新疆北部乡村道路实例分割图像数据集  

An dataset of rural roadway images of instance segmentation in Northern Xinjiang

作　　者：希仁娜 张太红[1,2,3] 姚芷馨 XI Renna;ZHANG Taihong;YAO Zhixin(Xinjiang Agricultural University,Urumqi 830052,P.R.China;Ministry of Education Engineering Research Center for Intelligent Agriculture,Urumqi 830052,P.R.China;Xinjiang Agricultural Informatization Engineering and Technology Research Center,Urumqi 830052,P.R.China)

机构地区：[1]新疆农业大学计算机与信息工程学院,乌鲁木齐830052 [2]智能农业教育部工程研究中心,乌鲁木齐830052 [3]新疆农业信息化工程技术研究中心,乌鲁木齐830052

出　　处：《中国科学数据（中英文网络版）》2024年第4期427-441,共15页China Scientific Data

基　　金：科技创新2030—“新一代人工智能”重大项目(2022ZD0115805);新疆维吾尔自治区重大科技专项(2022A02011)。

摘　　要：基于深度学习的乡村道路自动识别可助力于智能农机技术的发展,通过对道路及周边物体的识别,农机可实现自动导航,从而缓解农村劳动力短缺问题。目前实现机动车无人驾驶主要是依靠城市道路尤其是结构化道路数据集的支持,而针对乡村道路尤其是非结构化道路的数据集尚未发表。本数据集聚焦新疆北部地区,涵盖不同时间、不同场景的1285张高清乡村道路图像。对图像预处理后,设定了40种类别,包含20种实例类别,利用CVAT标注工具进行了细致的像素级别人工标注,共标注了10062个实例对象,通过多重检查保证标注数据的可靠性、完整性和统一性。随后与主流的道路数据集在静态属性上进行了对比,选取3个经典的实例分割模型进行训练与验证,利用多个评估指标进行了评估。本数据集由乡村道路高清图像文件、与原图一一对应的掩膜二值图像(掩码MASK图)和MS COCO格式的标注文件构成,可为研究乡村道路实例分割和其他深度学习下的任务提供数据支持。The automated identification of rural roadways based on deep learning methodologies contributes to the advancement of sophisticated agricultural machinery technologies.By discerning road networks and surrounding topographical features,agricultural equipment can seamlessly navigate rural landscapes,effectively mitigating the challenges posed by labor scarcity in these regions.This dataset,centered around Northern Xinjiang,encompasses 1,285 high-resolution images capturing various temporal and contextual dimensions of rural thoroughfares.Following meticulous image preprocessing,a taxonomy of 40 distinct categories,including 20 specific instance classifications,was established.Using the CVAT annotation tool,we meticulously performed pixel-level manual annotations,yielding a comprehensive dataset of 10,062 annotated instances.Rigorous quality assurance measures implemented to ensure the accuracy,completeness,and consistency of the annotated data.Subsequently,comparative analyses against prevailing road datasets were conducted,alongside training and validation using three seminal instance segmentation models,complemented by a battery of evaluative metrics.Featuring high-definition rural roadway images,corresponding binary mask images,and annotation files adhering to the MS COCO format,this dataset serves as a cornerstone for advancing research endeavors in rural road instance segmentation and related deep learning undertakings.

关 键 词：乡村道路 图像数据集 深度学习 实例分割 

分 类 号：U495[交通运输工程—交通运输规划与管理] TP391.41[交通运输工程—道路与铁道工程]