We discuss an archaeological research of employing deep learning (DL) based object detection on high-resolution aerial photographs to discover Nasca geoglyphs, which have been designated as a UNESCO World Heritage Site. Owing to extremely limited archaeological ground truth data and their differences in scale and design, it is difficult to detect new geoglyphs merely training DL on the known geoglyphs. Therefore, we developed a pipeline of DL to mine such data and address the challenges unique to archaeology. With this approach, we identified four new geoglyphs in the northern area of the Nasca Pampa, namely: a humanoid, a pair of legs, a fish, and a bird. The geoglyphs got verified through on-site surveys. We could identify new geoglyph's candidates approximately 21 times faster than with the naked eye alone. The approach would be beneficial for the future of archaeology in a new paradigm of combining field survey and AI.

我们讨论了一项考古研究，即在高分辨率航拍照片上使用基于深度学习 (DL) 的物体检测来发现被指定为联合国教科文组织世界遗产的纳斯卡地画。由于考古地面实况数据极其有限以及它们在规模和设计上的差异，仅在已知地理标志上训练 DL 很难检测到新的地理标志。因此，我们开发了 DL 管道来挖掘此类数据并解决考古学特有的挑战。通过这种方法，我们在 Nasca Pampa 北部地区确定了四个新的地理符号，即：人形、双腿、鱼和鸟。地画经现场勘察核实。我们可以比仅用肉眼快大约 21 倍来识别新的候选地理符号。