基于高效可擴(kuò)展改進(jìn)殘差結(jié)構(gòu)神經(jīng)網(wǎng)絡(luò)的艦船目標(biāo)識別技術(shù)

付哲泉; 李尚生; 李相平; 但波; 王旭坤

doi:10.11999/JEIT190913

基于高效可擴(kuò)展改進(jìn)殘差結(jié)構(gòu)神經(jīng)網(wǎng)絡(luò)的艦船目標(biāo)識別技術(shù)

doi: 10.11999/JEIT190913

海軍航空大學(xué) 煙臺(tái) 264001

詳細(xì)信息

作者簡介:
付哲泉：男，1992年生，博士生，研究方向?yàn)榫_制導(dǎo)技術(shù)及其智能化

李尚生：男，1965年生，教授，研究方向?yàn)閷?dǎo)彈制導(dǎo)技術(shù)

李相平：男，1963年生，教授，研究方向?yàn)榫_制導(dǎo)和目標(biāo)探測技術(shù)

但波：男，1985年生，講師，研究方向?yàn)槟繕?biāo)識別與選擇技術(shù)

王旭坤：男，1995年生，碩士生，研究方向?yàn)槔走_(dá)目標(biāo)識別技術(shù)

通訊作者:
付哲泉　fuzq2413@163.com

中圖分類號: TN957.51
計(jì)量
- 文章訪問數(shù): 5264
- HTML全文瀏覽量: 1500
- PDF下載量: 136
- 被引次數(shù): 0
出版歷程
- 收稿日期: 2019-11-14
- 修回日期: 2020-04-16
- 網(wǎng)絡(luò)出版日期: 2020-04-25
- 刊出日期: 2020-12-08

Ship Target Recognition Based on Highly Efficient Scalable Improved Residual Structure Neural Network

Naval Aviation University, Yantai 264001, China

摘要

摘要: 神經(jīng)網(wǎng)絡(luò)的深度在一定范圍內(nèi)與識別效果成正相關(guān)，為解決超出范圍后網(wǎng)絡(luò)層數(shù)增加識別準(zhǔn)確率卻下降的模型飽和問題，該文提出一種具有高效的微塊內(nèi)部結(jié)構(gòu)和殘差網(wǎng)絡(luò)結(jié)構(gòu)的神經(jīng)網(wǎng)絡(luò)模型，用于對艦船目標(biāo)基于高分辨距離像的分類識別。該方法利用具有小尺度卷積核的卷積模塊提取目標(biāo)的穩(wěn)定可分特征，同時(shí)利用聯(lián)合損失函數(shù)約束目標(biāo)特征的類內(nèi)距離提高識別能力。仿真結(jié)果表明，該模型相比于其他常見網(wǎng)絡(luò)結(jié)構(gòu)，在模型參數(shù)更少的情況下，識別效果更好，同時(shí)具有較強(qiáng)的噪聲魯棒性。
- 目標(biāo)識別 /
- 高分辨距離像 /
- 神經(jīng)網(wǎng)絡(luò) /
- 殘差結(jié)構(gòu)
Abstract: The depth of neural network is positively correlated with the recognition effect in a certain range. In order to solve the problem that model recognition accuracy decreases when the number of network layers increases after exceeding the range. A neural network model with efficient micro internal blocks structure and residual network structure is proposed, which is used for recognition of ship targets based on High Range Resolution Profile (HRRP) data. In this method, the convolution module with a small scale convolution kernel is used to extract automatically the stable and separable features of target. And the intra-class distance of the target is constrained by the joint loss function to improve the recognition ability. Simulation results show that compared with other common network structures, this model has better recognition performance and stronger noise robustness with fewer model parameters.
- Target recognition /
- High Range Resolution Profile (HRRP) /
- Neural Network (NN) /
- Residual Structure

HTML全文

圖 1 針對HRRP的CNN結(jié)構(gòu)示意圖

下載: 全尺寸圖片幻燈片

圖 2 殘差結(jié)構(gòu)

下載: 全尺寸圖片幻燈片

圖 3 卷積模塊結(jié)構(gòu)

