Infrared and Visible Image Registration for Power Equipments Based on Deep Learning
-
摘要: 针对现有电力设备红外与可见光图像配准难度大、配准时间长等问题,提出一种基于深度学习的电力设备红外与可见光图像配准融合的方法。本文将特征提取与特征匹配联合在深度学习框架中,直接学习图像块对与匹配标签之间的映射关系,用于后续的配准。此外为了缓解训练时红外图像样本不足的问题,提出一种利用红外图像及其变换图像学习映射函数的自学习方法,同时采用迁移学习来减少训练时间,加速网络框架。实验结果表明:本文方法与其他4种配准算法相比性能指标均有显著提升,本文平均准确率为89.909,同其余4种算法相比分别提高了2.31%、3.36%、2.67%、0.82%,本文平均RMSE(Root Mean Square Error)为2.521,同其余4种配准算法相比分别降低了14.68%、15.24%、4.90%、1.04%,算法平均用时为5.625 s,较其余4种算法分别降低了5.57%、6.82%、2.45%、1.75%,有效提高了电力设备红外与红外可见光图像配准的效率。Abstract: A registration fusion method of infrared and visible images of power equipment based on deep learning is proposed that aims at problems with difficult and long registration time of infrared and visible images of existing power equipment. In this study, feature extraction and feature matching are combined in a deep learning framework to directly learn the mapping relationship between image block pairs and matching labels for subsequent registration. In addition, a self-learning method using infrared image and its transform image to learn the mapping function is proposed to alleviate the problem of insufficient infrared image samples during training Simultaneously, transfer learning is used to reduce the training time and accelerate the network framework. The experimental results show that the performance index of this method is significantly improved compared with the other four registration algorithms. The average accuracy of this method is 89.909, which is 2.31%, 3.36%, 2.67%, and 0.82% higher than that of the other four algorithms, respectively. The average RMSE of this method is 2.521. Compared with the other four registration algorithms, the algorithm is reduced by 14.68%, 15.24%, 4.90%, and 1.04%, respectively. The average time of the algorithm is 5.625 s, which is reduced by 5.57%, 6.82%, 2.45%, and 1.75% respectively. The efficiency of infrared and visible image registration of the power equipment must be effectively improved.
-
Keywords:
- image registration /
- deep learning /
- self-study /
- deep neural network /
- power equipment
-
0. 引言
