Automatic extraction method of air bubbles in icing microscopic images
-
摘要:
针对传统图像分割方法提取结冰显微图像中的气泡漏检率高和无法分离粘连气泡的问题，提出深度神经网络和传统分割算法相结合的方法。基于Attention U-Net网络，采用双分支融合预测策略对结冰显微图像中的气泡进行提取。针对部分气泡粘连问题，引入直方图均衡化和局部极小值，采用基于距离变换的分水岭算法，对结冰显微图像中粘连气泡进行二次分割。实验结果表明：通过双分支融合预测的Attention U-Net网络，对不同结冰显微图像中的气泡提取更精确，特别是对于较小气泡的检出率更高。测试图像的像素精度、平均像素精度、平均交并比和频权交并比分别达到0.9767、0.8916、0.8188和0.9575。基于距离变换的分水岭算法在粘连气泡分割中也展现了良好的性能，为后续统计气泡个数、面积等特征提供可量化的数据支撑。
Abstract:In view of the problems of high missed detection rate and inability to separate adhering bubbles in the extraction of icing microscopic images by traditional image segmentation methods, a method combining deep neural network and traditional segmentation algorithm was proposed. Based on the attention U-Net network, a two-branch fusion prediction strategy was adopted to extract the bubbles in the icing microscopic images. For some bubble adhesion problems, histogram equalization and local minima were introduced, and a watershed algorithm based on distance transformation was used to segment the adhesion bubbles in icing microscopic images twice. The experimental results showed that the two-branch prediction prediction Attention U-Net network was more accurate for the extraction of bubbles in different icing microscopic images, especially the detection rate for smaller bubbles was higher. The pixel accuracy, mean pixel accuracy, mean inter-section over union and frequency weighted intersection over union of the test images reached 0.9767, 0.8916, 0.8188 and 0.9575, respectively. The watershed algorithm based on distance transformation also showed good performance in the segmentation of sticky bubbles, providing quantifiable data support for the subsequent statistics of the number and area of bubbles.
-
图 3 结冰显微图像中气泡的自动提取方法流程图
Figure 3. Flow chart of automatic extraction method of air bubbles in icing microscopic image
图 7 结冰显微图像中的气泡在不同模型中提取结果
Figure 7. Bubbles in icing microscopy images extract results in different models
图 8 大尺度结冰显微图像气泡提取结果
Figure 8. Bubble extraction results from large-scale icing microscopic images
表 1 各种方法的量化指标
Table 1. Quantitative indicators of various methods
方法/指标 Qpa Qmpa Rmiou Rfwiou U-Net 0.9700 0.7720 0.7407 0.9428 Attention U-Net 0.9755 0.8271 0.7912 0.9535 R2U-Net 0.9697 0.7882 0.7473 0.9432 本文方法 0.9767 0.8916 0.8188 0.9575 
下载: 导出CSV
-
[1] 郭林亮,王国智,裴彬彬,等. 非对称结冰情形下飞机横航向系统稳定性分析方法研究[J]. 空军工程大学学报(自然科学版),2020,21(5): 23-28.GUO Linliang,WANG Guozhi,PEI Binbin,et al. Research on the stability analysis method of aircraft lateral heading system under asymmetric icing conditions[J]. Journal of Air Force Engineering University (Natural Science Edition),2020,21(5): 23-28. (in Chinese) [2] KIND R J,POTAPCZUK M G,FEO A,et al. Experimental and computational simulation of in-flight icing phenomena[J]. Progress in Aerospace Sciences,1998,34(5/6): 257-345. [3] CHANG Shinan,LENG Mengyao,WU Hongwei,et al. Aircraft ice accretion prediction using neural network and wavelet packet transform[J]. Aircraft Engineering and Aerospace Technology,2016,88(1): 128-136. doi: 10.1108/AEAT-05-2014-0057 [4] 何磊,钱炜祺,易贤,等. 基于转置卷积神经网络的翼型结冰冰形图像化预测方法[J]. 国防科技大学学报,2021,43(3): 98-106. doi: 10.11887/j.cn.202103013HE Lei,QIAN Weiqi,YI Xian,et al. Image prediction method of airfoil icing based on transposed convolutional neural network[J]. Journal of National University of Defense Technology,2021,43(3): 98-106. (in Chinese) doi: 10.11887/j.cn.202103013 [5] DONG Yiqun,AI Jianliang. Research on inflight parameter identification and icing location detection of the aircraft[J]. Aerospace Science and Technology,2013,29(1): 305-312. doi: 10.1016/j.ast.2013.03.012 [6] 伍强, 徐浩军, 魏扬, 等. 结冰条件下飞机气动/运动耦合特性研究[J]. 航空学报, 2022, 43(8): 1-14WU Qiang, XU Haojun, WEI Yang, et al. Research on the aerodynamic/kinematic coupling characteristics of aircraft under icing conditions[J]. Acta Aeronautica et Astronautica Sinica, 2022,43(8): 1-14. (in Chinese) [7] 杜雁霞,李明,桂业伟,等. 飞机结冰热力学行为研究综述[J]. 航空学报,2017,38(2): 30-41.DU Yanxia,LI Ming,GUI Yewei,et al. A review of research on thermodynamic behavior of aircraft icing[J]. Acta Aeronautica et Astronautica Sinica,2017,38(2): 30-41. (in Chinese) [8] 李伟斌,宋超,易贤,等. 