Images for Inland navigation and vessel optimal path planning
收藏资源简介:
Laboratoire d'Automatique et de Robotique interactive, Université du Québec à Chicoutimi, Québec, Canada LAR.i - UQAC, 02 Feb. 2024 855 satellite images and their ground truth masks. The original satellite images could be accessed by email. Manual acquisition of 60 unlabelled aerial satellite images with varying dimensions (area ranging from 500 meters to 5 kilometers) from Google Maps. A manual segmentation is executed into three labels: water, land, and obstacle. Parametric augmentation pipeline: -random cropping, -horizontal and vertical flipping, -rotation, -brightness and contrast, -Contrast Limited Adaptive Histogram Equalization (CLAHE) and -Grid Distortion, or Optical Distortion.
加拿大魁北克省魁北克大学希库蒂米分校交互式自动化与机器人实验室(Laboratoire d'Automatique et de Robotique interactive,简称LAR.i - UQAC)于2024年2月2日发布的数据集包含855幅卫星图像及其真值掩膜(ground truth masks)。原始卫星图像可通过邮件申请获取。此外,数据集还包含60幅未标注航空卫星图像,通过手动从谷歌地图(Google Maps)采集,这些图像尺寸各异,单幅覆盖区域范围为500米至5千米。研究人员对所有图像执行手动分割,将其标注为三类:水体、陆地与障碍物。数据集采用参数化数据增强流程,具体包括:随机裁剪、水平与垂直翻转、旋转、亮度与对比度调整、对比度受限自适应直方图均衡化(Contrast Limited Adaptive Histogram Equalization,简称CLAHE),以及网格畸变(Grid Distortion)或光学畸变(Optical Distortion)。



