2026-10-14T06:30:05Z https://api.figshare.com/v2/oai

oai:figshare.com:article/33964987 2026-10-01T11:47:43Z category_28897 category_28864 item_type_3 month_year_10_2026

<b>A UAV image dataset for multimorphology power line detection and segmentation in outdoor environments</b> Xingzhong Zhang (25088539) Yang Yuyuanting (25102160) Computer vision and multimedia computation not elsewhere classified Artificial intelligence not elsewhere classified MPLD power lines UAV imagery power-line detection semantic segmentation binary ground-truth masks <p dir="ltr">MPLD (Multimorphology Power Line Dataset) is a visible-light UAV image dataset developed for power-line structure detection and semantic segmentation in open environments. It contains 1,000 RGB images collected by the research team in the spring of 2024 in a suburban area of a city in eastern China. The images were captured using a DJI Mini 3 UAV at flight heights of approximately 25–40 m and were manually screened, cropped, and resized to 540 × 360 pixels.</p><p dir="ltr">Five annotators used LabelMe to annotate the visible power lines. Each power-line instance is represented as an independent linestrip object defined by an ordered sequence of keypoints. Every image is accompanied by a LabelMe JSON annotation file and a corresponding single-channel PNG binary ground-truth mask. The masks were generated by connecting adjacent keypoints and rasterizing the resulting line segments at a width of 10 pixels using Pillow. In each mask, 0 denotes the background and 255 denotes power-line pixels. The dataset contains 2,079 annotated power-line instances and 17,529 ordered keypoints.</p><p dir="ltr">MPLD includes power lines with low curvature, pronounced curvature, and crossing configurations. It also covers sky, vegetation, forest, and built backgrounds, together with challenging conditions such as low contrast, occlusion, and interference from tower structures. The released files are organized into an 800-image train set and a 200-image test set. In the benchmark experiments reported in the accompanying study, the development set was further divided into 640 training images and 160 validation images. Images, JSON annotations, and binary masks with the same filename stem correspond to the same sample.</p><p dir="ltr">MPLD is provided under the Creative Commons Attribution 4.0 International license (CC BY 4.0). The dataset is provided as is, without warranties regarding its use or any results derived from it.</p> 2026-10-01T11:47:43Z Dataset Dataset 10.6084/m9.figshare.33964987.v1 https://figshare.com/articles/dataset/_b_A_UAV_image_dataset_for_multimorphology_power_line_detection_and_segmentation_in_outdoor_environments_b_/33964987 CC BY 4.0