带注意力机制的神经网络用于跑道线检测

doi:10.3778/j.issn.1002-8331.2112-0067

摘要/Abstract

摘要： 为了提高无人机着陆的有效性，设计了一种带有注意力机制的神经网络算法（LineNet）实现无人机跑道线检测。根据无人机着陆场景仿真出样本数据，并利用标注工具对数据进行标注，使用Shuffle Conv模块缓解特征融合计算量的占用问题，并引入空洞空间金字塔池化注意力机制（ASPP-SA），ASPP-SA模块中添加残差的跳跃结构获取更多的图像信息，对LineNet模型的检测结果做形态学处理并结合连通区域约束对跑道线特征点进行分类，对相同类别的特征点通过最小二乘法进行跑道线拟合。经仿真数据验证，设计的方法可以有效地检测和识别出正确的跑道线，其平均检测精度为94.36%，相较于LaneNet算法、SegNet算法分别提高了12.82和9.95个百分点，单帧检测时间17.2 ms，是CNN+Hough变化算法1.5倍左右，可以满足无人机着陆的响应时间需求，在无人机着陆的研究过程中有重要意义。

关键词: 跑道线检测, 连通区域, 最小二乘法, 通道转换卷积, 注意力机制

Abstract: To improve the effectiveness of UAV landing, a neural network algorithm（LineNet） with attention mechanism is designed to realize the detection of UAV runway lines. Firstly, the sample data is simulated according to the landing scene of the drone, and the data is marked with the labeling tool, Secondly, the Shuffle Conv module is used to alleviate the problem of feature fusion computation, and the hole space pyramid pooling attention mechanism（ASPP-SA） is introduced, it adds residual skip structure in ASPP-SA module to obtain more image information. Then, the detection results of the LineNet model are morphologically processed, and the feature points of the runway line are classified according to the constraints of the connected region. Finally, the feature points of the same category are fitted by the least square method. Verified by simulation data, the designed method can effectively detect and identify the correct runway line and its average detection accuracy is 94.36%. Compared with the LaneNet algorithm and the SegNet algorithm, the detection accuracy is improved by 12.82 and 9.95 percentage points. The detection time of a single frame is 17.2 ms, which is about 1.5 times that of the CNN+Hough change algorithm, which can meet the response time requirements of UAV landing and is of great significance in the research process of UAV landing.

Key words: runway line detection, connected area, least square method, Shuffle Conv, attention

程国建, 卞晨亮, 陈琛, 杨倬. 带注意力机制的神经网络用于跑道线检测[J]. 计算机工程与应用, 2023, 59(5): 169-175.

CHENG Guojian, BIAN Chenliang, CHEN Chen, YANG Zhuo. Neural Network with Attention Mechanism for Runway Line Detection[J]. Computer Engineering and Applications, 2023, 59(5): 169-175.

参考文献

[1] ZENG F C，SHI H Q，WANG H.The object recognition and adaptive threshold selection in the vision system for landing an unmanned aerial vehicle[C]//IEEE International Conference on Robotics and Automation，2009：117-122.
[2] LI J，MA X D，CHEN H M，et al.Real-time detection and tracking method of landmark based on UAV visual navigation[J].Journal of Northwestern Polytechnical University，2018，36（2）：294-301.
[3] 刘文婷，卢新明.基于计算机视觉的Transformer研究进展[J].计算机工程与应用，2022，58（6）：1-16.
LIU Wenting，LU Xinming.Research progress of Transformer based on computer vision[J].Computer Engineering and Applications，2022，58（6）：1-16.
[4] 张馨月，降爱莲.融合特征增强和自注意力的SSD小目标检测算法[J].计算机工程与应用，2022，58（5）：247-255.
ZHANG Xinyue，JIANG Ailian.SSD small target detection algorithm combining feature enhancement and self-attention[J].Computer Engineering and Applications，2022，58（5）：247-255.
[5] 饶颖露，邢金昊，张恒，等.基于视觉的无人机板载自主实时精确着陆系统[J].计算机工程，2021，47（10）：290-297.
RAO Yinglu，XING Jinhao，ZHANG Heng，et al.Vision-based autonomous real-time precise landing system for UVA-borne processors[J].Computer Engineering，2021，47（10）：290-297.
[6] 梅立春，王彩云，赵元富，等.无人机自主着陆地标实时检测方法[J].系统工程与电子技术，2019，41（10）：2157-2162.
MEI Lichun，WANG Caiyun，ZHAO Yuanfu，et al.Real-time detection method of landmark in UAV autonomous landing[J].Systems Engineering and Electronics，2019，41（10）：2157-2162.
[7] 赵珊，郑爱玲.判别相关分析双注意力机制的目标检测算法[J].计算机工程与应用，2022，58（17）：120-129.
ZHAO Shan，ZHENG Ailing.Object detection based on dual attention mechanism combined with discriminant correlation analysis[J].Computer Engineering and Applications，2022，58（17）：120-129.
[8] JIN R，OWAIS H M，LIN D，et al.Ellipse proposal and convolutional neural network discriminant for autonomous landing marker detection[J].Journal of Field Robotics，2019，36（1）：6-16.
[9] LIU C，LIU L，HU G，et al.A P3P problem solving algorithm for landing vision navigation[J].Navigation Positioning & Timing，2018，5（1）：64-67.
[10] GOODFELLOW I，BENGIO Y，COURVILLE A.Deep learning[M].Cambridge，MA：MIT Press，2016.
[11] 成昊天，丁荣莉，胡博文，等.基于HLS的红外遥感图像连通域快速提取方法[J].上海航天（中英文），2021，38（4）：144-151.
CHENG Haotian，DING Rongli，HU Bowen，et al.Fast extraction method for connected domain of infrared remote sensing image based on high level synthesis[J].Aerospace Shanghai，2021，38（4）：144-151.
[12] LECUN Y，BOTTOU L，BENGIO Y，et al.Gradient-based learning applied to document recognition[J].Proceedings of the IEEE，1998，86：2278-2324.
[13] KINGMA D，BA J.Adam：a method for stochastic optimization[J].arXiv：1412.6980，2014.
[14] G[O]TTEN F，HAVERMANN M，BRAUN C，et al.Wind-tunnel and CFD investigations of UAV landing gears and turrets-improvements in empirical drag estimation[J].Aerospace Science and Technology，2020，107：106306.
[15] DANG L，TEWOLDE G，ZHANG X，et al.Reduced reso-lution lane detection algorithm[C]//2017 IEEE AFRICON.Washington DC：IEEE Computer Society，2017：1459-1464.
[16] 邓天民，王琳，杨其芝，等.基于改进SegNet算法的车道线检测方法[J].科学技术与工程，2020，20（36）：14988-14993.
DENG Tianmin，WANG Lin，YANG Qizhi，et al.Lane line detection method based on improved SegNet algorithm[J].Science Technology and Engineering，2020，20（36）：14988-14993.
[17] 王帅康.基于深度强化学习的无人机自主降落方法研究[D].哈尔滨：哈尔滨工业大学，2020.
WANG Shuaikang.Research on the autonomous landing method based on deep reinforcement learning[D].Harbin：Harbin Institute of Technology，2020.
[18] 邓弘志.基于深度学习的智能化无人机视觉系统的设计与仿真[D].成都：电子科技大学，2020.
DENG Hongzhi.Design and simulation of intelligent UAV vision system based on deep learning[D].Chengdu：University of Electronic Scienceand Technology of China，2020.