设为首页
加入收藏
Lightning Channel Image Recognition Based on Line Support Region
-
摘要: 提出了一种基于线支持区域的闪电通道识别算法 (LLSR)，该算法首先应用对比度拉伸和高斯匹配滤波方法对闪电通道图像进行预处理，以增强闪电通道的对比度；然后自动检测出包含闪电通道的线支持区域，并用最小外接矩形包含这些区域；最后在各个矩形区域内分别使用最大类间方差Otsu阈值法进行分割，得到闪电通道识别结果。试验结果表明:LLSR具有良好的局部特性和自适应性，它不仅能自动提取低对比度闪电图像的通道，还能自动提取具有复杂背景闪电图像的通道，自动提取结果在视觉上与人眼观测结果一致。且定量评估结果表明:LLSR相比传统算法具有更好的分割精度。Abstract: Lightning channel coordinates in digital images are often manually extracted to analyze the development and morphology of lightning channels, but this method is not efficient and its result is often subjective. Therefore, more and more researchers start to investigate approaches to recognize lightning channel information automatically. In general, lightning channel images are complex and diverse because of low contrast, occlusion of clouds, and interference of other environmental factors, so most traditional lightning channel segmentation algorithms do not work well for such lightning images.A new lightning channel segmentation algorithm named LLSR is brought forward based on line support regions. First, Gauss matched filtering and contrast stretching method are applied to enhance the contrast of lightning channels, according to the gray distribution characteristics of cross section of lightning channels. Second, line support regions, which include lightning channels within a minimum enclosing rectangle, are extracted as foreground area by a line segment detection method. In addition, line support regions are expanded in both the main direction and its perpendicular direction. In general, a line support region contains a segment of a lightning channel. Furthermore, it has better contrast between lightning channel and background. Finally, Otsu thresholding method is applied in each line support region to extract lightning channels, because the gray level distribution of each line support region is bimodal. Therefore, lightning channels are segmented from complicated background.A dataset including various types of lightning channel images, are constructed and manually marked to evaluate the proposed algorithm LLSR. Compared with traditional algorithms, global thresholding method (GThres), local thresholding method (LThres), and Canny thresholding method (CThres), the proposed LLSR has higher precision for lightning channel segmentation, and it obtains a better balance between recall rate and false positive rate. Besides, experiment results show that traditional algorithms are not robust enough for all types of lightning images, but the new method demonstrates better generality. LLSR can recognize not only the lightning channels with low contrast but also the lightning channels with complicated background, and the segmentation result is visually consistent with human eyes.
-
Key words:
- lightning image;
- channel identify;
- line support region;
- Otsu;
- thresholding algorithm
-
图 1 图像预处理示例
(a) 闪电通道原图，(b) 对比度拉伸图像，(c) 高斯匹配滤波图像
Fig. 1 Example of image preprocessing results
(a) the original lightning channel image, (b) the image after contrast stretching, (c) the image after Gauss matched filtering
图 3 闪电通道线支持区域矩形扩展示例
(a) 扩展前，(b) 扩展后
Fig. 3 Examples of the rectangular expansion of lightning channel line support region
(a) before expansion, (b) after expansion
图 4 闪电通道图像中三个矩形区域及其对应的灰度直方图
(a) 闪电通道图像中3个区域示例, (b) 区域1的灰度直方图, (c) 区域2的灰度直方图, (d) 区域3的灰度直方图
Fig. 4 Three rectangular regions and their corresponding gray histogram
