无人机遥感影像 影像分割数据集

<p>DroneDeploy无人机遥感影像 影像分割数据集</p><p>数据来源：公开网站</p><p>数据特点：无人机拍摄的图像和相关地理信息数据，包含7种类型的遥感地物类型提取自航空影像drones，由DroneDeploy于2019年发布</p><p><br></p><p><br></p><ul><li>数据类型：包含无人机航拍场景的图像数据，具体包括 RGB 图像（images目录）、高程数据（elevation目录，单通道浮点型 TIFF，像素值代表海拔高度）和标签数据（labels目录，PNG 格式，用 7 种颜色标记不同类别）。</li><li>标签体系：标签涵盖 6 个类别（BUILDING、CLUTTER、VEGETATION、WATER、GROUND、CAR），通过特定颜色编码（如(75, 25, 230)代表BUILDING），需通过color2class等函数转换为类别索引。</li><li>数据处理：原始数据需转换为 300x300 的图像块（chips）用于模型训练，支持数据增强（如水平 / 垂直翻转、旋转等）。</li><li>分辨率：地面分辨率为 10 厘米 / 像素，属于高分辨率航拍数据。</li></ul><p>数据规模：140条数据，其中dataset-sample：0.5GB，为小型样本数据集，适合快速测试模型；dataset-medium：9.0GB，规模更大，包含更多场景和样本。包含训练集、验证集和测试集，通过train_ids、val_ids、test_ids划分</p><p>应用场景：无人机测绘、农业监测、建筑检查、城市规划（建筑物、植被识别）、环境监测（水域、地面覆盖分析）、基础设施管理（clutter区域识别）等领域</p>

DroneDeploy Drone Remote Sensing Image Segmentation Dataset

Data Source: Publicly available websites

Data Characteristics: Images and associated geospatial data captured by drones, containing 7 types of remote sensing land features extracted from drone aerial imagery, released by DroneDeploy in 2019.

- Data Type: Image data for drone aerial survey scenarios, including RGB images (stored in the `images` directory), elevation data (stored in the `elevation` directory: single-channel floating-point TIFF files, where pixel values represent elevation) and annotation data (stored in the `labels` directory, in PNG format, marked with 7 distinct colors for different categories).
- Annotation Schema: The annotations cover 6 categories: BUILDING, CLUTTER, VEGETATION, WATER, GROUND, CAR, with specific color coding (e.g., (75, 25, 230) represents BUILDING). Conversion to class indices is required via functions such as color2class.
- Data Preprocessing: Raw data needs to be cropped into 300×300 image chips for model training, and supports data augmentation techniques including horizontal/vertical flipping, rotation, etc.
- Spatial Resolution: The ground sampling distance is 10 cm per pixel, which belongs to high-resolution aerial survey data.

Data Scale: There are 140 data entries in total. The dataset includes two variants: dataset-sample (0.5GB), a small sample dataset suitable for rapid model testing; and dataset-medium (9.0GB), a larger-scale dataset with more scenarios and samples. It is split into training, validation and test sets via train_ids, val_ids and test_ids respectively.

Application Scenarios: Fields including drone mapping, agricultural monitoring, building inspection, urban planning (building and vegetation recognition), environmental monitoring (water body and land cover analysis), infrastructure management (clutter area recognition), and other related fields.