Please wait a minute...
浙江大学学报(工学版)
浙江大学学报(工学版)  2022, Vol. 56 Issue (1): 47-55    DOI: 10.3785/j.issn.1008-973X.2022.01.005
计算机技术、信息与电子工程     
基于线性逻辑矢量模式的遥感图像目标检测
陈雪云(),黄金汉(),胡子灿,岑升才
广西大学 电气工程学院，广西 南宁 530000
Remote sensing image target detection based on linearly logic vector pattern
Xue-yun CHEN(),Jin-han HUANG(),Zi-can HU,Sheng-cai CEN
School of Electrical Engineering, Guangxi University, Nanning 530000, China
 全文: PDF(1386 KB)   HTML
摘要：

针对局部线性模式(LLP)在遥感图像目标检测中存在维度较高及没有考虑相邻像素间联系的缺点，提出线性逻辑矢量模式特征. 通过在2条互相垂直的方向上进行采样，选取出用于阈值比较的像素点，利用逻辑矢量变换的原理进行特征降维. 通过阈值函数构建中心像素与采样点间的联系，提出基于中心的线性逻辑矢量模式特征(LLVP(C))，通过相邻点阈值比较模式联系相邻像素间的信息提出变型（LLVP(A)). 为了糅合中心与相邻这2类特征信息，提出对LLVP(C)和LLVP(A)进行按位的逻辑融合得到新的LLVP. 在遥感图像数据库上进行车辆、树木及建筑物的检测实验表明，LLVP优于其他的改进型LBP特征，表明应用LLVP特征再进行检测能够以较短的训练时长达到高精度及广适应性的双重标准.
关键词： 遥感检测;  逻辑矢量变换;  二值模式;  纹理特征;  逻辑融合    
Abstract:

Linear logic vector pattern features were proposed aiming at the shortcomings of local line pattern (LLP) in target detection of remote sensing images, such as high dimension and no connection between adjacent pixels. The pixels used for threshold comparison were selected by sampling in two mutually perpendicular directions, and the feature dimension was reduced by using the principle of logical vector transformation. A central based linear logic vector pattern (LLVP (C)) was proposed by constructing the relationship between the center pixel and the sampling point through the threshold function. An improved based linear logic vector pattern (LLVP (A)) was proposed by connecting the information between the adjacent pixels through the threshold comparison pattern of the adjacent points. A new LLVP was proposed based on the bit wise logical fusion of LLVP (C) and LLVP (A) in order to combine the center and adjacent feature information. The experiment of vehicle, tree and building detection on remote sensing image database shows that LLVP is obviously better than other improved LBP features. LLVP features can achieve high precision and wide adaptability with short training time.
Key words: remote sensing detection    logical vector transformation    binary mode    textural feature    logic fusion
收稿日期: 2021-02-25 出版日期: 2022-01-05
CLC:  TP 751  
基金资助: 国家自然科学基金资助项目（62061002）
作者简介: 陈雪云（1969—），男，副教授，博士，从事机器学习与模式识别的研究. orcid.org/0000-0001-5276-1707. E-mail： cxy177@163.com
服务　　
把本文推荐给朋友
加入引用管理器
E-mail Alert
作者相关文章　　
陈雪云
黄金汉
胡子灿
岑升才

引用本文:

陈雪云,黄金汉,胡子灿,岑升才. 基于线性逻辑矢量模式的遥感图像目标检测[J]. 浙江大学学报(工学版), 2022, 56(1): 47-55.

Xue-yun CHEN,Jin-han HUANG,Zi-can HU,Sheng-cai CEN. Remote sensing image target detection based on linearly logic vector pattern. Journal of ZheJiang University (Engineering Science), 2022, 56(1): 47-55.

