InsPLAD: A Dataset and Benchmark for Power Line Asset Inspection in UAV Images
收藏资源简介：
This article has been accepted for publication in the International Journal of Remote Sensing, published by Taylor & Francis. InsPLAD is a Power Line Asset Inspection Dataset and Benchmark containing 10,607 high-resolution Unmanned Aerial Vehicles color images. The dataset contains seventeen unique power line assets captured in the real world and annotated for object detection. In addition, five assets presented six types of defects, annotated into normal and defective samples on an image level. We thoroughly evaluate state-of-the-art and popular methods for three computer vision tasks covered by InsPLAD: object detection, image classification, and anomaly detection. The first aims to detect power line assets in UAV images, and the other two to classify defects in cropped power line asset images. InsPLAD offers a wide range of vision-related challenges, such as multi-scale objects, multi-size class instances, multiple objects per image, intra-class variation, cluttered background, uncontrolled environment, distinct point-of-views, perspective distortion, occlusion, and varied lighting conditions. Our benchmark indicates considerable room for improvement in the state-of-the-art methods. InsPLAD is the first large real-world dataset for power line asset inspection with multiple components and defects.
本文已被Taylor & Francis出版的《国际遥感学报（International Journal of Remote Sensing）》录用待刊。InsPLAD是一款电力线路资产巡检数据集与基准测试集，包含10,607张高分辨率无人机（Unmanned Aerial Vehicles, UAV）彩色图像。该数据集涵盖17类真实采集的电力线路资产，并针对目标检测任务完成了标注。此外，其中5类资产存在6种缺陷类型，且在图像层面被标注为正常与缺陷两类样本。本研究针对InsPLAD覆盖的三大计算机视觉任务——目标检测、图像分类与异常检测——全面评估了当前主流与前沿的算法模型。其中目标检测任务旨在从无人机航拍图像中定位电力线路资产，后两项任务则需对裁剪后的电力线路资产图像进行缺陷分类。InsPLAD涵盖了诸多计算机视觉领域的典型挑战，包括多尺度目标、多尺寸类别实例、单图多目标、类内差异、复杂背景、非受控拍摄环境、多样化拍摄视角、透视畸变、目标遮挡以及多变光照条件等。本基准测试结果显示，当前前沿算法仍存在较大的性能提升空间。InsPLAD是首个面向电力线路资产巡检的大规模真实场景数据集，涵盖多类资产与多种缺陷类型。