湿法烟气脱硫利用石灰石浆液吸收烟气中的SO2,具有高效率和高可靠性等优势,已成为当前火电厂和化工厂脱硫的主力技术[1]。但是火电厂运行过程中,由于煤燃烧和SO2氧化,特别是在选择性催化还原脱销系统中催化剂作用下,SO2会更有利于氧化成生成SO3,造成烟气中SO3浓度显著增加[2-3]。烟气进入脱硫塔系统后,温度迅速冷却至酸露点以下,形成细小SO3酸雾,而单脱硫塔的SO3脱除效率仅为30%~40%[4]。
烟气中的SO3、SO2,HF及其它酸性物质会导致塔体金属发生化学腐蚀,脱硫塔内还存在电化学腐蚀、磨蚀、结晶腐蚀、垢下腐蚀和氯离子腐蚀[5]。在以上多种方式的共同作用下,受到内涂玻璃鳞片、聚烯烃共聚物、改性聚脲或纳米复合涂料等物质防护的脱硫塔仍可能发生腐蚀。
目前,对脱硫塔的检测主要有外观损伤、钢材厚度、力学性能、焊缝缺陷检测和构件变形等项目[6]。以上检测项目能够有效对停机后的脱硫塔健康状态进行评测。但是,对运行中的脱硫塔腐蚀状态进行有效检测未见文献报道。结合脱硫塔运行参数和结构参数,本文提出了采用传热学反演的方法根据表面红外热像进行运行中脱硫塔壁厚定量检测,并以某厂的烟气脱硫塔为对象进行了实验验证。
1. 脱硫塔传热模型
研究对象为图 1所示的脱硫塔,其内部环境复杂,无法布置有效的温度测量装置;脱硫过程是一个包含了传热、传质和化学反应的复杂过程,难以准确地用数学语言描述。为了建立脱硫塔的传热模型,本文进行以下简化:
1)脱硫塔为轴对称结构,内壁面热流沿周向分布均匀;
2)脱硫塔内部有玻璃鳞片防腐涂层,涂层质地均匀且热物性参数为各向同性;
3)烟气和石灰浆液对脱硫塔的传热,可等效为对脱硫塔内壁施加有沿轴向分布的加热热流;
4)忽略脱硫塔内部喷嘴和支撑结构对脱硫塔壁面温度分布的影响;
5)忽略脱硫塔内壁和防腐涂层的接触热阻;
6)忽略脱硫塔的轴向导热;
7)脱硫塔的温度场为稳态。
脱硫塔浆液区是腐蚀的重灾区,本文以浆液区段为研究对象,简化后的脱硫塔浆液区如图 2所示,具体几何参数及热物性参数见表 1。脱硫塔内壁s1存在轴向分布热流,外表面s2与环境存在对流换热,且s2面上的温度可以直接测量,其他表面s3绝热。记x=(ϕ, θ, z)为空间坐标向量,脱硫塔温度场的控制方程为:
表 1 脱硫塔几何参数及热物性参数Table 1. Geometric parameters and thermophysical parameters ofdesulfurization towerMaterial Thickness /
mmDiameter/
mmThermal conductivity/
[W/(m·K)]Anticorrosive
coating4 5520 0.35 The tower wall 14 5528 48.85 $$\frac{1}{r}\frac{\partial }{{\partial r}}(\lambda (\mathit{\boldsymbol{x}})r\frac{{\partial T(\mathit{\boldsymbol{x}})}}{{\partial r}}) + \frac{1}{{{r^2}}}\frac{\partial }{{\partial \varphi }}(\lambda (\mathit{\boldsymbol{x}})\frac{{\partial T(\mathit{\boldsymbol{x}})}}{{\partial \varphi }}) + \frac{\partial }{{\partial z}}(\lambda (\mathit{\boldsymbol{x}})\frac{{\partial T(\mathit{\boldsymbol{x}})}}{{\partial z}}) = 0$$ (1) 边界条件为:
$$ - \lambda (\mathit{\pmb{x}})\frac{{\partial T(\mathit{\pmb{x}})}}{{\partial n}} = q(z)\;\quad \mathit{\pmb{x}} \in {s_1}$$ (2) $$ - \lambda (\mathit{\pmb{x}})\frac{{\partial T(\mathit{\pmb{x}})}}{{\partial n}} = h(T(\mathit{\pmb{x}}) - {T_f})\;\quad \mathit{\pmb{x}} \in {s_2}$$ (3) $$ - \lambda ({\mathit{\pmb{x}}})\frac{{\partial T({\mathit{\pmb{x}}})}}{{\partial n}} = 0\;\;\;\;\;{\mathit{\pmb{x}}} \in {s_3}$$ (4) 式中:q(z)为内壁沿轴向z的热流密度分布;h为表面s2的对流换热系数;Tf为环境温度;n为表面的外法线方向。
若已知脱硫塔的几何结构和热边界条件,利用有限元法(finite element method, FEM)求解公式(1)~(4),可以确定脱硫塔的温度场T(x),作为壁厚检测的基础。
2. 基于反问题的脱硫塔壁厚检测