下載: 全尺寸圖片幻燈片

圖 4 本文所提模型框圖

下載: 全尺寸圖片幻燈片

圖 5 某艘艦船模型圖及其對應(yīng)幅值歸一化后的HRRP圖

下載: 全尺寸圖片幻燈片

圖 6 HRRP數(shù)據(jù)平移截取示意圖

下載: 全尺寸圖片幻燈片

圖 7 信噪比為15 dB時(shí)不同模型的特征可視化圖

下載: 全尺寸圖片幻燈片

表 1 模型A各階段參數(shù)情況

階段	輸出	結(jié)構(gòu)		參數(shù)個(gè)數(shù)
初始卷積層	128×1×9	7×1, 9, s=2		99
		左側(cè)支路	右側(cè)支路
卷積模塊1	64×1×18	1×1, 9 3×1, 3, s=2, x=3 1×1, 12	1×1, 15, s=2	585
卷積模塊2	32×1×36	1×1, 18 3×1, 6, s=2, x=3 1×1, 24	1×1, 30, s=2	1980
卷積模塊3	16×1×72	1×1, 36 3×1, 12, s=2, x=3 1×1, 48	1×1, 60, s=2	7200
卷積模塊4	8×1×144	1×1, 72 3×1, 24, s=2, x=3 1×1, 96	1×1, 120, s=2	27360
全連接層1	144	全局平均池化+全局最大值池化		0
全連接層2	2			288
輸出層	13	SL+CL		26
	參數(shù)總數(shù)		37538

下載: 導(dǎo)出CSV

表 2 不同復(fù)雜度模型在不同信噪比數(shù)據(jù)集下的識別準(zhǔn)確率(%)

模型名稱	識別時(shí)間(μs)	信噪比(dB)
模型名稱	識別時(shí)間(μs)	0	5	10	15
模型A	258	60.42	89.41	98.21	99.83
模型B	326	72.95	94.41	99.15	99.89
模型C	323	73.78	93.71	99.07	99.86

下載: 導(dǎo)出CSV

表 3 CNN模型結(jié)構(gòu)和參數(shù)明細(xì)

階段	輸出維度	網(wǎng)絡(luò)結(jié)構(gòu)	參數(shù)個(gè)數(shù)
卷積層1	256×1×8	3×1, 8, s=1	64
池化層1	128×1×8	2×1, s=2	0
卷積層2	128×1×16	3×1, 16, s=1	464
池化層2	64×1×16	2×1, s=2	0
卷積層3	64×1×32	3×1, 32, s=1	1696
池化層3	32×1×32	2×1, s=2	0
卷積層4	32×1×64	3×1, 64, s=1	6464
池化層4	16×1×64	2×1, s=2	0
卷積層5	16×1×64	1×1, 64, s=1	4416
池化層5	8×1×64	2×1, s=2	0
全連接層1	64		32832
全連接層2	2		130
輸出層	13	SL	39
參數(shù)總數(shù)		46105

下載: 導(dǎo)出CSV

表 4 SDSAE&KNN模型結(jié)構(gòu)和參數(shù)明細(xì)

階段	輸出維度	參數(shù)個(gè)數(shù)
隱藏層1	150×1	38550
隱藏層2	100×1	15100
隱藏層3	50×1	5050
隱藏層4	10×1	510
參數(shù)總數(shù)	59210

下載: 導(dǎo)出CSV

表 5 SCAE模型結(jié)構(gòu)和參數(shù)明細(xì)

階段	輸出維度	網(wǎng)絡(luò)結(jié)構(gòu)	參數(shù)個(gè)數(shù)
卷積層1	256×1×128	5×1, 128, s=1	768
池化層1	128×1×128	2×1, s=2	0
卷積層2	128×1×64	5×1, 64, s=1	41024
池化層2	64×1×64	2×1, s=2	0
卷積層3	64×1×32	3×1, 32, s=1	6176
池化層3	32×1×32	2×1, s=2	0
卷積層4	32×1×16	3×1, 16, s=1	1552
池化層4	16×1×16	2×1, s=2	0
卷積層5	16×1×8	1×1, 8, s=1	136
池化層5	8×1×8	2×1, s=2	0
輸出層	13	SL	845
參數(shù)總數(shù)		50501

下載: 導(dǎo)出CSV

表 6 不同信噪比條件下本節(jié)模型與對比模型識別準(zhǔn)確率(%)

