Image segmentation and density clustering for moving object patches extraction in remote sensing image
-
摘要:
大幅宽遥感图像的动目标检测研究中，卷积神经网络虽然取得了显著效果，但算法存在目标搜索空间庞大、模型极其消耗时间及计算资源的问题，因此本文从目标区域预筛选的角度给出了针对性优化方法。首先，基于局部误差处理的策略，改进了现有的图像分割算法来粗糙地提取动目标可能存在的区域。然后，以相邻区域合并、减少总数量和面积为目的，设计了一种基于空间约束的密度聚类算法——SC-DBSCAN，其以分治思想来降低问题的规模，通过空间尺寸的先验约束自适应地将数据划分为多个相互独立的簇，并针对簇的复杂程度选择相应的合并策略，在复杂簇中，考虑到合并结果与对象遍历顺序相关，易陷入局部最优，引入基于模拟退火思想的随机扰动有效提升了输出的图像块质量。最终，通过减少模型推断次数及避免目标的重复检测，显著地改进动目标检测的整体效率。
Abstract:Recently, moving object detection in large-scale remote sensing images achieves outstanding performance by fully convolutional neural network. However, handling such data is very time-consuming because the search space is extremely large. This paper proposes a specific improved method from the point of candidate region proposals. First, irregular candidate areas are roughly extracted by neighborhood differencing and local errors handling. Then a spatial-constraint based density cluster algorithm (SC-DBSCAN) is proposed to merge adjacent areas into patches as CNN input, which aims to reduce final outputs' amount and area. Through the priori of space constraints, this algorithm can adaptively divide data into multi types of clusters, and choose different merging strategies according to the complexity of clusters. For complicated clusters, the outputs are related to traverse sequence of each object, and thus a random search strategy based on simulated annealing is applied to avoid local optima and improve the patches' quality. Finally, by reducing the times of model inferences and avoiding redundant object detections, the detection efficiency of proposed method is significantly improved.
-
图 2 面向图像块的局部不精确分割效果比较
Figure 2. Comparsion of patch-oriented local imprecise segmentation effect
图 5 低阈值分割和OTSU阈值分割的效果比较
Figure 5. Comparsion of low-threshold segmentation and OTSU threshold segmentation
图 7 不同算法在区域1、2的合并结果
Figure 7. Merging results by different algorithms in Region 1 and Region 2
图 8 参数γ对最终图像块数量和面积的影响
Figure 8. Impact of parameter γ on final image patches' amount and area
表 1 候选运动目标质量比较
Table 1. Qualitative comparison in candidate moving objects
% 算法 精度 召回率 GMM 92.72 83.53 S-3frame 89.06 75.76 N-3frame 91.37 87.73 本文算法 90.68 93.30 
下载: 导出CSV
表 2 最终图像块质量比较
Table 2. Qualitative comparison of final image patches
算法 精度/% 召回率/% 数量比 面积比 重复比 基线 90.98 93.96 1.2241 1.2241 1.4804 HC 98.60 94.25 0.3876 1.1049 0.0843 DBSCAN 98.59 94.15 0.4018 1.1053 0.0999 本文算法 98.79 94.34 0.3818 1.0586 0.0756
下载: 导出CSV
表 3 参数γ对结果的影响
Table 3. Impact of parameter γ on result
权重γ 数量比 面积比 -0.05 1.1963 1.2008 -0.02 0.7792 0.9626 0 0.6108 0.9299 0.03 0.4947 0.952 0.06 0.4352 0.9921 0.1 0.3818 1.0586 0.3 0.2973 1.341 0.55 0.2953 1.356
下载: 导出CSV
-
[1] 徐伟, 金光, 王家骐.吉林一号轻型高分辨率遥感卫星光学成像技术[J].光学精密工程, 2017, 25(8):1969-1978.XU W, JIN G, WANG J Q.Optical imaging technology of JL-1 lightweight high resolution multispectral remote sensing satellite[J].Optics & Precision Engineering, 2017, 25(8):1969-1978(in Chinese). [2] MARK K, LUIS G, CHRIS S.Real-time tracking of low-resolution vehicles for wide-area persistent surveillance[C]//Workshop on Application of Computer Vision.Piscataway, NJ: IEEE Press, 2013: 441-448. [3] IMRAN S, MUBARAK S.Multiframe many-many point correspondence for vehicle tracking in high density wide area aerial videos[J].International Journal of Computer Vision, 2013, 104(2):198-219. [4] LALONDE R, ZHANG D, SHAH M.ClusterNet: Detecting small objects in large scenes by exploiting spatio-temporal information[C]//Computer Vision and Pattern Recognition (CVPR).Piscataway, NJ: IEEE Press, 2018: 4003-4012. [5] GAO F, ZHANG Y, WANG J, et al.Visual attention model based vehicle target detection in synthetic aperture radar images:A novel approach[J].Cognitive Computation, 2015, 7(4):434-444. doi: 10.1007/s12559-014-9312-x [6] GAO F, MA F, ZHANG Y, et al.Biologically inspired progre-ssive enhancement target detection from heavy cluttered SAR images[J].Cognitive Computation, 2016, 8(5):955-966. doi: 10.1007/s12559-016-9405-9 [7] BLASCH E, CHEN G, LING H.Vehicle classification in WAMI imagery using deep network[C]//Sensors and Systems for Space Applications.New York: SPIE, 2016: 9838-9846. [8] REN S, HE K, GIRSHICK R, et al.Faster R-CNN: Towards real-time object detection with region proposal networks[C]//International Conference on Neural Information Processing Systems.Cambridge: MIT Press, 2015: 91-99. [9] REDMON J, DIVVALA S, GIRSHICK R, et al.You only look once: Unified, real-time object detection[C]//IEEE Conference on Computer Vision and Pattern Recognition.Piscataway, NJ: IEEE Press, 2016: 779-788. [10] LIU W, ANGUELOV D, ERHAN D, et al.SSD: Single shot multibox detector[C]//European Conference on Computer Vision (ECCV).Berlin: Springer, 2016: 21-37. [11] DAI J, LI Y, HE K, et al.R-FCN: Object detection via region-based fully convolutional networks[C]//Advances in Neural Information Processing Systems.Piscataway, NJ: IEEE Press, 2016: 379-387. [12] SOMMER L W, TEUTSCH M, SCHUCHERT T, et al.A survey moving object detection for wide area motion imagery[C]//Winter Conference on Applications of Computer Vision (WACV).Piscataway, NJ: IEEE Press, 2016: 1-9. [13] 罗群明, 施霖.图像拼接方法综述[J].传感器与微系统, 2017, 36(12):4-6.LUO Q M, SHI L.Review on image stitching methods[J].Transducer & Microsystem Technologies, 2017, 36(12):4-6(in Chinese). [14] THOMAS P, MATTHEW A.Detecting and tracking all moving objects in wide-area aerial video[C]//Computer Vision and Pattern Recognition Workshops.Piscataway, NJ: IEEE Press, 2012: 15-22. [15] ZIVKOVIC Z.Improved adaptive Gaussian mixture model for background subtraction[C]//International Conference on Pa-ttern Recognition.Piscataway, NJ: IEEE Press, 2004: 28-31. [16] ESTER M, KRIEGEL H P, SANDER J, et al.Density-based spatial clustering of applications with noise[C]//International Conference on Knowledge Discovery and Data Mining.Reston: AIAA, 1996: 226-231. [17] 张小朋, 钱海忠, 岳辉丽, 等.基于模拟退火的空间聚类算法[J].测绘科学技术学报, 2010, 27(4):306-309. doi: 10.3969/j.issn.1673-6338.2010.04.018ZHANG X P, QIAN H Z, YUE H L, et al.Simulated-annealing based spatial clustering algorithm[J].Journal of Geomatics Science & Technology, 2010, 27(4):306-309(in Chinese). doi: 10.3969/j.issn.1673-6338.2010.04.018 [18] KIRKPATRICK S.Optimization by simulated annealing:Quantitative studies[J].Journal of Statistical Physics, 1984, 34(5-6):975-986. doi: 10.1007/BF01009452 [19] AFRL.Wright-patterson air force base (wpafb) dataset[EB/OL].[2017-12-10]. [20] BASHARAT A, TUREK M, XU Y L, et al.Real-time multi-target tracking at 210 megapixels/second in wide area motion imagery[C]//IEEE Winter Conference on Applications of Computer Vision.Piscataway, NJ: IEEE Press, 2014: 839-846. [21] GAO F, YOU J, WANG J, et al.A novel target detection method for SAR images based on shadow proposal and saliency analysis[J].Neurocomputing, 2017, 267:220-231. [22] GAO F, MA F, WANG J, et al.Visual saliency modeling for river detection in high-resolution SAR imagery[J].IEEE Access, 2018, 6:1000-1014. doi: 10.1109/ACCESS.2017.2777444 -
下载:
点击查看大图
计量
- 文章访问数: 748
- HTML全文浏览量: 101
- PDF下载量: 447
- 被引次数: 0