运行中脱硫塔壁厚d是影响脱硫塔表面温度分布的关键因素之一,如果脱硫塔壁面腐蚀减薄,则传热热阻减小,在腐蚀部位对应的塔外表面形成局部高温区域,如图 1(b)所示。因此,可通过求解导热反问题根据脱硫塔表面红外热像进行脱硫塔壁厚检测。
2.1 壁厚检测方案
在如图 2所示的脱硫塔表面热像图中,选取高温区域的K个温度测点,依据此测量信息Tkmea(k=1, 2, …, K)采用共轭梯度方法(conjugate gradient method,CGM)求解多变量稳态传热反问题[7-8],进行壁厚d的反演;由于脱硫塔为薄壁结构,热扩散效应较弱,则外表面高温区域可看作与内壁腐蚀区域接近。为了便于问题讨论,在本文中,内壁腐蚀区域用圆柱近似。
然而,壁厚反演过程中正问题的计算需要已知脱硫塔内壁热边界条件如热流q(z)。而内壁热流q(z)难以直接测量。如果直接同时反演壁厚和内壁热流q(z),可能因为测量信息不能够同时对壁厚和热流具有较大的灵敏度,造成检测系统的病态程度加剧。
因此,检测方案包括了两步:先进行内壁热流定量识别,再定量识别壁厚。内壁热流可采用CGM反演得到:在脱硫塔外表面高温区域附近同等高度的温度正常区域沿周向选取M个测点,以该测点的温度信息Tmmea(m=1, 2, …, M)反演该位置处的脱硫塔内壁热流q;内壁热流沿高度方向变化,但是周向分布均匀,以第一步反演得到的内壁热流q作为壁厚反演中正问题的已知热边界条件,以高温区域的温度测量信息作为壁厚反演的依据,提高了测量信息对壁厚的灵敏度,有利于削弱壁厚检测问题的病态程度。
2.2 共轭梯度算法
利用共轭梯度算法求解壁厚反问题,通过迭代优化使得目标函数J(d)足够小或者达到最大迭代步imax,对应的壁厚d即为所求。目标函数J(d)可表示为:
$$J({d_i}) = \sum\limits_{k = 1}^K {{{[T_k^{{\rm{cal}}}({d_i}) - T_k^{{\rm{mea}}}]}^2}} \le \varepsilon $$ (5) 式中:Tkmea为在脱硫塔红外热像图上提取的第K个温度测量值;di为第i次迭代得到的壁厚的猜测值;Tkcal(di)是根据di进行正问题计算得到的第k个测量位置处的温度计算值。K为在红外热像图上提取的温度测量值的数目。停机标准ε可由下式表示:
$$ \varepsilon = K{\sigma ^2} $$ (6) 式中:σ为测量误差的标准差。
CGM沿着已知点处的梯度所构造出的共轭方向迭代搜索目标函数的极小点,迭代过程中对壁厚猜测值的修正可表示为:
$$ {d_i}_{ + 1} = {d_i} - {\alpha _i}{\mathit{\boldsymbol{\gamma }}_i} $$ (7) 式中:αi为搜索步长;γi为搜索方向。
搜索步长αi表示为:
$${\alpha _i}{\rm{ = }}\sum\limits_{k = 1}^K {[T_k^{{\rm{cal}}}({d_i}) - T_k^{{\rm{mea}}}]\nabla T_k^{{\rm{cal}}}({d_i}){\mathit{\boldsymbol{\gamma }}_i}} /\sum\limits_{k = 1}^K {\nabla T_k^{{\rm{cal}}}({d_i}){\mathit{\boldsymbol{\gamma }}_i}} $$ (8) 搜索方向γi可由下式表示:
$${\boldsymbol{\gamma} _i}{\rm{ = }}\nabla J({d_i}) + {\beta _i}{d_{i - 1}}$$ (9) 式中:▽J(di)为目标函数的梯度;βi为共轭系数,可根据式(10)计算:
$${\beta _i}{\rm{ = }}{\left[ {\nabla J({d_i})/\nabla J({d_{i - 1}})} \right]^2}$$ (10) 利用CGM根据红外热像图中正常区域温度反演该位置处的脱硫塔内壁热流q,其过程可参考公式(6)~(10),在此就不一一赘述。
2.3 迭代求解流程
应用CGM根据脱硫塔红外热像图反演壁厚的计算步骤如下:
1)根据红外热像图,对异常区域进行辨识;
2)反演异常区域脱硫塔内壁热流q;
3)给出壁厚初始猜测值d0;
4)通过求解公式(1)~(4),得到测点处的计算温度Tkmea(k=1, 2, …, K),并代入公式(5):
$$J({d_i}) = \sum\limits_{k = 1}^K {{{[T_k^{\rm{cal}}({d_i}) - T_k^{\rm{mea}}]}^2}} \leqslant \varepsilon $$ (11) 如果满足上述条件,di即为所求,停止迭代;否则继续;
5)按公式(8)~(10)对CGM里的参数进行更新;
6)根据公式(7)更新壁厚d的猜测值,并返回步骤4)。
3. 脱硫塔检测结果及分析
选取环境温度Tf=20℃,对流换热系数h=10 W/(m2·K)。异常区域温度测点数量K=3;正常区域温度M=2。考虑到实际测量过程中,温度测量误差是无法消除的,通过现场标定,测量误差σ=0.055℃,ε=0.01。
1)数值实验验证
