基于改进卷积神经网络的无人机异常飞行检测  

UAV abnormal flight detection based on improved convolutional neural network

作　　者：宋煜[1] 黄祥 张欣[1] 王海楠 SONG Yu;HUANG Xiang;ZHANG Xin;WANG Hai-nan(Jiangsu Frontier Electric Technology Co.,Ltd.,Nanjing 211102,China)

机构地区：[1]江苏方天电力技术有限公司,南京211102

出　　处：《信息技术》2023年第4期51-57,62,共8页Information Technology

基　　金：江苏方天电力技术有限公司科技项目资助(YF202012)。

摘　　要：在无人机异常飞行姿态检测过程中,受到姿态识别模型的影响,导致算法的时间复杂度较高。因此,提出了基于改进卷积神经网络的轻小型无人机异常飞行姿态检测算法。通过无人机姿态坐标的转换,获取图像采集位置。针对采集图像进行处理和分割,提高了图像特征采集精度。基于改进卷积神经网络构建识别模型,完成飞行姿态快速识别。最后,运用高斯混合模型聚类方法建立异常姿态判别规则,实现无人机异常飞行姿态检测。实验结果证明,比较两种对比方法,文中设计的检测算法分时间复杂度降低了48.27%和67.81%。In the process of UAV abnormal flight attitude detection,the time complexity of the algorithm is high because of the influence of attitude recognition model.Therefore,an improved convolution neural network based abnormal attitude detection algorithm for light and small UAV is proposed.Through the transformation of UAV attitude coordinates,the image acquisition position is obtained.The processing and segmentation of the acquired image could improve the accuracy of image feature acquisition.Based on the improved convolution neural network,the recognition model is constructed to realize the fast recognition of flight attitude.Finally,the Gaussian hybrid model clustering method is used to establish the abnormal attitude discrimination rules to realize the abnormal attitude detection.Experiment results show that the proposed algorithm reduces the time complexity by 48.27%and 67.81%after comparing with the two contrast methods.

关 键 词：改进卷积神经网络 飞行姿态 图像分割 无人机监管模块 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]