随着卫星遥感数据井喷式涌现，无人机获取影像的效率与质量稳步提升，无人机遥感技术与卫星SAR在应用领域都取得了长足的发展。无人机遥感技术与卫星SAR均具有一定的技术局限性，通常在灾害识别、监测、评估以及特殊地形下的地形、地物边界提取与监测中需要协同使用，这对影像处理技术的准确性和时效性提出新的挑战。影像处理技术中影像匹配是一个重要的环节，也是国内外研究人员研究的热点，影像匹配的效果直接影响后续影像处理的效果及精度。为实现对无人机影像、SAR影像快速且精确的影像匹配，本文选择两种运算速度较快的ORB算法和SURF算法进行改进，针对ORB算法在进行影像匹配时出现的精匹配点数量少、匹配正确率低的不足，通过将原算法的描述符替换成二进制BEBLID描述符对算法进行改进，针对SURF算法进行影像匹配时精匹配点数量少的缺点，通过将原算法的描述符替换成DEEPDESC描述符对算法进行改进。为验证本文两种改进算法的匹配性能及鲁棒性，本文使用两种改进算法、两种原算法再加上常规的AKAZE算法与基于深度学习LIFT算法做对比验证，共六种算法进行无人机影像匹配实验、卫星SAR中的哨兵一号影像匹配实验和基于graffiti数据集的鲁棒性验证，从匹配结果、匹配时间、精匹配点对数、匹配正确率和匹配精度五个方面对本文改进算法、原算法、常规算法和基于深度学习的算法进行比较分析。主要结论如下。 （1）探究ORB改进算法与SURF改进算法在进行无人机影像匹配时的匹配性能和适用性。实验结果表明在进行无人机影像匹配时，ORB改进算法平均精匹配点对数比ORB算法多197对，增加了22.26%，平均精匹配率高6.46%；SURF改进算法的平均匹配时间相比SURF算法快了25.59%，平均精匹配点对数比SURF算法多124对，增加了10.86%。六种算法相比较，ORB改进算法是完成对时间效率要求较高的无人机影像匹配最合适的算法；SURF改进算法的匹配点数量最多，匹配精度最高，但匹配时间不够快速，虽然较SURF算法有所减少，但是不及ORB改进算法与ORB算法，SURF改进算法适用于对时间效率要求一般，对精匹配点对数要求较高的情况。 （2）探究ORB改进算法与SURF改进算法在进行哨兵一号影像匹配时的匹配性能和适用性。实验结果表明在进行哨兵一号影像匹配时，ORB改进算法平均精匹配点对数比ORB算法多184对，增加了20.67%，平均匹配率提升了9.23%；SURF改进算法平均匹配正确率比SURF算法高3.28%，平均精匹配点对数比SURF算法多244对，增加了19.67%。六种算法相比较，ORB改进算法是对时间效率要求较高的哨兵一号影像匹配时六种算法中最优秀的算法；SURF改进算法的匹配点数量最多，匹配精度最高，对这两方面具有较高需求时SURF改进算法是最优的选择；当哨兵一号影像中存在大面积水体区域时SURF改进算法与LIFT算法均有良好表现。 （3）验证各算法进行不同程度亮度、模糊、旋转和仿射变形畸变下的鲁棒性。通过实验发现，ORB算法能够很好的适应亮度畸变，对模糊、旋转、仿射变形的适应性不强；ORB改进算法对亮度、模糊、旋转和仿射变形时的适应性和匹配精度都比ORB算法有所提升；SURF算法对模糊和仿射变形的适应性不高。SURF改进算法对仿射变形适应性最强，匹配精度优于SURF算法。AKAZE算法对模糊畸变的适应性最好，对旋转、亮度和仿射变形的适应性在六种算法中比较中庸，但是匹配精度一直都比较稳定；LIFT算法对亮度畸变的适应性最好。

