基于区域卷积神经网络的空中飞行物识别算法  

Flying object recognition algorithm based on region CNN

作　　者：刘聪聪 应捷[1] 杨海马[1] 刘瑾[2] 李筠[1] LIU Congcong;YING Jie;YANG Haima;LIU Jin;LI Jun(School of Optical-Electrical and Computer Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China;School of Electronic and Electrical Engineering,Shanghai University ofEngineering Science,Shanghai 201620,China)

机构地区：[1]上海理工大学光电信息与计算机工程学院,上海200093 [2]上海工程技术大学电子电气工程学院,上海201620

出　　处：《传感器与微系统》2021年第1期110-113,117,共5页Transducer and Microsystem Technologies

基　　金：国家自然科学基金天文联合基金资助项目(U1831133);上海市自然科学基金资助项目(17ZR1443500)。

摘　　要：针对飞行物检测中,由于目标背景复杂、姿态不一导致的识别准确率低的问题,提出改进的基于区域卷积神经网络Faster R-CNN的空中飞行物识别算法,使用消色差折射式望远镜ETX80和个人电脑(PC)构建空中飞行物识别系统。首先,构建包含无人机、飞机和飞鸟三种飞行物的数据集,对数据集进行标注和划分;然后,利用ResNet101深度残差网络提取图像特征,并输入构建的深度学习网络模型进行训练,网络模型包括区域建议网络、感兴趣区域池化层和分类层。测试结果证明:该方法能够在不同背景下的无人机、飞机、飞鸟三类空中飞行物的识别上达到良好的效果,检测平均准确率为96.7%,比FasterR-CNN算法提高3.1%。Aiming at the problem that in detection of flying objects,due to the complex background of target and low recognition accuracy caused by different postures,a flying object recognition algorithm based on improved Faster R-CNN is proposed.The recognition algorithm uses the achromatic refracting telescope ETX80 and PC to construct the flying object recognition system.A dataset containing three kinds of flying objects of UAV,airplanes and birds is constructed,and the dataset is marked and divided.ResNet101 deep residual network is used to extract the image features,and input constructed deep learning network model for training.The network model includes the regional proposal network,the ROI pooling layer and the classification layer.The test results show that the method can achieve good results in the identification of three types of airborne objects such as uavs,airplanes and birds in different backgrounds.The mean average precision is 96.7%,which is 3.1%higher than that of Faster R-CNN algorithm.

关 键 词：卷积神经网络 飞行物检测 目标识别 深度学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]