深度强化学习在多传感器融合下的无人机视觉感知巡检中的应用

Application of Deep Reinforcement Learning in Drone Visual Perception Inspection Under Multi-sensor Fusion

文中利用深度强化学习,实现了基于视觉感知的无人机全光谱自主巡检功能。在全光谱巡检系统技术部分,详细介绍了多传感器融合的数学模型和公式,并深入探讨了卷积神经网络(CNN)在目标检测中的应用,提出了一种基于特征金字塔网络(FPN)的多尺度目标识别方法。最后,进一步阐述了无人机云台相机的稳定性控制方法,包括PID控制器,并分析了如何将目标检测结果与飞行控制系统相结合,实现自动化的云台相机联动拍摄技术。通过对实验结果的分析,深入讨论了各项技术的优劣势,并提出了改进方向,为未来实现无人机自主巡检提供了参考,为实现高效、精准的工程巡检提供了新的技术路径。

In this paper,the drone full-sPectrum autonomous inspection function based on visual perception is realized by using deep reinforcement learning.In the technical part of the full-spectrum inspection system,the mathermatical model and formula of multi-sensor fusion are introduced in detail,and the application of convolutional neural networks(CNN)in object detection is deeply discussed.A multiscale target recognition method based on feature Pyramid network(FPN)is proposed.Finally,the stability control method of drone pan-tilt camera,including PID controller,is further elaborated,and how to combine the object detection results with the flight control system to realize the automatic pan-tilt camera linkage shooting technology is analyzed.Through the analysis of the experimental results,the advantages and disadvantages of each technology are discussed in depth,and the improvement direction is proposed,which provides a reference for the realization of drone autonomous inspection in the fu-ture,and provides a new technical path for the realization of eficient and accurate engineering inspection.