With the emergence of satellite remote sensing data, the efficiency and quality of UAV image acquisition have been improved steadily. UAV remote sensing technology and satellite SAR have made great progress in the application field. UAV remote sensing technology and satellite SAR both have certain technical limitations, and usually need to be used collaboratively in disaster identification, monitoring, assessment, and terrain feature extraction and monitoring under special terrain, which poses a new challenge to the accuracy and timeliness of image processing technology. Image matching is an important link in image processing technology, and it is also a hotspot researched by researchers at home and abroad. In order to achieve fast and accurate image matching for UAV images and SAR images, two fast ORB algorithm and SURF algorithm are selected to improve the speed. Aiming at the problems of few exact matching points and low matching accuracy in the ORB algorithm, it was improved by replacing the descriptor of the original algorithm with BEBLID descriptor. And aiming at the problem of accurate matching points in SURF algorithm, it was improved by replacing the descriptors of original algorithm with DEEPDESC descriptors. In order to verify the matching performance and robustness performance of the two improved algorithms, this paper conducts the UAV image matching experiment, sentinel one image matching experiment in satellite SAR and robustness verification based on graffiti data set, by using the improved algorithm, the original algorithm, the conventional algorithm and the deep learning-based LIFT algorithm. The six algorithms are analyzed and compared from five aspects: matching result, matching time, precise matching points, matching accuracy and matching precision. The main conclusions are as follows: (1)Explore the matching performance and applicability of improved ORB algorithm and improved SURF algorithm in UAV image matching. The experimental results show that when performing UAV image matching, there are 197 more pairs of precision matching points in the improved ORB algorithm than that in the ORB algorithm, which increases by 22.26% , and the average precision matching rate increased 6.46%; The average matching time of the improved SURF algorithm is 25.59% faster than that of SURF algorithm, and the average number of precise matching points is 124 pairs more than that of SURF algorithm, which increases by 10.86% . Compared with the six algorithms, the improved ORB algorithm is the most suitable one for UAV image matching with high time efficiency. The improved SURF algorithm has the most matching points and the highest matching precision, but the matching time is not fast enough. Although it is faster than SURF, it is not as good as improved ORB algorithm and ORB algorithm. The improved SURF algorithm is suitable for the case where the logarithm of the exact matching point is high but the request of the time efficiency is not so rigid. (2)Explore the matching performance and applicability of improved ORB algorithm and improved SURF algorithm in Sentinel-1 image matching. The experimental results show that when performing image of Sentinel-1 matching, there are 184 more pairs of mean exact matching points in the improved ORB algorithm than in the ORB algorithm, which increases by 20.67% and the average matching rate increases by 9.23%; the average matching rate of the improved SURF algorithm is 3.28% higher than that of the SURF algorithm, the average number of matching point pairs is 244 more than that of SURF algorithm, which increases by 19.67%. Compared with the six algorithms, the improved ORB algorithm is the best one among the six algorithms for the high level time-efficient Sentinel-1 image matching, and the improved SURF algorithm has the largest number of matching points and the highest matching accuracy. The improved SURF algorithm is the best choice when there is high demand for these two aspects, and the improved SURF algorithm and LIFT algorithm both have good performance when there is large area of water body in Sentinel-1 image. (3) The robustness of the six algorithms is verified under different luminance, blur, rotation and affine distortion. The experiments show that the ORB algorithm can adapt to the brightness distortion well, but not to the blur, rotation and affine distortion; The adaptability and matching accuracy of the improved ORB algorithm for brightness, ambiguity, rotation and affine deformation are improved; SURF algorithm is not suitable for fuzzy and affine deformation; The improved SURF algorithm has the best adaptability to affine deformation, and its matching precision is better than SURF algorithm; AKAZE algorithm has the best adaptability to fuzzy distortion, and the adaptability to rotation, brightness and affine distortion is moderate among the six algorithms, but the matching accuracy can keep stable; The LIFT algorithm has the best adaptability to the luminance distortion.

[1]朱述龙. 遥感图像处理与应用[M]. 北京: 科学出版社, 2006.

[2] Armin G. Development and Status of Image Matching in Photogrammetry[J]. The Photogrammetric Record, 2012,27(137).

[3] 何瑞瑞, 赵凤君, 曾玉婷, 等. 多源遥感影像在森林火灾监测中的应用[J]. 世界林业研究, 2022, 35(2): 59-63.

[4] 吴普侠, 冯伟, 王硕, 等. 遥感新技术在秦岭保护行动中的应用[J]. 西北林学院学报, 2021, 36(6): 211-219.

[5] 付佳妮, 刘洪华, 董杰, 等. 遥感技术在青岛市城市地质调查中的应用[J]. 海洋地质前沿, 2021, 37(9): 69-78.

[6] 胡博文, 李予红, 张聚斌. 无人机遥感3D建模在露天矿山环境综合治理勘查设计中的应用[J]. 桂林理工大学学报, 2020, 40(4): 726-734.