为了验证检测系统的有效性和精确性,本文先进行脱硫塔内部缺陷检测的数值实验。在数值实验中,脱硫塔的热边界条件、几何参数和热物性参数均与实际过程相同,假设真实壁厚de=16 mm。
数值实验中设置不同大小的测量误差,以考察测量误差对缺陷检测结果的影响。实验结果如表 2所示。
表 2 不同测量误差时的检测结果Table 2. The detection results of the different measurementerrorsMeasurement
error σDetect wall
thickness d/mmRelative
error/ %0.055 15.92 0.50 0.1 16.35 2.19 0.2 17.14 7.13 从表 2可以看出,随着测量误差的增大,壁厚检测结果的精确性下降。如σ=0.2℃时,相对误差为7.13%,在工程上可以接受。
2)依据现场红外热像反演
① 高温区域1
如图 3所示,高温区域1的最高温度为48.2℃,区域的最大温差为1.4℃。利用基于导热反问题的脱硫塔壁厚检测方法,对高温区域1的壁厚进行计算,结果为d=14.6 mm,即该区域玻璃鳞片厚度为0.6 mm。考虑到计算误差,可判定为防腐涂层已磨损殆尽或脱落,若不处理,塔壁金属将受到快速腐蚀。在检测后30天左右,脱硫塔停机检修,发现高温区域1的防腐涂层已脱落,证实了本方法的正确性。
② 高温区域2
如图 4所示,高温区域2的最高温度为48.3℃,区域的最大温差为1.2℃。对高温区域2的壁厚进行反演:d=15.8 mm,即该区域防腐涂层厚度为1.8 mm,可判定为防腐涂层已减薄。
③ 高温区域3
如图 5所示,高温区域3的最高温度为48.1℃,区域的最大温差为3.7℃。对高温区域3的壁厚进行反演:d=6.5mm。表明该区域脱硫塔金属塔壁已发生腐蚀,减薄了7.5mm,应尽快排查、检修。
4. 结论
本文采用导热反问题的方法,根据红外热像图对运行中的脱硫塔壁厚进行了检测。其中,脱硫塔壁厚和脱硫塔内壁面热流的反演均采用共轭梯度法。首先通过数值实验,验证了本方法的可行性。然后,依据红外热像进行反演,发现脱硫塔筒体腐蚀1处,防腐涂层脱落1处,防腐涂层减薄1处。在后续的停机检修时对上述部进行了复核,均验证了上述检测结果,表明了基于表面红外热像的脱硫塔壁厚定量检测方法的有效性和准确性。
防腐涂层厚度的不一致,脱硫塔内介质分布的不均匀,红外热像仪精度以及环境等因素,可能会给壁厚检测结果引入误差,造成识别精度下降。如何提高壁厚检测精度,仍是下一步研究的方向。
-
表 1 不同图像大小对模型的影响结果
Table 1 Influence results of different image sizes on model %
Image block ACC P R F-measure 18×18 87.62 87.88 87.28 87.58 34×34 92.27 95.21 89.01 92.01 50×50 92.19 98.21 85.94 91.67 66×66 94.10 97.92 90.10 93.85 表 2 不同配准算法配准结果对比
Table 2 Comparison of registration results of different registration algorithms
Methods SIFT Improvement CSS DNN SURF Ours Text 1 ACC 89.142 88.652 88.149 90.011 90.178 RMSE 2.117 2.032 1.713 1.649 1.642 Text 2 ACC 84.492 79.635 80.542 87.799 87.826 RMSE 4.893 5.126 3.854 3.601 3.584 Text 3 ACC 83.816 82.873 85.361 83.931 86.507 RMSE 4.023 3.893 3.901 3.878 3.852 Text 4 ACC 88.920 88.972 88.688 88.998 89.717 RMSE 2.494 2.613 2.598 2.501 2.450 Text 5 ACC 93.028 94.813 95.099 95.158 95.316 RMSE 1.247 1.208 1.189 1.108 1.077 表 3 不同匹配算法时间对比
Table 3 Time comparison of different matching algorithms
s Time SIFT CSS DNN SURF OURS Test 1 6.021 6.634 6.127 6.031 5.827 Test 2 7.138 7.041 6.732 6.644 6.521 Test 3 9.152 9.037 8.802 8.668 8.646 Test 4 6.071 6.083 5.853 5.973 5.837 Test 5 1.401 1.386 1.317 1.309 1.293 -
