基于深度学习的古代圆形墓葬检测算法比较研究  

Comparison of the Deep Learning Algorithms for Detecting Circular Ancient Tombs

作　　者：田明艾 于丽君 蒋廷臣 朱建峰 蔡丹路 黄思懿 张渊智 聂跃平 王辉[5] Mingai TIAN;Lijun YU;Tingchen JIANG;Jianfeng ZHU;Danlu CAI;Siyi HUANG;Yuanzhi ZHANG;Yueping NIE;Hui WANG(School of Marine Technology and Geomatics,Jiangsu Ocean University,Lianyungang 222005,China;Joint Laboratory of Remote Sensing Archaeology,Aerospace Information Research Institute,Chinese Academy of Sciences,Beijing 100094,China;School of Resources and Environment,University of Chinese Academy of Sciences,Beijing 100049,China;School of Astronomy and Space Science,University of Chinese Academy of Sciences,Beijing 100049,China;Institute of Archaeological Science,Fudan University,Shanghai 200433,China)

机构地区：[1]江苏海洋大学海洋技术与测绘学院,江苏连云港222005 [2]中国科学院空天信息创新研究院遥感考古联合实验室,北京100094 [3]中国科学院大学资源与环境学院,北京100049 [4]中国科学院大学天文与空间科学学院,北京100049 [5]复旦大学科技考古研究院,上海200433

出　　处：《遥感技术与应用》2024年第5期1249-1260,共12页Remote Sensing Technology and Application

基　　金：国家重点研发计划(2020YFC1521900);国家自然科学基金项目(41801134);江苏海洋科技创新项目(JSZRHYKJ202201);连云港市科技局科技项目(2020058)共同资助。

摘　　要：古代墓葬作为反映早期各社会阶层及其生活状态的典型遗迹,是考古学的重要研究对象,目前有大量墓葬遗迹尚未发现。常规的考古调查手段工作效率不高,尤其在大范围、复杂环境区域难以实现全覆盖。基于深度学习的自动识别方法能够提高考古探测的效率,在遥感考古调查中发挥了重要作用。以新疆阿勒泰地区的古代圆形墓葬为研究对象,构建了首个古代圆形墓葬遥感数据集,选取Faster R-CNN、Cascade R-CNN、YOLOv5和YOLOv7四种典型的深度学习算法开展古代圆形墓葬检测的比较研究。结果表明,YOLOv5和YOLOv7的平均精度超过0.8,YOLOv7最高达到0.87,展现了出色的鲁棒性和广泛的适应性。基于WorldView2、WorldView3和国产GF2卫星数据的进一步测试结果显示,YOLOv7算法在识别具有斑点和环状特征的墓葬时具有明显优势。实验开展的比较研究为实现古代墓葬高精度的自动检测提供了有效方法,同时也为其他考古遗迹的自动检测提供了有价值的参考和借鉴。Ancient tombs,as significant relics that mirror the early social class and its lifestyle,constitute crucial objects of archaeological research.However,numerous ancient tombs remain undiscovered.Traditional archaeological investigations of ancient tombs,especially those conducted in vast and complex environments,are often inefficient.The automatic identification method based on deep learning algorithms has the potential to enhance archaeological detection efficiency and thus plays an important role in remote sensing archaeological investigation.This paper focuses on ancient circular tombs in the Altai region of Xinjiang as the subject of research.We have built the first dataset of ancient circular tombs and conducted a comparative study utilizing four mainstream algorithms,Faster R-CNN,Cascade R-CNN,YOLOv5 and YOLOv7,to detect these tombs.The result reveals that the average precision of YOLOv5 and YOLOv7 exceeds 0.8,with YOLOv7 achieving a peak accuracy of 0.87,showcasing the robustness and adaptability of the YOLO algorithms.Additionally,we have tested the four deep learning algorithms on remote sensing data of varying resolutions,including WorldView2,WorldView3 and GF2 satellites.YOLOv7 emerges as the superior algorithm in terms of both efficiency and accuracy for identifying ancient tombs with blob and ring archaeological markers.This comparative study,not only offers an effective algorithm for the automatic detection of ancient tombs but also provides valuable insights for the automatic detection of other archaeological relics.

关 键 词：遥感考古 深度学习 阿勒泰 古代圆形墓葬 目标检测 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]