising results are obtained. In addition, GPR is applied to two blocks of a rock type known commercially as Crema Marfil. The 3D fracture maps obtained from the proposed approach are compared with the stone slabs from the cutting process. The results show that the proposed approach applied to GPR radargrams is an effective method for determining the internal structure of stone materials, particularly for detecting and classifying fractures.

中文翻译：

---

基于 SI-PLCA 解卷积技术应用于 GPR 信号的石灰石块体内部缺陷检测和分类

摘要 在这项工作中，一种新的偏移方法应用于探地雷达 (GPR) 数据，以检测装饰石块的内部缺陷。为了根据裂缝的空间方向对裂缝进行检测和分类，提出了一种移位不变概率潜在成分分析（SI-PLCA）。GPR 模拟是使用建模软件进行的，以测试岩块中的几种类型的裂缝（具有不同的位置、厚度和长度），并训练几种模式作为 SI-PLCA 方法的输入。800 MHz 天线用于评估模拟数据和真实数据。对所提出的方法的准确率进行了评估，并与经典的检测迁移方法进行了比较，并与用于分类的模板匹配方法进行了比较；取得了可喜的成果。此外，GPR 应用于商业上称为 Crema Marfil 的岩石类型的两个块。从所提出的方法获得的 3D 断裂图与切割过程中的石板进行了比较。结果表明，所提出的方法应用于探地雷达雷达图是确定石材内部结构的有效方法，特别是用于检测和分类裂缝。