[1] 张旭, 魏娟, 赵冬梅, 等. 电网故障诊断的研究历程及展望[J]. 电网技术, 2013, 37(10): 2745-2553. https://www.cnki.com.cn/Article/CJFDTOTAL-DWJS201310011.htm ZHANG Xu, WEI Juan, ZHAO Dongmei, et al. Research history and prospect of power grid fault diagnosis [J]. Power Grid Technology, 2013, 37(10): 2745-2553. https://www.cnki.com.cn/Article/CJFDTOTAL-DWJS201310011.htm
[2] 刘晓康, 万曦, 涂文超, 等. 基于红外可见光图像配准的电力设备分割算法[J]. 计算机与现代化, 2020(8): 26-30. https://www.cnki.com.cn/Article/CJFDTOTAL-JYXH202008005.htm LIU Xiaokang, WAN Xi, TU Wenchao, et al. Power equipment segmentation algorithm based on infrared and visible image registration[J]. Computer and modernization, 2020(8): 26-30. https://www.cnki.com.cn/Article/CJFDTOTAL-JYXH202008005.htm
[3] 王鲲鹏, 徐一丹, 于起峰. 红外与可见光图像配准方法分类及现状[J]. 红外技术, 2009, 31(5): 270-274. DOI: 10.3969/j.issn.1001-8891.2009.05.007 WANG Kunpeng, XU Yidan, YU Qifeng. Classification and status quo of infrared and visible image registration methods[J]. Infrared Technology, 2009, 31(5): 270-274. DOI: 10.3969/j.issn.1001-8891.2009.05.007
[4] 赵洪山, 张则言. 基于文化狼群算法的电力设备红外和可见光图像配准[J]. 光学学报, 2020, 40(16): 1-14. https://www.cnki.com.cn/Article/CJFDTOTAL-GXXB202016007.htm ZHAO Hongshan, ZHANG Zeyan Infrared and visible image registration of power equipment based on cultural wolf swarm algorithm[J]. Acta Optica Sinica, 2020, 40(16): 1-14. https://www.cnki.com.cn/Article/CJFDTOTAL-GXXB202016007.htm
[5] MA J Y, JIANG X Y, FAN A X, et al. Image matching from handcrafled to deep features: A survey [J]. International Journal of Computer Vision, 2021, 129(1): 23-79. DOI: 10.1007/s11263-020-01359-2
[6] 徐全飞, 冯旗. 基于SURF和矩阵乘法的超大规模遥感图像亚像素配准算法研究[J]. 红外技术, 2017, 39(1): 44-52. http://hwjs.nvir.cn/article/id/hwjs201701009 XU Quanfei, FENG Qi Research on subpixel registration algorithm of super large scale remote sensing image based on surf and matrix multiplication[J]. Infrared Technology, 2017, 39(1): 44-52. http://hwjs.nvir.cn/article/id/hwjs201701009
[7] WU G R, KIM M, WANG Q, GAO Y Z, et al. Unsupervised deep feature learning for deformable registration of MR brain image[C]//Proceedings of the 16th International Conference on Medical Image Computing and Computer-Assisted Intervention, 2013: 649-656.