模型名稱	識別時(shí)間(μs)	信噪比(dB)
模型名稱	識別時(shí)間(μs)	0	5	10	15
模型A	258	60.42	89.41	98.21	99.83
CNN	69	58.22	86.91	95.51	98.79
SCAE	47	54.78	86.58	94.44	98.78
SDSAE&KNN	68	46.50	83.94	93.44	98.65

下載: 導(dǎo)出CSV

參考文獻(xiàn)(18)

魏存?zhèn)? 段發(fā)階, 劉先康. 基于寬帶雷達(dá)HRRP艦船目標(biāo)長度估計(jì)算法[J]. 系統(tǒng)工程與電子技術(shù), 2018, 40(9): 1960–1965. doi: 10.3969/j.issn.1001-506X.2018.09.10

WEI Cunwei, DUAN Fajie, and LIU Xiankang. Length estimation method of ship target based on wide-band radar’s HRRP[J]. Systems Engineering and Electronics, 2018, 40(9): 1960–1965. doi: 10.3969/j.issn.1001-506X.2018.09.10

賀思三, 趙會(huì)寧, 張永順. 基于時(shí)頻域聯(lián)合濾波的中段群目標(biāo)信號分離[J]. 雷達(dá)學(xué)報(bào), 2015, 4(5): 545–551. doi: 10.12000/JR15008

HE Sisan, ZHAO Huining, and ZHANG Yongshun. Signal separation for target group in midcourse based on time-frequency filtering[J]. Journal of Radars, 2015, 4(5): 545–551. doi: 10.12000/JR15008

吳佳妮, 陳永光, 代大海, 等. 基于快速密度搜索聚類算法的極化HRRP分類方法[J]. 電子與信息學(xué)報(bào), 2016, 38(10): 2461–2467. doi: 10.11999/JEIT151457

WU Jiani, CHEN Yongguang, DAI Dahai, et al. Target recognition for polarimetric HRRP based on fast density search clustering method[J]. Journal of Electronics &Information Technology, 2016, 38(10): 2461–2467. doi: 10.11999/JEIT151457

李建偉, 曲長文, 彭書娟, 等. 基于生成對抗網(wǎng)絡(luò)和線上難例挖掘的SAR圖像艦船目標(biāo)檢測[J]. 電子與信息學(xué)報(bào), 2019, 41(1): 143–149. doi: 10.11999/JEIT180050

LI Jianwei, QU Changwen, PENG Shujuan, et al. Ship detection in SAR images based on generative adversarial network and online hard examples mining[J]. Journal of Electronics &Information Technology, 2019, 41(1): 143–149. doi: 10.11999/JEIT180050

杜蘭, 魏迪, 李璐, 等. 基于半監(jiān)督學(xué)習(xí)的SAR目標(biāo)檢測網(wǎng)絡(luò)[J]. 電子與信息學(xué)報(bào), 2020, 42(1): 154–163. doi: 10.11999/JEIT190783

DU Lan, WEI Di, LI Lu, et al. SAR target detection network via semi-supervised learning[J]. Journal of Electronics &Information Technology, 2020, 42(1): 154–163. doi: 10.11999/JEIT190783

羅會(huì)蘭, 盧飛, 孔繁勝. 基于區(qū)域與深度殘差網(wǎng)絡(luò)的圖像語義分割[J]. 電子與信息學(xué)報(bào), 2019, 41(11): 2777–2786. doi: 10.11999/JEIT190056

LUO Huilan, LU Fei, and KONG Fansheng. Image semantic segmentation based on region and deep residual network[J]. Journal of Electronics &Information Technology, 2019, 41(11): 2777–2786. doi: 10.11999/JEIT190056

XING Shihong and ZHANG Shaokang. Ship model recognition based on convolutional neural networks[C]. 2018 IEEE International Conference on Mechatronics and Automation, Changchun, China, 2018: 144-148. doi: 10.1109/ICMA.2018.8484362.