动态结冰孔隙结构三维建模方法[J]. 化工学报,2020,71(3): 1009-1017.LI Weibin,SONG Chao,YI Xian,et al. Three-dimensional modeling method of dynamic icing pore structure[J]. CIESC Journal,2020,71(3): 1009-1017. (in Chinese) [9] 李伟斌,马洪林,易贤,等. 温度对动态结冰微观结构特性影响定量分析[J]. 空气动力学学报,2019,37(5): 748-753.LI Weibin,MA Honglin,YI Xian,et al. Quantitative analysis of the effect of temperature on the microstructure characteristics of dynamic icing[J]. Acta Aerodynamica Sinica,2019,37(5): 748-753. (in Chinese) [10] 杜雁霞,桂业伟,柯鹏,等. 飞机结冰冰型微结构特征的分形研究[J]. 航空动力学报,2011,26(5): 997-1002. doi: 10.13224/j.cnki.jasp.2011.05.015DU Yanxia,GUI Yewei,KE Peng,et al. Fractal study on microstructure characteristics of aircraft icing[J]. Journal of Aerospace Power,2011,26(5): 997-1002. (in Chinese) doi: 10.13224/j.cnki.jasp.2011.05.015 [11] 李伟斌,魏东,杜雁霞,等. 动态结冰微观孔隙结构定量分析[J]. 航空学报,2018,39(2): 112-119.LI Weibin,WEI Dong,DU Yanxia,et al. Quantitative analysis of dynamic icing microscopic pore structure[J]. Acta Aeronautica et Astronautica Sinica,2018,39(2): 112-119. (in Chinese) [12] SHELHAMER E,LONG J,DARRELL T. Fully convolutional networks for semantic segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence,2017,39(4): 640-651. [13] BADRINARAYANAN V,KENDALL A,CIPOLLA R. Segnet: a deep convolutional encoder-decoder architecture for image segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence,2017,39(12): 2481-2495. doi: 10.1109/TPAMI.2016.2644615 [14] ZHAO Hengshuang, SHI Jianping, QI Xiaojuan, et al. Pyramid scene parsing network[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway, US: IEEE, 2017: 6230-6239. [15] ZHAO Hengshuang, QI Xiaojuan, SHEN Xiaoyong, et al. Icnet for real-time semantic segmentation on high-resolution images[C]//Proceedings of the European Conference on Computer Vision. Salt Lake City, US: Springer, 2018: 405-420. [16] RONNEBERGER O, FISCHER P, BROX T. U-net: Convolutional networks for biomedical image Segmentation[C]//International Conference On Medical Image Computing and Computer-assisted Intervention. Munich, German: Springer, 2015: 234-241. [17] ALOM M Z, YAKOPCIC C, HASAN M, et al. Recurrent residual U-Net for medical image segmentation[J]. Journal of Medical Imaging, 2019, 6(1): 014006. [18] OKTAY O, SCHLEMPER J, FOLGOC L L, et al. Attention U-Net: learning where to look for the pancreas[EB/OL]. [2022-01-15]. https://doi.org/10.48550/arXiv.1804.03999. [19] CASTLEMAN K R. Digital image processing[M]. New Jersey, US: Prentice Hall Press, 1996. [20] WU Yanpeng,PENG Xiaoqi,RUAN Kai,et al. Improved image segmentation method based on morphological reconstruction[J]. Multimedia Tools and Applications,2017,76(19): 19781-19793. doi: 10.1007/s11042-015-3192-2 [21] 王娅. 血液红细胞图像自适应标记分水岭分割算法[J]. 中国图象图形学报,2017,22(12): 1779-1787. doi: 10.11834/jig.170330WANG Ya. Watershed segmentation algorithm for adaptive labeling of blood red blood cell images[J]. Chinese Journal of Image Graphics,2017,22(12): 1779-1787. (in Chinese) doi: 10.11834/jig.170330 [22] AMANKWAH A, ALDRICH C. Automatic ore image segmentation using mean shift and watershed transform[C]//Proceedings of 21st International Conference Radioelektronika. Brno, Czech: IEEE, 2011: 1-4. [23] 雷旺雄,卢军. 基于全卷积网络与凹点搜索的重叠葡萄分割算法[J]. 光电子·激光,2021,32(3): 231-240.LEI Wangxiong,LU Jun. Overlapping grape segmentation algorithm based on fully convolutional network and pit search[J]. Journal of Optoelectronics·Laser,2021,32(3): 231-240. (in Chinese) [24] 游迎荣,范影乐,庞全. 基于距离变换的粘连细胞分割方法[J]. 计算机工程与应用,2005,41(20): 206-208. doi: 10.3321/j.issn:1002-8331.2005.20.061YOU Yingrong,FAN Yingle,PANG Quan. Adhesion cell segmentation method based on distance transformation[J]. Computer Engineering and Applications,2005,41(20): 206-208. (in Chinese) doi: 10.3321/j.issn:1002-8331.2005.20.061 [25] LIN T Y, GOYAL P, GIRSHICK R, et al. Focal loss for dense object detection[C]//2017 IEEE International Conference on Computer Vision. Piscataway, US: IEEE, 2017: 2999-3007. -
点击查看大图
计量
- 文章访问数: 372
- HTML浏览量: 316
- PDF量: 36
- 被引次数: 0