[8] CHENG X, ZHANG L, ZHENG Y F. Deep similarity learning for multimodal medical images[J]. Computer Methods in Biomechanics and Biomedical Engineering: Imaging and Visualization, 2018, 6(3): 248-252.
[9] Simonovsky M, Gulierrez-Becker B, Mateus D, et al. A deep metric for multimodal registration[C]//Proceedings of the 19th International Conference on Medical Image Computing and Computer Assisted Intervention, 2016: 10-18.
[10] Kabus S, Netsch T, Fischer B, et al. B-spline registration of 3D images with Levenberg-Marquardt optimization[C]//Proceedings of SPIE, Medical Imaging, Image Processing, 2004, 5370: 304-313.
[11] TANG L S, Hamarneh G, Celler A. Co-registration of bone CT and SPECT images using mutual information[C]//IEEE International Symposium on Signal Processing and Information Technology, 2006: 116-121.
[12] MIAO S, WANG Z J, LIAO R. A CNN regression approach for real-time 2D/3D registration[J]. IEEE Transactions on Medical Imaging, 2016, 35(5): 1352-1363.
[13] HAN Xufeng, LEUNG T, JIA Y, et al. MatchNet: Unifying feature and metric learning for patch-based matching[C]//2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2015: 3279-3286, Doi: 10.1109/CVPR.2015.7298948.
[14] Krebs J, Mansi T, Delingette H, et al. Robust non-rigid registration through agent-based action learning[C]// Proceedings of the 20th International Conference on Medical Image Computing and Computer-Assisted Intervention, 2017: 344-35.
[15] Dosovitskiy A, Fischer P, Ilg E, et al. FlowNet: learning optical flow with convolutional networks[C]// Proceedings of IEEE International Conference on Computer Vision, 2015: 2758-2766.
[16] LI H W, WANG B H, LI L. Research on the infrared and visible power-equipment image fusion for inspection robots[C]//2010 1st International Conference on Applied Robotics for the Power Industry, 2010: 1-5.
[17] 李寒, 王库, 刘韶军. 基于灰度冗余和SURF算法的电气设备红外和可见光图像配准[J]. 电力系统保护与控制, 2011, 39(11): 111-115, 123. https://www.cnki.com.cn/Article/CJFDTOTAL-JDQW201111021.htm LI Han, WANG Ku, LIU Shaojun. Infrared and visible image registration of electrical equipment based on gray redundancy and surf algorithm[J]. Power System Protection and Control, 2011, 39(11): 111-115, 123. https://www.cnki.com.cn/Article/CJFDTOTAL-JDQW201111021.htm
[18] 姜骞, 刘亚东, 方健, 等. 基于轮廓特征的电力设备红外和可见光图像配准方法[J]. 仪器仪表学报, 2020, 41(11): 252-260. https://www.cnki.com.cn/Article/CJFDTOTAL-YQXB202011028.htm JIANG Qian, LIU Yadong, FANG Jian, et al. Infrared and visible image registration method of power equipment based on contour feature[J]. Journal of Instrumentation, 2020, 41(11): 252-260. https://www.cnki.com.cn/Article/CJFDTOTAL-YQXB202011028.htm
[19] 李云红, 罗雪敏, 苏雪平, 等. 基于改进CSS的电力设备红外与可见光图像配准[J]. 激光与光电子学进展, 2021, 11(24): 1-15. https://www.cnki.com.cn/Article/CJFDTOTAL-JGDJ202212014.htm LI Yunhong, LUO Xuemin, SU Xueping, et al. Infrared and visible image registration of power equipment based on improved CSS [J]. Progress in Laser and Optoelectronics, 2021, 11(24): 1-15. https://www.cnki.com.cn/Article/CJFDTOTAL-JGDJ202212014.htm
[20] Taylor S, Drummond T. Binary histogrammed intensity patches for efficient and robust matching[J]. International Journal of Computer Vision, 2011, 94(2): 241-265.
[21] MA W, ZHANG J, WU Y, et al. A novel two-step registration method for remote sensing images based on deep and local features[J]. IEEE Trans Geosci Remote Sen, 2019, 57: 4834-4843.