楊宏宇, 王峰巖. 基于深度卷積神經(jīng)網(wǎng)絡(luò)的氣象雷達(dá)噪聲圖像語義分割方法[J]. 電子與信息學(xué)報(bào), 2019, 41(10): 2373–2381. doi: 10.11999/JEIT190098

YANG Hongyu and WANG Fengyan. Meteorological radar noise image semantic segmentation method based on deep convolutional neural network[J]. Journal of Electronics &Information Technology, 2019, 41(10): 2373–2381. doi: 10.11999/JEIT190098

王鑫, 李可, 寧晨, 等. 基于深度卷積神經(jīng)網(wǎng)絡(luò)和多核學(xué)習(xí)的遙感圖像分類方法[J]. 電子與信息學(xué)報(bào), 2019, 41(5): 1098–1105. doi: 10.11999/JEIT180628

WANG Xin, LI Ke, NING Chen, et al. Remote sensing image classification method based on deep convolution neural network and multi-kernel learning[J]. Journal of Electronics &Information Technology, 2019, 41(5): 1098–1105. doi: 10.11999/JEIT180628

郭晨, 簡濤, 徐從安, 等. 基于深度多尺度一維卷積神經(jīng)網(wǎng)絡(luò)的雷達(dá)艦船目標(biāo)識別[J]. 電子與信息學(xué)報(bào), 2019, 41(6): 1302–1309. doi: 10.11999/JEIT180677

GUO Chen, JIAN Tao, XU Congan, et al. Radar HRRP target recognition based on deep multi-scale 1D convolutional neural network[J]. Journal of Electronics &Information Technology, 2019, 41(6): 1302–1309. doi: 10.11999/JEIT180677

王容川, 莊志洪, 王宏波, 等. 基于卷積神經(jīng)網(wǎng)絡(luò)的雷達(dá)目標(biāo)HRRP分類識別方法[J]. 現(xiàn)代雷達(dá), 2019, 41(5): 33–38. doi: 10.16592/j.cnki.1004-7859.2019.05.007

WANG Rongchuan, ZHUANG Zhihong, WANG Hongbo, et al. HRRP classification and recognition method of radar target based on convolutional neural network[J]. Modern Radar, 2019, 41(5): 33–38. doi: 10.16592/j.cnki.1004-7859.2019.05.007

劉興旺. 一種多層預(yù)訓(xùn)練卷積神經(jīng)網(wǎng)絡(luò)在圖像識別中的應(yīng)用[D]. [碩士論文], 中南民族大學(xué), 2018.

LIU Xingwang. The application of a multi-layers pre-training convolutional neural network in image recognition[D]. [Master dissertation], South-Central University for Nationalities, 2018.

趙飛翔, 劉永祥, 霍凱. 基于棧式降噪稀疏自動(dòng)編碼器的雷達(dá)目標(biāo)識別方法[J]. 雷達(dá)學(xué)報(bào), 2017, 6(2): 149–156. doi: 10.12000/JR16151

ZHAO Feixiang, LIU Yongxiang, and HUO Kai. Radar target recognition based on stacked denoising sparse autoencoder[J]. Journal of Radars, 2017, 6(2): 149–156. doi: 10.12000/JR16151

KRIZHEVSKY A, SUTSKEVER I, and HINTON G E. ImageNet classification with deep convolutional neural networks[C]. The 25th International Conference on Neural Information Processing Systems, Red Hook, United States, 2012: 1097–1105.

SIMONYAN K and ZISSERMAN A. Very deep convolutional networks for large-scale image recognition[C]. The 3rd International Conference on Learning Representations, San Diego, United States, 2015: 1–14.

HE Kaiming, ZHANG Xiangyu, REN Shaoqing, et al. Deep residual learning for image recognition[C]. 2016 IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, United States, 2016: 770–778. doi: 10.1109/CVPR.2016.90.

SZEGEDY C, IOFFE S, VANHOUCKE V, et al. Inception-v4, inception-ResNet and the impact of residual connections on learning[C]. The 31st AAAI Conference on Artificial Intelligence, San Francisco, United States, 2017: 4278–4284.

WEN Yandong, ZHANG Kaipeng, LI Zhifeng, et al. A discriminative feature learning approach for deep face recognition[C]. The 14th European Conference on Computer Vision, Amsterdam, Netherlands, 2016: 499–515. doi: 10.1007/978-3-319-46478-7_31.

相關(guān)文章

施引文獻(xiàn)

資源附件(0)

訪問統(tǒng)計(jì)