Aerial Image Dataset for Construction (AIDCON)

AIDCON 是一个面向施工现场的无人机高空图像数据集，由 UAV 无人机俯视航拍采集构成，旨在支持施工机械检测与分割等任务。共采集 2,155 幅图像，覆盖 30 个不同施工场地（包括开挖、钢结构、运输等真实工况），并标注了 9,563 个施工机械实例。标注采用像素级标注（instance segmentation），包括 9 类对象：倾卸卡车（dump truck）、挖掘机、装载机、压路机、推土机、平地机、轿车及其他类别。图像拍摄海拔高度约在 10–150 米之间，数据跨越 2015–2023 年多个施工项目。AIDCON 数据集伴随发表在 Remote Sensing 期刊的基准论文评估了多种最先进的深度学习检测与分割模型（如 Mask R-CNN、Cascade Mask R-CNN、Hybrid Task Cascade 等），说明其对施工场景下目标检测与语义分割模型的研究价值。该数据集适合用于施工区域识别、重型机械识别与地域覆盖分析等视觉任务。

AIDCON is an unmanned aerial vehicle (UAV) top-down aerial image dataset for construction sites, collected via UAV overhead photography. It aims to support tasks such as construction machinery detection and segmentation. The dataset contains a total of 2,155 images covering 30 distinct construction sites, including real-world working conditions like excavation, steel structure construction, and material transportation, with 9,563 annotated construction machinery instances. The annotations are pixel-level instance segmentation annotations, covering 9 object categories: dump truck, excavator, loader, road roller, bulldozer, grader, sedan, and other categories. The images were captured at an altitude of approximately 10–150 meters, and the data spans multiple construction projects from 2015 to 2023. The AIDCON dataset is accompanied by a benchmark paper published in the journal *Remote Sensing*, which evaluates multiple state-of-the-art deep learning detection and segmentation models such as Mask R-CNN, Cascade Mask R-CNN, Hybrid Task Cascade, etc., demonstrating its research value for target detection and semantic segmentation models in construction scenarios. This dataset is suitable for visual tasks including construction area recognition, heavy machinery recognition, and geographic coverage analysis.

搜集汇总

背景与挑战

背景概述

AIDCON是一个无人机高空图像数据集，包含2,155幅施工现场图像，标注了9,563个施工机械实例，适用于机械检测与分割任务。数据集覆盖30个不同施工场地，包含9类机械对象，支持施工区域识别和重型机械识别等视觉任务。