作者：范玉莹1, 何赟泽1, 孙高森1, 王洪金1, 刘昊2, 李杰2
作者单位：1. 湖南大学，湖南 长沙 410082;
2. 中南勘测设计研究院有限公司，湖南 长沙 410014
关键词：无人机;风力发电机;改进YOLOv5;CBAM;检测跟踪;卡尔曼滤波
摘要：
无人机自主巡检风力发电机的首要步骤是识别和定位风机对象，对目标巡检持续跟踪。无人机运动过程中涉及到风机对象的实时检测与跟踪。该文在无人机高空捕获图像的限制条件下，选择风机轮毂作为特征组件，研究基于改进YOLO方法与卡尔曼滤波器的风机轮毂检测跟踪方法。基于YOLOv5深度卷积神经网络算法自主提取风机轮毂图像特征，设计融合通道和空间信息的CBAM注意力机制，提高目标特征关注度，改进YOLOv5的Backbone主干网络，提高了检测效率和精度，通过建立轮毂数据集，训练回归网络得到风机轮毂检测模型，算法精度达99.5%，推理速度达212 F/s。针对运动目标无人机设计卡尔曼滤波器实现对定位结果的连续稳定跟踪，将主流卷积神经网络与经典滤波器算法结合，得到稳定高效的特征跟踪结果。最后，在实际风场环境中利用无人机与云台相机拍摄进行实验，验证了该方案的可行性与优越性。
Wind turbine hub center detection and tracking based on UAV images
FAN Yuying1, HE Yunze1, SUN Gaosen1, WANG Hongjin1, LIU Hao2, LI Jie2
1. Hunan University, Changsha 410082, China;
2. Zhong Nan Engineering Corporation Limited, Changsha 410014, China
Abstract: The first step for UAV to inspect the wind turbine independently is to identify and locate the wind turbine object and continuously track the target. The motion process of UAV involves the real-time detection and tracking of fan object. In this paper, we select the wind turbine hub as the feature component and study the wind turbine hub detection and tracking method based on the improved YOLO method with Kalman filter under the limitation of UAV overhead capture images. Based on YOLOv5 deep convolutional neural network algorithm to extract wind turbine hub image features autonomously, design CBAM attention mechanism that fuses channel and spatial information to improve target feature attention, improve Backbone network of YOLOv5 to improve detection efficiency and accuracy, the fan hub detection model is obtained by establishing wheel hub data set and training regression network, the algorithm accuracy reaches 99.5% and inference speed up to 212 F/s. Design Kalman filter for motion target UAV to achieve continuous and stable tracking of localization results and combine mainstream convolutional neural network with classical filter algorithm to obtain stable and efficient feature tracking results. Finally, experiments are conducted in a real wind field environment using UAV and gimbal camera shots to verify the feasibility and superiority of the scheme.
Keywords: UAV;wind turbine;improved YOLOv5;CBAM;detection and trace;Kalman filter
2022, 48(7):90-96 收稿日期: 2021-12-20;收到修改稿日期: 2022-02-06
基金项目:
作者简介: 范玉莹（1996-），河南驻马店市人，硕士研究生，专业方向为机器视觉、无人机视觉控制
参考文献
[1] ARAAR O, AOUF N. Visual servoing of a Quadrotor UAV for autonomous power lines inspection[C]//22nd Mediterranean Conference on Control and Automation. IEEE, 2014: 1418-1424.
[2] GU W, HU D, CHENG L, et al. Autonomous Wind Turbine Inspection using a Quadrotor[C]//2020 International Conference on Unmanned Aircraft Systems (ICUAS). 2020: 709-715.
[3] CAR M, MARKOVIC L, IVANOVIC A, et al. Autonomous Wind-Turbine Blade Inspection Using LiDAR-Equipped Unmanned Aerial Vehicle[J]. Access, 2020, 8: 1-1
[4] RAO Y, XIANG B J, HUANG B, et al. Wind Turbine Blade Inspection Based on Unmanned Aerial Vehicle(UAV) Visual Systems[C]//2019 IEEE 3rd Conference on Energy Internet and Energy System Integration (EI2). IEEE, 2019: 708-713.
[5] CAR M, MARKOVIC L, IVANOVIC A, et al. Autonomous Wind-Turbine Blade Inspection Using LiDAR-Equipped Unmanned Aerial Vehicle[J]. IEEE Access, 2020, 8: 131380-131387
[6] FAN T. Research and realization of video target detection system based on deep learning[J]. International Journal of Wavelets, Multiresolution and Information Processing, 2020, 18(01): 280-292
[7] GIRSHICK R. Fast R-CNN[C]//2015 IEEE International Conference on Computer Vision (ICCV), 2015: 1440-1448.
[8] HE K, ZHANG X, REN S, et al. Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition[J]. IEEE transactions on pattern analysis and machine intelligence, 2015, 37(9): 1904-1916
[9] REN S, HE K, GIRSHICK R, et al. Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2017, 39(6): 1137-1149
[10] WU Z, CHEN X, GAO Y, et al. Rapid Target Detection in High Resolution Remote Sensing Images Using YOLO Model[J]. International archives of the photogrammetry, remote sensing and spatial information sciences, 2018, XLII-3: 1915-1920
[11] 高继东, 焦鑫, 刘全周, 等. 机器视觉与毫米波雷达信息融合的车辆检测技术[J]. 中国测试, 2021, 47(10): 33-40
[12] WOO S, PARK J, LEE J-Y, et al. CBAM: Convolutional Block Attention Module[C]//European Conference on Computer Vision. Cham: Springer International Publishing, 2018: 3-19.
[13] 吴军, 石改琴, 卢帅员, 等. 采用无人机视觉的飞机蒙皮损伤智能检测方法[J]. 中国测试, 2021, 47(11): 119-126
[14] 李青青, 李陈斌, 李镇宇, 等. 基于改进YOLOv5的叉车能效优化方法研究[J]. 现代计算机, 2021(21): 64-71