[7] 路尧, 丁小牛. 无人机遥感在地震灾害损失预评估中的应用[J]. 测绘通报, 2021(S1): 170-172.

[8] San J, Jiang W. On-Board GNSS/IMU Assisted Feature Extraction and Matching for Oblique UAV Images[J]. Remote Sensing,2017,9(8).

[9] 郭庆华, 胡天宇, 刘瑾, 等. 轻小型无人机遥感及其行业应用进展[J]. 地理科学进展, 2021, 40(9): 1550-1569.

[10]王昆, 杨鹏, 吕文生, 等. 无人机遥感在矿业领域应用现状及发展态势[J]. 工程科学学报, 2020,42(9): 1085-1095.

[11]孙艳. 哨兵一号星载SAR宽幅交叉极化数据去噪与海冰覆盖检测算法研究[D]. 北京: 中国科学院大学(中国科学院空天信息创新研究院), 2020.

[12]张成龙, 李振洪, 余琛, 等. 利用GACOS辅助下InSAR Stacking对金沙江流域进行滑坡监测[J]. 武汉大学学报(信息科学版), 2021, 46(11): 1649-1657.

[13]张建新, 钱罕林, 何薇, 等. 无人机遥感的滑坡地质灾害监测[J]. 新疆地质, 2021, 39(3): 507-509.

[14]高思远, 黄诗峰, 孙亚勇, 等. Sentinel-1 SAR数据在四川茂县山体滑坡灾害监测中的应用研究[J]. 中国水利水电科学研究院学报, 2019, 17(2): 132-138.

[15]龙林丽, 刘英, 张旭阳, 等. 无人机在矿区表土特征及地质灾害监测中的应用[J]. 煤田地质与勘探, 2021, 49(6): 200-211.

[16]王志一, 徐素宁, 王娜, 等. 高分辨率光学和SAR遥感影像在地震地质灾害调查中的应用——以九寨沟M7.0级地震为例[J]. 中国地质灾害与防治学报, 2018, 29(5): 81-88.

[17]吕素娜, 薛思涵, 谢婷, 等. 哨兵一号SAR数据在鄱阳湖洪涝灾害监测中的应用[J]. 卫星应用, 2021(8): 51-55.

[18]张晓宁, 温奇, 王薇. 河南郑州“7·20”特大暴雨灾害无人机遥感监测评估[J]. 中国减灾, 2021(23): 12-13.

[19]刘国祥, 张波, 张瑞, 等. 联合卫星SAR和地基SAR的海螺沟冰川动态变化及次生滑坡灾害监测[J]. 武汉大学学报(信息科学版), 2019, 44(7): 980-995.

[20]张波, 张瑞, 刘国祥, 等. 基于SAR影像的贡巴冰川末端冰湖年际变化监测及溃决规律分析[J]. 武汉大学学报(信息科学版), 2019, 44(7): 1054-1064.

[21]符茵, 刘巧, 刘国祥, 等. 基于无人机影像的冰面流速与高程变化提取方法[J]. 地理学报, 2021, 76(5): 1245-1256.

[22]贾迪, 朱宁丹, 杨宁华, 等. 图像匹配方法研究综述[J]. 中国图象图形学报, 2019, 24(5): 677-699.

[23]孙钰珊, 张力, 艾海滨, 等. 倾斜影像匹配与三维建模关键技术发展综述[J]. 遥感信息, 2018, 33(2): 1-8.

[24]敖成欢. 基于ASIFT改进的倾斜航空影像特征匹配算法研究[D]. 贵阳: 贵州师范大学, 2020.

[25]陈晓勇, 何海清, 周俊超, 等. 低空摄影测量立体影像匹配的现状与展望[J]. 测绘学报, 2019, 48(12): 1595-1603.

[26]宋伟. 基于改进AKAZE特征的无人机影像匹配算法研究[D]. 徐州: 中国矿业大学, 2019.

[27]张亮格. 基于局部仿射不变特征的倾斜影像匹配技术研究[D]. 合肥: 合肥工业大学, 2020.

[28]Dubuisson M P, Jain A K. A Modified Hausdorff Distance for Object Matching[C]// PatternRecognition, 1994. Vol. 1 - Conference A: Computer Vision & Image Processing.Proceedings of the 12th IAPR International Conference on. IEEE, 1994.

[29]Masry S E, An Antomatic Method for Height Profile Determination[J]. The Photogrammetric Record, 1973, 7(42):728-730.

