Synthetic Data for Deep Learning: Segmentation of PCB X-Ray Images
Synthetic Data for Deep Learning: Segmentation of PCB X-Ray Images
复制标题
用于深度学习的合成数据：PCB X 射线图像的分割
DOI：
10.1093/micmic/ozad067.979
复制
发表时间：
2023
影响因子：
2.8
通讯作者：
Tavousi, Pouya
中科院分区：
文献类型：
--
作者：
Phoulady, Adrian;Choi, Hongbin;May, Nicholas;Shahbazmohamadi, Sina;Tavousi, Pouya
逆向工程、缺陷分析和检测是电子设备领域的重要任务。在电子器件检测过程中，X射线成像已被证明是获得高分辨率样品三维图像的最有效的无损技术之一。然而，X射线成像获得的数据一般都很大，手动检查数据可能是一项耗时的任务。为了克服这一挑战，可以实施自动分割系统来辅助检查。为了检查印刷电路板(PCB)，应该首先对电路板进行分割，以区分铜和印刷电路板中的其他材料，如玻璃纤维。这项任务可以使用不同的图像处理技术，包括Otsu的阈值处理[1]、活动轮廓模型[2]和k-均值聚类[3]。虽然这些方法能够有效地分割印刷电路板X射线扫描中的电路，但在准确区分节点和连接以便自动提取网表方面仍然存在重大挑战。
Reverse engineering, defect analysis, and inspection are important tasks in the field of electronic devices. In the process of electronic device inspection, X-ray imaging has proven to be one of the most effective non-destructive techniques for obtaining a 3D view of the sample with good resolution. However, the data obtained by X-ray imaging is generally large, and manual inspection of the data can be a time-consuming task. To overcome this challenge, an automatic segmentation system can be implemented to assist with the inspection.In order to inspect a printed circuit board (PCB), the board should first be segmented to differentiate between copper and other materials in the PCB, such as glass fiber. Different image processing techniques can be used for this task, including Otsu’s thresholding [1], active contour models [2], and k-means clustering [3]. While these methods are capable of effectively segmenting the circuits in a PCB X-ray scan, there is still a significant challenge in accurately differentiating between nodes and connections in order to automatically extract the net list.