无人机智能识别巡查路面磨损箭头数据

应用场景：针对城市道路、河道、城郊公路等，无人机凭借 70 至 100 米的中低空优势，结合高分辨率传感器与 AI 算法，可高效识别多种异常情况，包括：道路标线问题（如模糊车道线、褪色斑马线、磨损箭头）、各类垃圾与堆积物（如路面泥沙）、典型道路病害（如裂缝、坑洞）以及水域异常（如水面垃圾、违规垂钓、浮游植物、水体不洁）。算法通过分析图像的纹理、边缘特征，能够精准识别上述目标并记录坐标。可高效覆盖长距离路段，弥补人工巡检效率低、存在视野盲区的短板。无人机巡检广泛应用于：
1、城市道路日常巡检：发现道路垃圾、泥沙堆积等，立即记录坐标并同步至环卫调度平台，引导清洁车优先清理，避免垃圾散落或泥沙堆积影响通行。
2、河道水域智能监测：精确识别水面垃圾、水体不洁、浮游植物、违规垂钓及岸边各类垃圾。实时回传位置信息至河道管理处，协助安排打捞船定点作业，防止垃圾堵塞水闸或腐烂污染水质。
3、道路公路养护巡查：精确检测路面问题，如模糊车道线、褪色斑马线、磨损箭头，并捕捉裂缝、坑洞及少量垃圾。整合垃圾分布、路面标线、病害数据，生成综合报告，为养护部门提供 “边清理垃圾边修复路面”的联动作业依据。1、数据来源: 数据来源于本企业无人机智能巡查系统。2、高分辨率图像通过无人机采集，记录丰富元信息，包括图像标识ID、图像分辨率(pix)、相机型号、记录时间、文件路径、焦距(mm)、经纬度、海拔(m)、边界框组、置信度阈值、置信度、一级标签和二级标签，经人工清洗剔除噪声，确保数据可靠性。数据采用两级标签：初始细化目标类型标为二级标签（磨损箭头），记录边界框及属性，后映射为一级标签（路面标识不清）。3、算法基于YOLOv8m，集成SAHI通过图像分片优化小目标检测，结合迁移学习加载预训练权重。预训练权重使用在COCO数据集上预训练的模型参数yolov8m.pt，适用于通用目标检测任务。微调时，保留骨干网络低层次特征提取层权重，冻结部分层以防过拟合，调整检测头参数，设置学习率0.01、批量大小8。4、在推理阶段，设定置信度阈值为0.5，目标置信度由模型输出，反映目标检测的可信水平。仅保留高于此阈值的检测结果作为目标，即置信度输出大于等于0.5的检测框视为正样本（磨损箭头），小于0.5的检测框视为背景负样本。

Application Scenarios: For urban roads, rivers, suburban highways and other scenarios, drones leverage their mid-low altitude advantage of 70 to 100 meters, combined with high-resolution sensors and AI algorithms, to efficiently identify various abnormal conditions, including: road marking issues (such as blurred lane lines, faded zebra crossings, worn arrows), various garbage and accumulations (such as sediment on the road surface), typical road diseases (such as cracks, potholes), and water area abnormalities (such as floating garbage on the water surface, illegal fishing, phytoplankton, unclean water bodies). By analyzing the texture and edge features of images, the algorithm can accurately identify the above targets and record their coordinates. It can efficiently cover long-distance road sections, making up for the shortcomings of manual inspection with low efficiency and blind visual areas. UAV inspection is widely used in:
1. Daily inspection of urban roads: Detect road garbage, sediment accumulation, etc., immediately record the coordinates and synchronize them to the sanitation scheduling platform, guiding cleaning vehicles to clean first, so as to avoid garbage scattering or sediment accumulation affecting traffic.
2. Intelligent monitoring of river water areas: Accurately identify floating garbage on the water surface, unclean water bodies, phytoplankton, illegal fishing and various garbage on the banks. Real-time return of location information to the river management department, assisting in arranging salvage ships to conduct fixed-point operations, preventing garbage from blocking sluices or rotting and polluting water quality.
3. Road and highway maintenance inspection: Accurately detect road surface problems such as blurred lane lines, faded zebra crossings, worn arrows, and capture cracks, potholes and a small amount of garbage. Integrate garbage distribution, road marking and disease data to generate a comprehensive report, providing a basis for collaborative operations of "cleaning garbage while repairing road surfaces" for maintenance departments.
1. Data Source: The data comes from the enterprise's UAV intelligent inspection system.
2. High-resolution images are collected by drones, with rich metadata recorded, including image ID, image resolution (pix), camera model, recording time, file path, focal length (mm), longitude and latitude, altitude (m), bounding box group, confidence threshold, confidence, primary label and secondary label. Noise is eliminated through manual cleaning to ensure data reliability. A two-level labeling system is adopted: the initially refined target type is marked as a secondary label (e.g., worn arrows), with bounding boxes and attributes recorded, and then mapped to the primary label (e.g., unclear road markings).
3. The algorithm is based on YOLOv8m, integrated with SAHI to optimize small target detection via image slicing, and combined with transfer learning to load pre-trained weights. The pre-trained weights use the model parameters yolov8m.pt pre-trained on the COCO dataset, which is suitable for general object detection tasks. During fine-tuning, the weights of the low-level feature extraction layers of the backbone network are retained, some layers are frozen to prevent overfitting, the detection head parameters are adjusted, with a learning rate of 0.01 and a batch size of 8.
4. In the inference stage, the confidence threshold is set to 0.5. The target confidence is output by the model, reflecting the credibility level of object detection. Only detection results higher than this threshold are retained as valid targets: detection boxes with confidence output greater than or equal to 0.5 are regarded as positive samples (e.g., worn arrows), while those with confidence less than 0.5 are classified as background negative samples.

搜集汇总

背景与挑战

背景概述

该数据集聚焦于无人机智能识别巡查路面磨损箭头数据，包含502条结构化数据，每日更新，采用xlsx格式，涵盖图像标识、经纬度、标签等13个字段。其特点在于利用YOLOv8m算法和SAHI技术优化小目标检测，高效识别道路磨损箭头等路面标识问题，适用于城市道路巡检和养护管理，提升人工巡检效率。