[30]Barnea D I, Silverman H F. A Class of Algorithms for Fast Digital Image Registration[J]. IEEE Transactions on Computers, 1972, C-21(2):179-186.

[31]Ackermann F, High Precision Digital Image Correlation[C]// Proceedings of the 39tb Photogrammetric Week. Stuutgart: University of Stuttgart, 19.

[32]谢维达, 周宇恒, 寇若岚. 一种改进的快速归一化互相关算法[J]. 同济大学学报(自然科学版), 2011, 39(8): 1233-1237.

[33]吴鹏, 徐洪玲, 宋文龙. 结合小波金字塔的快速NCC图像匹配算法[J]. 哈尔滨工程大学学报, 2017, 38(5): 791-796.

[34]戴帅龙, 王家禹, 王智鑫. 基于GPU加速的改进NCC算法[J]. 中国新通信, 2018, 20(1): 174-175.

[35]牛牧野, 周春霞, 刘婷婷. 基于改进NCC算法的东南极极记录冰川流速提取研究[J]. 极地研究, 2016, 28(2): 243-249.

[36]王万同, 刘鹏飞, 韩志刚. 基于SIFT和NCC的多源遥感影像配准方法[J]. 计算机工程与应用, 2012, 48(3): 10-12+16.

[37]徐晓伟. 无人机影像匹配算法研究[D]. 西安: 长安大学, 2019.

[38]闫倩倩. 基于改进AKAZE和K-VFC无人机影像匹配算法研究[D]. 西安: 西安科技大学, 2021.

[39]Moravec H P. Towards Automatic Visual Obstacle Avoidance[C]// Proceedings of the 5th international joint conference on Artificial intelligence - Volume 2. Morgan Kaufmann Publishers Inc. 1977.

[40]C Harris, M Stephens. A Combined Corner and Edge Detector[J]. In Proc. of Fourth Alvey Vision.

[41]Lowe D G. Distinctive Image Features from Scale-Invariant Keypoints[J]. International Journal of Computer Vision, 2004, 60(2):91-110.

[42]Bay H, Tuytelaars T, Gool L V. SURF: Speeded up Robust Features[J]. 2006.

[43]Rosten E, Porter R, Drummond T. Faster and Better: a Machine Learning Approach to Corner Detection[J]. IEEE Trans Pattern Anal Mach Intell, 2008, 32(1):105-119.

[44]Leutenegger S, Chli M, Siegwart R Y. BRISK: Binary Robust Invariant Scalable Keypoints[J]. 2011.

[45]Rublee E, Rabaud V, Konolige K, et al. ORB: An Efficient Alternative to SIFT or SURF[J]. 2011.

[46]Alcantarilla P F, Bartoli A, Davison A J, KAZE Features[M]. Springer Berlin Heidelberg, 2012:214-227.

[47]Pablo Fernández Alcantarilla, Bartoli A, Davison A J. KAZE Features[M]// Computer Vision – ECCV 2012. Springer Berlin Heidelb.

[48]刘利锋, 马燕, 张相芬, 等. 采用二值SIFT特征描述的图像匹配方法[J]. 计算机应用与软件, 2016, 33(12): 152-155+210.

[49]甘小红, 覃志东, 蔡勇, 等. 一种基于SIFT的改进优化特征匹配算法[J]. 智能计算机与应用, 2021, 11(11): 5-9.

[50]毛克乐. 改进SURF和Delaunay三角网在图像匹配中应用[J]. 沈阳工业大学学报, 2021, 43(4): 432-438.

[51]陈雪松, 陈秀芳, 毕波, 等. 基于改进SURF的图像匹配算法[J]. 计算机系统应用, 2020, 29(12): 222-227.

[52]杨倩兰, 宋丽梅, 黄浩珍, 等. 基于改进ORB算法的图像匹配方法[J/OL]. 计算机集成制造系统, 2022-02-21.

[53]邹斌, 赵小虎, 尹智帅. 基于改进ORB的图像特征匹配算法研究[J]. 激光与光电子学进展, 2021, 58(2): 96-103.

[54]Lecun Y, Bengio Y, Hinton G. Deep Learning[J]. Nature, 2015, 521(7533):436-444.

[55]Schmidhuber J. Deep Learning in Neural Networks: an Overview[J]. Neural Networks, 2015, 61:85-117.