[22] YE F, SU Y, XIAO H, et al. Remote sensing image registration using convolutional neural network features[J]. IEEE Geosci. Remote Sens Lett. , 2018, 15: 232-236.
[23] LIU X L, JIANG D S, WANG M N, et al. Image synthesis based multi-modal image registration framework by using deep fully convolutional networks[J]. Medical & Biological Engineering & Computing, 2019, 57(5): 1037-1048.
[24] 周微硕, 安博文, 赵明, 等. 基于几何不变性和局部相似特征的异源遥感图像配准算法[J]. 红外技术, 2019, 41(6): 561-571. http://hwjs.nvir.cn/article/id/hwjs201906012 ZHOU Weishuo, AN Bowen, ZHAO Ming, et al A registration algorithm for heterogeneous remote sensing images based on geometric invariance and local similarity [J]. Infrared Technology, 2019, 41(6): 561-571. http://hwjs.nvir.cn/article/id/hwjs201906012
[25] Salakhutdinov R, Hinton G E. Deep Boltzmann machines[C]// Proceedings of the 12th International Conference on Artificial Intelligence and Statistics, 2009: 448-455.
[26] 杨琼楠, 马天力, 杨聪锟, 等. 基于优化采样的RANSAC图像匹配算法[J]. 激光与光电子学进展, 2020, 57(10): 259-266. https://www.cnki.com.cn/Article/CJFDTOTAL-JGDJ202010029.htm YANG Qiongnan, MA Tianli, YANG Congkun, et al. RANSAC image matching algorithm based on optimized sampling[J]. Progress in Laser and Optoelectronics, 2020, 57(10): 259-266. https://www.cnki.com.cn/Article/CJFDTOTAL-JGDJ202010029.htm
[27] 李航. 统计学习方法[M]. 北京: 清华大学出版社, 2012. LI Hang. Statistical Learning Method[M]. Beijing: Tsinghua University Press, 2012.
[28] 朱英宏, 李俊山, 汤雨. 基于CSS角点提取的红外与可见光图像匹配算法[J]. 系统工程与电子技术, 2011, 33(11): 2540-2545. https://www.cnki.com.cn/Article/CJFDTOTAL-XTYD201111037.htm ZHU Yinghong, LI Junshan, TANG Yu. Infrared and visible image matching algorithm based on CSS corner extraction[J]. Systems Engineering and Electronic Technology, 2011, 33(11): 2540-2545. https://www.cnki.com.cn/Article/CJFDTOTAL-XTYD201111037.htm
[29] 岳国华, 邢晓利. 基于卷积神经网络和校正网络相结合的遥感图像配准方法研究[J]. 计算机应用与软件, 2021, 38(11): 185-190. https://www.cnki.com.cn/Article/CJFDTOTAL-JYRJ202111031.htm YUE Guohua, XING Xiaoli. Research on remote sensing image registration method based on convolution neural network and correction network[J]. Computer Applications and Software, 2021, 38(11): 185-190. https://www.cnki.com.cn/Article/CJFDTOTAL-JYRJ202111031.htm
[30] FU J, LI W S, XU L M. DSAGAN: A generative adversarial network based on dual-stream attention mechanism for anatomical and functional image fusion [J]. Information Sciences, 2021, 576: 484-506.
[31] 王少杰, 武文波, 徐其志. VGG与DoG结合的光学遥感影像精确配准方法[J]. 航天返回与遥感, 2021, 42(5): 76-84. https://www.cnki.com.cn/Article/CJFDTOTAL-HFYG202105010.htm WANG Shaojie, WU Wenbo, XU Qizhi. An accurate registration method of optical remote sensing images based on VGG and DoG[J]. Aerospace Return and Remote Sensing, 2021, 42(5): 76-84. https://www.cnki.com.cn/Article/CJFDTOTAL-HFYG202105010.htm
-
期刊类型引用(1)
1. 单幼芳. 基于深度学习算法的太赫兹人体安检图像处理技术研究. 电脑编程技巧与维护. 2024(07): 127-129 . 百度学术
其他类型引用(1)