无人机AI智能识别路面养护车数据

针对城市道路清洁作业，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，能快速识别作业中的环卫车辆。给环卫企业智能化管理各种环卫车辆提供有力的数据支撑。可广泛应用于：
1、作业车辆轨迹监督
通过搭载的 AI 图像识别模块，无人机可快速区分洒水车、压缩车、三轮清运车、快保车、厨余垃圾清运车等各类环卫作业车辆，并实时比对其预设作业路线与实际轨迹，自动标记并推送至指挥中心，结合 GPS 位置数据生成辅助考核管理。
2、车辆故障状况判别
当环卫车辆（如路面养护车、新能源机扫车、压缩车等）发生故障，自动定位并向指挥中心发送精确坐标，联系维修人员赶赴现场尽快修理。
3、违规车辆监管
对作业时段外运行、未按规划路线行驶、违规停靠等行为（常见于洒水车、机扫车、压缩车等），无人机即时上传车辆编号与位置至指挥中心，自动生成整改工单。
4、作业时长统计
系统可有效统计各类车辆（如雾炮车、厨余垃圾清运车等）的作业时长、覆盖面积，判定有效作业时段，排除空驶、停靠等非作业状态。生成量化考核报表，分析不同车型在各路段的作业效率。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（路面养护车），记录边界框及属性，后映射为一级标签（车辆）。3、算法基于YOLOv8m，集成P2层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0.45，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0.45的检测框视为正样本（路面养护车），小于0.45的检测框视为背景负样本。

For urban road cleaning operations, UAVs leverage their low-to-medium altitude advantage (70 to 100 meters) combined with high-resolution sensors and AI algorithms to rapidly identify sanitation vehicles in operation, providing robust data support for sanitation enterprises to intelligently manage their fleet of sanitation vehicles. It can be widely applied in: 1. Operation Vehicle Trajectory Supervision: Equipped with an AI image recognition module, UAVs can quickly distinguish various sanitation operation vehicles including water sprinkler trucks, compactor trucks, three-wheeled garbage collection trucks, quick maintenance vehicles, kitchen waste collection trucks, etc., compare their preset operational routes with actual trajectories in real time, automatically mark and relay the information to the command center, and generate auxiliary assessment and management reports using GPS location data. 2. Vehicle Fault Diagnosis: When sanitation vehicles (such as road maintenance vehicles, new-energy street sweepers, compactor trucks, etc.) malfunction, UAVs can automatically locate the faulty vehicles and send precise coordinates to the command center, notifying maintenance personnel to rush to the site for repairs as soon as possible. 3. Non-compliant Vehicle Oversight: For behaviors such as operating outside scheduled hours, driving without following planned routes, and unauthorized parking (common in water sprinkler trucks, street sweepers, compactor trucks, etc.), UAVs will immediately upload the vehicle number and location to the command center and automatically generate rectification work orders. 4. Operation Duration Statistics: The system can effectively count the operation duration and covered area of various vehicles (such as fog cannon trucks, kitchen waste collection trucks, etc.), identify valid operation periods, and exclude non-operation states such as empty cruising and parking, generate quantitative assessment reports, and analyze the operational efficiency of different vehicle types on each road segment. 1. Data Source: The data is collected from the enterprise's own UAV intelligent inspection system. 2. High-resolution images are collected by UAVs, 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 score, primary label and secondary label. Noise is removed via manual cleaning to ensure data reliability. A two-level annotation scheme is adopted: the initially refined target type is labeled as the secondary label (e.g., road maintenance vehicle), with bounding boxes and corresponding attributes recorded, and then mapped to the primary label (vehicle). 3. The algorithm is based on YOLOv8m, which integrates the P2 layer to optimize small-object detection via image slicing, and combines transfer learning to load pre-trained weights. The pre-trained weights are sourced from the yolov8m.pt model parameters pre-trained on the COCO dataset, which is tailored 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, and the learning rate is set to 0.01 with a batch size of 8. 4. During the inference phase, a confidence threshold of 0.45 is set. The target confidence score is output by the model, reflecting the credibility level of object detection. Only detection results with a confidence score greater than or equal to 0.45 are retained as valid targets; that is, detection boxes with a confidence score ≥0.45 are regarded as positive samples (e.g., road maintenance vehicles), while those with a confidence score <0.45 are classified as background negative samples.

搜集汇总

背景与挑战

背景概述

该数据集由浙江利珉环境科技有限公司登记，包含502条无人机采集的路面养护车图像数据，每日更新，采用xlsx格式，记录图像元信息和AI识别结果。数据集基于YOLOv8m算法实现智能检测，置信度阈值为0.45，主要用于城市道路清洁作业的车辆轨迹监督、故障判别和违规监管，提升环卫管理效率。