(a) example of three rectangular regions in lightning channel image, (b) gray histogram of region 1, (c) gray histogram of region 2, (d) gray histogram of region 3
图 5 闪电图像及其对应的人工标注图像示例
Fig. 5 Examples of lightning images and their corresponding manually marked images
表 1 不同σ取值对应的LLSR性能
Table 1 Performance of LLSR with different σ
σ Rrecall Ppre Fmeasure 0.7 0.5439 0.8182 0.6397 1.6 0.7347 0.7621 0.7352 2.3 0.7462 0.6655 0.6777 
下载: 导出CSV
表 2 不同τ取值对应的性能表
Table 2 Performance of LLSR with different τ
τ/(°) Rrecall Ppre Fmeasure 14 0.6952 0.7635 0.7101 15 0.7042 0.7566 0.7116 18 0.7224 0.7390 0.7118 20 0.7290 0.7303 0.7106
下载: 导出CSV
表 3 不同的算法性能表
Table 3 Performance for different algorithms
算法 Rrecall Ppre Fmeasure GThres 0.6252 0.6724 0.6227 LThres 0.8709 0.3090 0.4455 CThres 0.8834 0.2243 0.3534 LLSR 0.7224 0.7390 0.7118
下载: 导出CSV
-
[1] 王道洪.雷电与人工引雷.上海:上海交通大学出版社, 2000:1-12. http://www.cnki.com.cn/Article/CJFDTOTAL-SYQY201603027.htm [2] 刘欣生, 肖庆复.人工引发雷电的静态摄影及特性分析.高原气象, 1998, 17(1):106-110. http://www.cnki.com.cn/Article/CJFDTOTAL-GYQX801.011.htm [3] 张义军, 孟青, 马明, 等.闪电探测技术发展和资料应用.应用气象学报, 2006, 17(5):611-620. doi: 10.11898/1001-7313.20060504 [4] 李俊, 张义军, 吕伟涛, 等.一次多回击自然闪电的高速摄像观测.应用气象学报, 2008, 19(4):401-411. doi: 10.11898/1001-7313.20080403 [5] Lu W, Chen L, Ma Y, et al.Lightning attachment process involving connection of the downward negative leader to the lateral surface of the upward connecting leader.Geophys Res Lett, 2013, 40(20):5531-5535. doi: 10.1002/2013GL058060 [6] 吕伟涛, 张义军, 周秀骥, 等.火箭触发闪电通道的亮度特征分析.气象学报, 2008, 65(6):983-993. http://www.cnki.com.cn/Article/CJFDTOTAL-QXXB200706015.htm [7] Zhou E, Lu W, Zhang Y, et al.Correlation analysis between the channel current and luminosity of initial continuous and continuing current processes in an artificially triggered lightning flash.Atmos Res, 2013, 129:79-89. http://www.sciencedirect.com/science/article/pii/S0169809512003468 [8] 张义军, 周秀骥.雷电研究的回顾和进展.应用气象学报, 2006, 17(6):829-834. doi: 10.11898/1001-7313.20060619 [9] 谢盟, 张阳, 张义军, 等.两种类型M分量物理特征和机制对比.应用气象学报, 2015, 26(4):451-459. doi: 10.11898/1001-7313.20150407 [10] 张阳, 张义军, 孟青, 等.北京地区正地闪时间分布及波形特征.应用气象学报, 2010, 21(4):442-449. doi: 10.11898/1001-7313.20100407 [11] Otsu N.A threshold selection method from gray-level histograms.IEEE Transactions on Systems Man & Cybernetics, 1979, 9(1):62-66. http://ieeexplore.ieee.org/document/4310076/?reload=true&arnumber=4310076 [12] 周恩伟.触发闪电放电过程的光电同步观测与分析.合肥:中国科学技术大学, 2010. http://www.cnki.com.cn/Article/CJFDTOTAL-SYQY201603027.htm [13] 杨欣怡, 吕伟涛, 杨俊, 等.3种阈值方法在闪电通道图像识别中的应用.应用气象学报, 2014, 4(4):427-435. doi: 10.11898/1001-7313.20140405 [14] von Gioi R G, Jakubowicz J.LSD:A fast line segment detector with a false detection control.IEEE Trans Pattern Anal Mach Intell.IEEE Transactions on Software Engineering, 2010, 32(4):722-732. http://ieeexplore.ieee.org/document/4731268/ [15] von Gioi R G, Jakubowicz J, Morel J M, et al.LSD:A line segment detector.Image Processing on Line, 2012, 2:35-55. doi: 10.5201/ipol [16] Chaudhuri S, Chatterjee S, Katz N, et al.Detection of blod vessels in retinal images using two-dimensional matched filters.IEEE Trans Medical Imaging, 1989, 8(3):263-269. doi: 10.1109/42.34715 [17] 王晓红, 赵于前, 廖苗, 等.基于多尺度2D Gabor小波的视网膜血管自动分割.自动化学报, 2015, 41(5):970-980. http://www.cnki.com.cn/Article/CJFDTOTAL-MOTO201505011.htm -
计量
- 摘要浏览量: 3123
- HTML全文浏览量: 1043
- PDF下载量: 257
- 被引次数: 0
