應用卷積神經網路進行雷達自動目標識別;Applying Convolutional Neural Network to Automatic Recognition of Radar Targets

NCUIR > College of Electrical Engineering & Computer Science > Software Engineer > Electronic Thesis & Dissertation > Item 987654321/80952

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/80952

Title:	應用卷積神經網路進行雷達自動目標識別;Applying Convolutional Neural Network to Automatic Recognition of Radar Targets
Authors:	林宛諭;Lin, Wan-Yu
Contributors:	軟體工程研究所
Keywords:	人工智慧;卷積神經網路;深度學習;高解析度距離輪廓圖;Artificial Intelligence;Convolution Neural Network;Deep Learning;HRRP
Date:	2019-07-24
Issue Date:	2019-09-03 15:21:18 (UTC+8)
Publisher:	國立中央大學
Abstract:	隨著雷達設備及訊號接收器的發展，微波信號在各種目標物的識別應用越來越廣泛，例如在海上環境中，對於船艦的識別有著諸多用途，包括：海岸警戒控制、海上救護、航道管理和軍事作戰。一般而言，在使用微波信號描述船艦的特徵方面，多使用衛星圖、紅外線圖、光學圖、合成孔徑雷達（Synthetic Aperture Radar, SAR）圖及逆合成孔徑雷達（Inverse Synthetic Aperture Radar, ISAR）圖作為辨識的輸入數據。相較於前敘的影像類型，高解析度距離輪廓圖（High Resolution Range Profile, HRRP）更能夠凸顯目標物的幾何特徵，所以不同於使用模擬數據的研究，本論文使用真實高解析度距離輪廓圖做為船艦辨識的輸入數據，並以卷積神經網路作為辨識的核心模型。除了原始資料之高解析度距離輪廓圖，也使用二維灰階圖，另外由於高解析度距離輪廓圖在視覺上可以大致看出各分類間之區別，因此提出具視覺特性之二值圖，並使用經常應用於圖像識別之卷積神經網路進行分類，所以本論文除了使用一維卷積神經網路應用於辨識高解析度距離輪廓圖，並使用灰階圖以及二值圖做為二維卷積神經網路的輸入，針對訓練好的卷積神經網路模型進行測試，並計算單次預測單筆資料及多筆資料各自所需要的時間。實驗結果證明，卷積神經網路在雷達信號辨識上是可行的，能有效地減輕對雜波及目標方位面的敏感性，並能提升微波信號目標辨識能力，達到成功分類目標的目的，且在所有實驗中，測試集準確率皆可高達97%以上。 ;With the progress of radar devices and signal receiver, radio frequency has been used widely for target recognition applications such as military and civilian ship identification for coastal warnings, sea rescue, navigation management and naval warfare. Generally, satellite imagery, infrared image, optical image, Synthetic aperture radar (SAR) and Inverse synthetic aperture radar (ISAR) are usually used to extract signal features. Alternatively, the high-resolution range profile emphasizes the geometric feature of the target, and it is considered a one-dimensional signature of an object, in this paper, the real high-resolution range profiles are used as input data. Furthermore, Convolution Neural Network (CNN) model is adopted as the primary method for ship recognition. Besides the original input one-dimensional high-resolution range profiles, two-dimensional Grayscale images are also used in this paper. Additionally, the high-resolution range profile can be roughly recognized by eyes. Hence, the high-resolution range profiles are also converted into a histogram in order to enhance the contrast of different signal and fed to the established CNN for classification as well. Experimental results exhibit that the CNN model achieves high performance in radar recognition. Furthermore, the CNN model can effectively reduce noise and aspect sensitivity of high-resolution range profiles, and also improves the ability of radio frequency recognition to achieve the goal of classification successfully. In all the experiments, accuracy for test case can reach more than 97%.
Appears in Collections:	[Software Engineer] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	255	View/Open

社群 sharing

Loading...