链接本文:

https://www.zjujournals.com/eng/CN/10.3785/j.issn.1008-973X.2022.01.005    或    https://www.zjujournals.com/eng/CN/Y2022/V56/I1/47

图 1  不同逻辑块变换的原理
φ1函数 逻辑类型 φ2函数 逻辑类型
φ1_or 或 φ2_or 或
φ1_and 与 φ2_and 与
φ1_nor 或非 φ2_nor 或非
φ1_nad 与非 φ2_nad 与非
表 1  变换函数φ对应的逻辑类型
图 2  LLVP(C)采样原理
图 3  LLVP(A)采样原理
图 4  逻辑融合的原理
图 5  部分遥感车辆、遥感树木正、负样本
模式 m 模式 m
LBP 5 376 LVP 4368
CLBP 5 376 LLVP(C) 2688
ULBP 1239 LLVP 2688
LLP 2478 ? ?
表 2  各种特征的维数
φ1函数 特征 m 遥感车辆 遥感树木
A/% F1 A/% F1
φ1_or LLVP(C) 672 90.84 0.921 89.31 0.918
φ1_and LLVP(C) 672 91.43 0.928 88.97 0.902
φ1_nor LLVP(C) 672 89.77 0.915 87.51 0.881
φ1_nad LLVP(C) 672 90.05 0.908 88.43 0.894
φ1_or LLVP(A) 672 86.22 0.905 86.71 0.885
φ1_and LLVP(A) 672 87.64 0.911 85.97 0.871
φ1_nor LLVP(A) 672 86.14 0.889 86.36 0.879
φ1_nad LLVP(A) 672 86.17 0.896 86.98 0.884
φ2_or LLVP(C) 2 688 92.31 0.924 91.77 0.914
φ2_and LLVP(C) 2 688 91.91 0.926 90.12 0.907
φ2_or LLVP(A) 2 688 86.42 0.871 87.22 0.861
φ2_and LLVP(A) 2 688 88.21 0.881 87.58 0.872
表 3  不同φ函数作逻辑变换的LLVP(C)、LLVP(A)在BP网络中的测试结果
φ函数融合的LLVP m 遥感车辆 遥感树木
A/% F1 A/% F1
LLVP1, [β_or]
LLVP1, [β_and]
LLVP1, [β_nor]
LLVP1, [β_nad] 		672
672
672
672 		94.03
94.96
93.47
93.88 		0.948
0.957
0.945
0.951 		92.43
92.44
90.51
91.89 		0.938
0.943
0.914
0.925
表 4  不同β函数作逻辑融合的LLVP1在BP网络上的测试结果
φ函数融合的LLVP m 遥感车辆 遥感树木
A/% F1 A/% F1
LLVP2, [β_or]
LLVP2, [β_and]
LLVP2, [β_nor]
LLVP2, [β_nad] 		2688
2688
2688
2688 		95.54
95.81
93.77
95.62 		0.961
0.967
0.943
0.952 		94.79
94.71
93.11
92.97 		0.955
0.945
0.939
0.945
表 5  不同β函数作逻辑融合的LLVP2在BP网络上的测试结果
算法 m ttr/h 遥感车辆 遥感树木
A/% F1 A/% F1
LBP
ULBP
LLP
LVP
YOLOv3
Faster R-CNN
LLVP(本文方法)		 5376
1239
2478
4368
?
?
2688		 0.39
0.12
0.2
0.36
18
22
0.23		 88.54
89.98
91.17
93.11
96.18
96.83
95.81		 0.891
0.901
0.923
0.937
?
?
0.967		 87.87
88.91
90.84
91.78
95.77
96.12
94.79		 0.881
0.887
0.914
0.927
?
?
0.955
表 6  经BP神经网络训练的各类纹理特征及YOLOv3、Faster R-CNN的测试结果
图 6  遥感数据库上的P-R曲线
图 7  遥感车辆的检测效果图
图 8  遥感树木的检测效果图
图 9  遥感建筑物的检测效果图
1 WANG Y Q, LEI M A, TIAN Y State-of-the-art of ship detection and recognition in optical remotely sensed imagery[J]. Acta Automatica Sinica, 2011, 37 (9): 1029- 1039
2 WANG X , BAN Y , GUO H, et al. Deep learning model for target detection in remote sensing images fusing multilevel features [C]// 2019 IEEE International Geoscience and Remote Sensing Symposium. Yokohama: IEEE, 2019.
3 刘丽, 赵凌君, 郭承玉, 等 图像纹理分类方法研究进展和展望[J]. 自动化学报, 2018, 44 (4): 584- 607
LIU Li, ZHAO Ling-jun, GUO Cheng-yu, et al Research progress and prospect of image texture classification[J]. Acta Automatica Sinica, 2018, 44 (4): 584- 607
