为提升对高坝大库在自然灾害作用下的结构表面缺陷识别能力，研究团队于2023-2025年期间不断完善构建了裂缝、渗漏、滑坡典型缺陷数据集。数据集共包含3532张高分辨率影像，涵盖滑坡、裂缝、渗水、析出物等典型缺陷。影像来源于无人机、轮式机器人、挂轨机器人及多类型摄像头等巡检设备拍摄，经过预处理与裁切后形成标准化样本。
", "ds_source": "数据集的影像一部分来自于CRACK500公开数据集，一部分来自高坝大库工程现场采集整理，现场采集影像由研究人员利用无人机、轮式机器人、挂轨机器人及多类型摄像头等巡检设备在高坝上下游表面、内部混凝土廊道、水库库岸边坡等部位不同气候、气象条件下获取。
", "ds_process_way": "（1）使用无人机、轮式机器人、挂轨机器人及多类型摄像头在不同气候、气象条件下采集库坝区域影像，共获得3532张含滑坡、裂缝、渗水、析出物等缺陷的高分辨率图像；\n
（2）对原始影像进行畸变校正、光照均衡与降噪处理，并裁剪为标准尺寸图像块（640×640像素），以满足深度学习模型的输入要求。
", "ds_quality": "数据集共包含3532张分辨率不低于640×640的影像，覆盖滑坡、裂缝、渗水、析出物等多类典型缺陷。影像由无人机、轮式机器人、挂轨机器人及多类型摄像头在不同光照、季节与环境条件下采集，具有良好的场景多样性和工程真实性。同时，影像经过预处理与裁剪，提高了数据质量一致性与模型训练适配性,为高坝大库结构表面缺陷识别与智能巡检研究提供了高质量的数据支持。
\n使用时，需根据研究需求自行完成标注工作。推荐使用流程为：首先，根据具体任务需求选择合适的标注工具，并制定统一的标注规范，针对裂缝、渗漏、滑坡、析出物等缺陷类别进行精确标注，建议由具备工程背景的人员执行以确保标注准确性；其次，将标注完成的数据集按合理比例划分为训练集、验证集和测试集；最后，根据研究目标选择适合的深度学习模型进行训练，在验证集上调优超参数，并在测试集上评估模型性能。
", "ds_acq_start_time": "2023-01-01 00:00:00", "ds_acq_end_time": "2025-10-31 00:00:00", "ds_acq_place": "", "ds_acq_lon_east": null, "ds_acq_lat_south": null, "ds_acq_lon_west": null, "ds_acq_lat_north": null, "ds_acq_alt_low": null, "ds_acq_alt_high": null, "ds_share_type": "login-access", "ds_total_size": 254822024, "ds_files_count": 0, "ds_format": "*.jpg (图像), *.png (标注掩码)", "ds_space_res": "", "ds_time_res": "", "ds_coordinate": "无", "ds_projection": "", "ds_thumbnail": "91a7e36a-58a0-4aa0-af0c-92340a0963d9.png", "ds_thumb_from": 0, "ds_ref_way": "", "paper_ref_way": "", "ds_ref_instruction": "", "ds_from_station": "", "organization_id": "c0a03a3b-e859-4f8b-8abe-a6a23a81aedb", "ds_serv_man": "毛莺池", "ds_serv_phone": "13951029973", "ds_serv_mail": "yingchimao@hhu.edu.cn", "doi_value": "", "subject_codes": [ "170.50" ], "quality_level": 0, "publish_time": "2026-06-11 17:47:32", "last_updated": "2026-06-11 17:47:32", "protected": false, "protected_to": "2027-12-01 00:00:00", "lang": "zh", "cstr": "", "i18n": { "en": { "title": "Typical defect dataset of cracks, leakage, and landslides in high dams and large reservoirs(2023-2025)", "ds_format": "*. jpg (image), *. png (annotation mask)", "ds_source": "Part of the images in the dataset are from the publicly available CRACK500 dataset, and another part are collected and organized from the site of the high dam and large reservoir project. The on-site collected images were obtained by researchers using inspection equipment such as drones, wheeled robots, rail mounted robots, and various types of cameras under different climatic and meteorological conditions on the upstream and downstream surfaces, internal concrete corridors, and reservoir bank slopes of the high dam.
", "ds_quality": "The dataset contains 3532 images with a resolution of no less than 640 × 640, covering various typical defects such as landslides, cracks, water seepage, and precipitates. The images are collected by drones, wheeled robots, rail mounted robots, and various types of cameras under different lighting, seasons, and environmental conditions, with good scene diversity and engineering authenticity. At the same time, the image has been preprocessed and cropped to improve data quality consistency and model training adaptability, providing high-quality data support for surface defect recognition and intelligent inspection research of high dam and large reservoir structures.
", "ds_ref_way": "", "ds_abstract": "In order to enhance the ability to identify structural surface defects of high dams and large reservoirs under natural disasters, the research team continuously improved and constructed typical defect datasets for cracks, leaks, and landslides from 2023 to 2025. The dataset contains a total of 3532 high-resolution images, covering typical defects such as landslides, cracks, water seepage, and precipitates. The images are taken by inspection equipment such as drones, wheeled robots, rail mounted robots, and various types of cameras, and are preprocessed and cropped to form standardized samples.
", "ds_time_res": "", "ds_acq_place": "", "ds_space_res": "", "ds_projection": "", "ds_process_way": "(1) Using drones, wheeled robots, rail mounted robots, and various types of cameras to collect images of the reservoir dam area under different climatic and meteorological conditions, a total of 3532 high-resolution images containing defects such as landslides, cracks, water seepage, and precipitates were obtained; \r\n
(2) Perform distortion correction, lighting balance, and noise reduction on the original image, and crop it into standard sized image blocks (640 × 640 pixels) to meet the input requirements of the deep learning model.
", "ds_ref_instruction": "" } }, "submit_center_id": "ncdc", "data_level": 0, "recommendation_value": 0, "license_type": "https://creativecommons.org/licenses/by/4.0/", "doi_reg_from": "reg_local", "cstr_reg_from": "reg_local", "doi_not_reg_reason": null, "cstr_not_reg_reason": null, "is_paper_in_submitting": false, "belong_to_nieer": false, "ds_topic_tags": [ "裂缝检测", "图像分割", "深度学习", "计算机视觉" ], "ds_subject_tags": [ "地质学" ], "ds_class_tags": [], "ds_locus_tags": [], "ds_time_tags": [ 2023, 2024, 2025 ], "ds_contributors": [ { "true_name": "毛莺池", "email": "yingchimao@hhu.edu.cn", "work_for": "河海大学", "country": "中国" } ], "ds_meta_authors": [ { "true_name": "毛莺池", "email": "yingchimao@hhu.edu.cn", "work_for": "河海大学", "country": "中国" } ], "ds_managers": [ { "true_name": "毛莺池", "email": "yingchimao@hhu.edu.cn", "work_for": "河海大学", "country": "中国" } ], "category": "其他" }