
一种无人机图像识别技术体系研究与应用
赵薛强
一种无人机图像识别技术体系研究与应用
The Study and Application of a UAV Image Recognition Technology System
为了实现水利工程巡检、河湖岸线、河湖环境监测等海量无人机图像异常特征物的智能化检测识别,提高管理效率,满足智慧水利建设的需求,基于YOLO v3算法框架,通过引入注意力模块SE,构建了高精度的YOLO v3-SE目标检测算法,形成了无人机图像识别技术体系,并将其成功应用于多个水利工程的海量无人机图像的异常特征物检测识别中。结果表明:通过自建66 000 张图片数据的训练集和35 514张图片数据的测试集,本算法与原始YOLO v3算法、改进的SKSet-YOLO v3算法和CBAM-YOLO v3算法相比,在积水、塌方、运输船、滑坡、聚集型垃圾和分散型垃圾等6类目标物的检测精度AP均有较大幅度的提升;平均检测精度mAP也分别从59.83%提升至90.17%、从79%提升至90.17%、从 72%提升至90.17%,精度得到明显提升,满足水利工程智慧化监控的需求。
In order to make the intelligent detection and identification of abnormal features of massive UAV images, such as water conservancy project inspection, river and lake shoreline, and the realization of river and lake environmental monitoring, the enhancement of management efficiency, the satisfaction of demands for intelligent water conservancy construction based on YOLO v3 algorithm framework, a high-precision YOLO v3-SE target detection algorithm is constructed by the introduction of the SE, the UAV image recognition technology system is formed and it is also successfully applied to the detection and identification of massive UAV images in multiple water conservancy projects. It is shown based on the results that are compared with the original YOLO v3. SKSet-YOLO v3 and CBAM-YOLO v3 algorithm are improved by a self-built 66 000 picture data training set and a test set of 35 514 picture data, for the algorithm, the detection accuracy of 6 types of targets such as water, landslide, carrier, landslide, aggregate garbage and decentralized garbage has been greatly improved. With regard to the average detection accuracy mAP, it has also increased from 59.83% to 90.17%, from 79% to 90.17%, and from 72% to 90.17%.
YOLO v3-SE / 图像识别 / 通道注意力模块 / 水利工程 / 技术体系 {{custom_keyword}} /
YOLO v3-SE / image identification / channel attention module / hydraulic works / technical system {{custom_keyword}} /
表1 异常目标物标准Tab.1 norm of abnormal target |
类别 | 描述 |
---|---|
g_garbage d_garbage | 聚集型垃圾(坝站设置的拦网或建筑处所形成的聚集型漂浮物) 分散型垃圾(针对河面飘散的不成堆,零散的漂浮物) |
trans_boat | 运输船 |
spoil | 弃渣(主要为施工区域的废弃建筑垃圾) |
stag_water | 积水(施工区域的积水) |
collapse | 塌方(河岸线、道路、护坡等区域的坍塌情况) |
表2 数据集参数表Tab.2 Parameters of Dataset |
类别名 | 图片数 | 目标数 |
---|---|---|
g_garbage | 10 530 | 17 750 |
d_garbage | 12 100 | 12 430 |
trans_boat | 14 050 | 33 910 |
spoil | 18 200 | 12 080 |
stag_water | 29 490 | 45 130 |
collapse | 17 170 | 19 000 |
表3 各算法的识别精度统计表 (%)Tab.3 Statistical table of recognition accuracy of each algorithm |
网络结构 | collapse | stag_water | trans_boat | d_garbage | spoil | g_garbage | mAP |
---|---|---|---|---|---|---|---|
YOLO v3 | 62 | 59 | 61 | 60 | 56 | 61 | 59.83 |
SKNet-YOLOV3 | 81 | 80 | 78 | 76 | 83 | 76 | 79.00 |
CBAM-YOLO v3 | 75 | 74 | 74 | 71 | 70 | 68 | 72.00 |
本文算法 | 95 | 89 | 93 | 91 | 81 | 92 | 90.17 |
1 |
王小刚,赵薛强,王建成.贴近摄影测量在水利工程监测中的应用[J].人民长江,2021,52():130-133.
