基于国产嵌入式智能计算平台的无人机检测方法

UAV Detection Method Based on Domestic Embedded Intelligent Computing Platform

面向陆地战场上对反无人机侦察的现实需求,提出一种基于国产嵌入式智能计算平台的无人机检测方法。针对无人机体型小、易受战场环境影响而不易察觉的难题,采用红外、可见光图像和视频流等多源输入进行目标检测;针对嵌入式平台算力和存储能力有限的特性,构建轻量化深度神经网络,通过将单次多盒检测器(SSD)中的特征提取网络替换为MobileNet进行模型优化;选用国产嵌入式平台比特大陆SE5智能计算盒进行验证,完成模型转换和移植。实验结果表明:所提基于轻量化深度神经网络MobileNet-SSD的无人机检测方法在国产嵌入式智能计算平台上能够准确判断出目标的类别,且平均识别精度和帧率与在开发环境中运行差距不大。充分表明所提方法在国产嵌入式智能计算平台上进行移植后,能够在速度和精度方面满足应用环境对无人机检测算法实时性与准确性的要求。

A UAV detection method based on a domestic embedded intelligent computing platform is proposed to meet the actual requirements of anti-unmanned aerial vehicle(UAV) reconnaissance on the land battlefield.For the problem that UAV is small in size and not easy to be detected in the battlefield environment,the detection method is to use infrared and visible light images and video streams inputs for target detection.For the limited computing power and storage capacity of embedded platform,a lightweight deep neural network is built,and the feature extraction network in single shot multi-box detector(SSD) is replaced with MobileNet for model compression.The embedded platform Bitmain SE5 intelligent computing box is selected for verification,and the model conversion and transplantation are achieved.The experimental result shows that the proposed UAV detection method based on the lightweight deep neural network MobileNet-SSD can accurately determine the type of targets on the embedded intelligent computing platform,and the mean recognition accuracy and frame rate are basically same with those running in the development environment.It fully shows that the detection method can meet the requirements of the real-time and accuracy of UAV detection algorithm in the application environment in terms of speed and accuracy after being transplanted on the embedded intelligent computing platform.