[56]季顺平, 罗冲, 刘瑾. 基于深度学习的立体影像密集匹配方法综述[J]. 武汉大学学报(信息科学版), 2021, 46(2): 193-202.

[57]Yi K M,Trulls E,Lepetitv, et al. LIFT: Learned invariant feature transform [C]∥Proceedings of the 14th European Conference on Computer Vision. Amsterdam, Netherlands, 2016:467-483.

[58]Daniel D, Tomasz M, Andrew R. Super Point: Self-su-pervised Interest Point Detection and Description[C]//The IEEE Conference on Computer Vision and Pattern Recognition(CVPR) Workshops. Salt Lake City, USA, 2018: 224-236.

[59]Dusmanu M, Rocco I, Pajdla T, et al. D2-Net: A Trainable CNN for Joint Detection and Description of Local Features[EB / OL]. ( 2019-05-09 ) .

[60]Zhang F, Prisacariu V, Yang R, et al. GA-Net:Guided Aggregation Net for End-to-End Stereo Matching [C]. IEEE Conference on Computer Vision and Pattern Recognition, Los Angeles, America, 2019.

[61]万尧. 基于深度学习的图像匹配算法研究[D]. 南昌: 南昌大学, 2019.

[62]霍门婕, 范大昭, 董杨. 异源多分辨率卫星影像匹配算法研究[J]. 测绘通报, 2019(6): 47-51.

[63]蓝朝桢, 施群山, 崔志祥, 等. 无人机影像与卫星影像配准的卷积神经网络方法[J]. 测绘科学技术学报, 2020, 37(1): 39-44.

[64]余美. 倾斜立体影像匹配若干问题研究[D]. 徐州: 中国矿业大学, 2018.

[65]吴冰冰. 基于视觉局部特征的遥感影像匹配方法研究[D]. 长沙: 国防科学技术大学, 2016.

[66]黄一昕. 基于改进SIFT算法的无人机倾斜影像匹配研究[J]. 铁道勘察, 2021, 47(1): 14-17+22.

[67]朱标. 基于优化ORB算法的图像角点特征匹配方法[J]. 重庆工商大学学报(自然科学版), 2021, 38(4): 42-51.

[68]薛国超, 刘龙龙, 高超. 基于SIFT算法及三角形约束的SAR影像匹配方法[J]. 太赫兹科学与电子信息学报, 2020, 18(5): 802-807.

[69]邓仕雄, 闫星光, 刘继庚, 等. 一种基于AKAZE算法的山区无人机影像匹配[J]. 矿山测量, 2020, 48(5): 105-109.

[70]李闯, 王晓红, 何志伟. 利用A-AKAZE算法进行喀斯特地区无人机影像匹配[J]. 测绘通报, 2020(6): 53-56.

[71]李瑞祥, 赵海涛, 葛小三, 等. 多匹配策略融合的无人机影像匹配方法[J]. 测绘科学, 2021, 46(3): 87-95.

[72]李钦, 游雄, 李科, 等. PCA-SIFT特征匹配算法研究[J]. 测绘工程, 2016, 25(4): 19-24.

[73]于翔舟, 王慧, 李烁, 等. 利用二进制特征描述符的改进SIFT影像匹配算法[J]. 海洋测绘, 2019, 39(1): 39-43.

[74]张培佩, 王永波, 宋伟. 基于改进SIFT算法的无人机影像匹配[J]. 湖南科技大学学报(自然科学版), 2019, 34(2): 90-95.

[75]薄单, 李宗春, 王晓南, 等. 适用于倾斜影像的加速KAZE-SIFT特征提取算法[J]. 计算机应用, 2019, 39(7): 2093-2097.

[76]曹留霞, 王晓红, 李闯, 等. BRISK-DAISY无人机影像匹配算法研究[J]. 测绘工程, 2020, 29(2): 17-22.

[77]代友林, 王晓, 李闯, 等. 基于BRISK-LATCH的喀斯特山区无人机影像匹配算法[J]. 地理与地理信息科学, 2021, 37(1): 26-32.

[78]Suárez, Iago, Ghesn Sfeir,et al. BEBLID: Boosted Efficient Binary Local Image Descriptor.[J] Pattern Recognition Letters. 2020: 366–72.

[79]Simo-Serra E,Trulls E,Ferraz L,et al. Discriminative Learning of Deep Convolutional Feature Point Descriptors [C]// Proceedings of 2015 IEEE International Conference on Computer Vision. San-tiago,Chile: IEEE,2015: 118-126.