增刊1
{{custom_citation.content}}
{{custom_citation.annotation}}
|
2 |
崔保春,徐言勋.基于模式识别技术的高光谱遥感图像检测[J].现代电子技术,2019,42(17):58-62.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
3 |
董淑娟,张志纲,丁爱萍.基于视觉图像的水利工程质量检测研究与仿真[J].计算机仿真,2012,29(6):274-277.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
4 |
C]//CVPR 2016: Proceedings of the 2016 IEEEConference on Computer Vision and Pattern Recognition.Washington, DC: IEEE Computer Society, 2016:779-788.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
5 |
杨高坤.单阶段法目标检测技术研究[J].电子世界,2021(3):77-78,81.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
6 |
WILLIAM.一文读懂YOLO v5与YOLO v4 [EB/OL].
{{custom_citation.content}}
{{custom_citation.annotation}}
|
7 |
李维刚,杨潮,蒋林,等.基于改进YOLOv4算法的室内场景目标检测[J/OL].激光与光电子学进展:1-19[2021-08-06].
{{custom_citation.content}}
{{custom_citation.annotation}}
|
8 |
李阿娟. YOLOv5算法改进及其现实应用[D].中北大学,2021.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
9 |
蔡鸿峰,吴观茂.一种基于改进YOLO v3的小目标检测方法[J].湖北理工学院学报,2021,37(2):33-36,47.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
10 |
顾晋,罗素云.基于改进的YOLO v3车辆检测方法[J].农业装备与车辆工程,2021,59(7):98-103.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
11 |
舒壮壮,单梁,马苗苗,等.基于YOLOv3的改进行人检测算法研究[J].南京理工大学学报,2021,45(3):259-264.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
12 |
袁小平,马绪起,刘赛.改进YOLOv3的行人车辆目标检测算法[J].科学技术与工程,2021,21(8):3 192-3 198.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
13 |
王浩,雷印杰,陈浩楠.改进YOLOV3实时交通标志检测算法[J/OL].计算机工程与应用:1-9[2021-08-06].
{{custom_citation.content}}
{{custom_citation.annotation}}
|
14 |
王飞,刘梦婷,刘雪芹,等.基于YOLOv3深度学习的海雾气象条件下海上船只实时检测[J].海洋科学,2020,44(8):197-204.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
15 |
江波,屈若锟,李彦冬,等.基于深度学习的无人机航拍目标检测研究综述[J].航空学报,2021,42(4):137-151.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
16 |
严开忠,马国梁,许立松,等.基于改进YOLOv3的机载平台目标检测算法[J].电光与控制,2021,28(5):70-74.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
17 |
孙奥,金鑫,管相源,等.基于YOLOv3的无人机建筑物空间特征提取方法研究[J/OL].土木建筑工程信息技术:1-7[2021-08-06].
{{custom_citation.content}}
{{custom_citation.annotation}}
|
18 |
邹鑫垚. 基于优化预选区域的二阶段目标检测算法[D].哈尔滨工程大学,2020.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
19 |
郭智超,丛林虎,刘爱东,等.基于SK-YOLOV3的遥感图像目标检测方法[J].兵器装备工程学报,2021,42(7):165-171.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
20 |
孙义博,张文靖,王蓉,等.基于通道注意力机制的行人重识别方法[J/OL].北京航空航天大学学报:1-10[2021-08-06].
{{custom_citation.content}}
{{custom_citation.annotation}}
|
21 |
鞠默然,罗江宁,王仲博,等.融合注意力机制的多尺度目标检测算法[J].光学学报,2020,40(13):132-140.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
{{custom_ref.label}} |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
/
〈 |
|
〉 |