4 OJALA T, PIETIKAINEN M, HARWOOD D. Performance evaluation of texture measures with classification based on Kullback discrimination of distributions[C]// Proceedings of 12th International Conference on Pattern Recognition. Jerusalem: IEEE, 1994.
5 OJALA T, PIETIKAINEN M, HARWOOD D A comparative study of texture measures with classification based on featured distributions[J]. Pattern Recognition, 1996, 29 (1): 51- 59
doi: 10.1016/0031-3203(95)00067-4
6 TAN X Y, TRIGGS B Enhanced local texture feature sets for face recognition under difficult lighting conditions[J]. IEEE Transactions on Image Processing, 2010, 19 (6): 1635- 1650
doi: 10.1109/TIP.2010.2042645
7 HEIKKILA M, PIETIKAINEN M, SCHMID C Description of interest regions with local binary patterns[J]. Pattern Recognition, 2009, 42 (3): 425- 436
doi: 10.1016/j.patcog.2008.08.014
8 BABER J, BAKHTYAR M, ULLAH I, et al Effective compression of center symmetric local binary pattern[J]. Mehran University Research Journal of Engineering and Technology, 2017, 36 (2): 209- 224
9 陈家华, 陈雪云, 阳理理 基于局部线段模式特征的城市道路视觉检测[J]. 广西大学学报: 自然科学版, 2018, 43 (6): 2206- 2215
CHEN Jia-hua, CHEN Xue-yun, YANG Li-li Urban road visual inspection based on local line pattern features[J]. Journal of Guangxi University: Natural Science Edition, 2018, 43 (6): 2206- 2215
10 OJALA T, PIETIKAINEN M, MAENPAA T Multiresolution gray-scale and rotation invariant texture classification with local binary patterns[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2002, 24 (7): 971- 987
doi: 10.1109/TPAMI.2002.1017623
11 李安然, 陈雪云 一种基于矢量变换的局部二值化模式分类方法[J]. 广西大学学报:自然科学版, 2020, 45 (3): 490- 503
LI An-ran, CHEN Xue-yun A local binary pattern classification method based on vector transformation[J]. Journal of Guangxi University: Natural Science Edition, 2020, 45 (3): 490- 503
12 郭寅, 尹仕斌, 崔鹏飞, 等. 改进的LBP特征提取方法: CN111862178A [P]. 2020-10-30.
GUO Yin, YIN Shi-bin, CUI Peng-fei, et al. Improved LBP feature extraction method: CN111862178A [P]. 2020-10-30.
13 LIU X, XUE F, TENG L. Surface defect detection based on gradient LBP [C]// 2018 IEEE 3rd International Conference on Image, Vision and Computing. Chongqing: IEEE, 2018.
14 BRODERSEN K H, CHENG S O, STEPHAN K E, et al. The binormal assumption on precision-recall curves [C]// International Conference on Pattern Recognition. Turkey: IEEE, 2010.
[1] 何智翔, 丁晓青, 方驰, 文迪. 基于LBP和CCS-AdaBoost的多视角人脸检测[J]. J4, 2013, 47(4): 622-629.
[2] 戴渊明, 韦巍, 林亦宁. 基于颜色纹理特征的均值漂移目标跟踪算法[J]. J4, 2012, 46(2): 212-217.
[3] 幸锐, 徐舒畅, 张三元, 竺乐庆. 融合复小波特征和局部二值模式的纹理聚类[J]. J4, 2010, 44(5): 910-914.
[4] 李晓丽 何勇 裘正军 吴迪 陈孝敬. 基于多光谱图像的不同品种绿茶的纹理识别[J]. J4, 2008, 42(12): 2131-2138.