Improvement and Application of Artificial Neural Networks to Cloud Classification
-
摘要: 采用2005—2009年FY-2C静止气象卫星可见光和红外自旋扫描辐射计的红外1(IR1)、红外2(IR2) 和水汽 (WV) 亮温资料,选取2449个云分类样本。设计两层嵌套的前向传递后向反馈 (BP) 人工神经网络模型,第1层网络选取IR1,IR2,WV亮温及IR1与WV亮温差和IR2与WV亮温差5个特征量,第2层网络选取特征量IR1与IR2亮温差,两层网络都采用一层隐含层且带有附加动量法的简单网络,降低了网络的冗余度。误差分析表明:嵌套BP人工神经网络模型的分类准确率在中云和薄卷云这两类上分别提高了42.7%和11.3%,整个分类模型的平均平方误差和标准化平均平方误差分别降低了6.1%和44.7%,相关系数提高了3.4%。通过3个个例的对比分析发现,嵌套模型的分类结果比传统模型的分类结果更合理,特别是在中低云和薄卷云的云量和位置分辨能力上有了较大提高。Abstract: 2449 cloud classification samples are artificially selected from infrared channel 1, infrared channel 2 and water vapor channel of VISSR on FY-2C geostationary meteorological satellite during 2005—2009. Different linear combination of three channels are selected as feature values, which are brightness temperature of IR1, IR2, WV, and brightness temperature difference of IR1 to IR2, IR1 to WV, IR2 to WV. According to statistical theory, the sample probability distribution is assumed to help to remove some apparent unreasonable data such as outliers, and to understand cloud normal features better. It is found that the brightness temperature difference of IR1 to IR2 is most sensitive to the amount of thin cirrus cloud in 6 selected values. On the other hand, the error of the BP neural network model mostly comes from the contradiction of this feature too. A nested BP artificial neural network model is designed, and it's composed of two layers. The first layer includes five features of brightness temperature of IR1, IR2, WV, and brightness temperature difference of IR1 to WV, IR2 to WV that are used to classify each pixel to one of four categories such as clear, mixed cloud, thick cirrus cloud and strong convective cloud. And the second layer includes just one feature, brightness temperature difference of IR1 to IR2 that are used to classify mixed cloud to low-level cloud, mid-level cloud or thin cirrus cloud. Finally, every pixel is classified into one of total 6 categories corresponding to each color. Both layers adopt a BP neural network, the most widely used algorithm for generating classifiers, with one hidden layer and the additional momentum method, not only accelerating the training speed, but also reducing the redundancy of the networks.Error analysis shows that the accuracy rates of the nested BP artificial neural network for types of mid-level cloud and thin cirrus cloud have increased by 42.6% and 11.3%, respectively. The mean square error and normalized mean square error of the whole classification model have decreased by 6.1% and 44.7% with the correlation coefficient increasing by 3.4%. By comparison of the classification results from 3 tests of tropical, subtropical areas and tropical cyclone, it shows that the nested model identifies thin cirrus cloud more accurately than the traditional model. Therefore, the results of the nested model are more reasonable than the traditional model.
-
表 1 VISSR通道特征
Table 1 Specifications of VISSR channels
通道
标号通道
名称光谱
范围/μm空间
分辨率/km1 IR1 10.3~11.3 5 2 IR2 11.5~12.5 5 3 WV 6.3~7.6 5 4 IR4 3.5~4.0 5 5 VIS 0.55~0.90 1.25 表 2 云分类和样本信息
Table 2 The information of classes and samples
分类 样本数 备注 强对流云 515 积雨云和发展旺盛的对流云 厚卷云 244 较厚的密卷云和积雨云的卷云毡 薄卷云 232 毛卷云、卷积云和较薄的密卷云 中云 214 高层云、高积云 低云 404 层积云、层云、雨层云 海洋 435 晴空下的海洋 陆地 405 晴空下的陆地 表 3 云分类特征量表
Table 3 The selected features of the cloud classification
云分类特征 描述 IR1亮温 红外1通道的云顶亮温 IR2亮温 红外2通道的云顶亮温 WV亮温 水汽通道的云顶亮温 IR1与IR2亮温差 红外1和红外2通道的分裂窗差 IR1与WV亮温差 红外1和水汽通道的亮温差 IR2与WV亮温差 红外2和水汽通道的亮温差 表 4 测试样本准确率 (单位:%)
Table 4 The accuracy rates of test samples (unit: %)
云分类 晴空 低云 中云 薄卷云 厚卷云 对流云 晴空 97.379 2.620 0 0 0 0 低云 3.465 92.574 3.960 0 0 0 中云 0 31.250 67.708 0 1.042 0 薄卷云 0 9.538 5.660 82.915 1.887 0 厚卷云 0 0 0.806 0 87.903 11.290 对流云 0 0 0 0 6.179 93.820 表 5 云分类模型参数
Table 5 The parameters of each cloud classifier
网络类型 输出层 隐含层 附加动量 步长 层数 神经元个数 步长 传统BP网络 0.1 1 16 1 0.7 嵌套网络第1层 0.1 1 9 1 0.7 嵌套网络第2层 0.1 1 6 1 0.7 表 6 准确率对比表
Table 6 Accuracy rates of the traditional and nested models
网络类型 EMS ENMS r 传统网络 0.0479 0.1133 0.9385 嵌套网络 0.045 0.0627 0.9708 -
[1] Shenk W E, Neff R A, Holub R J. A multispectral cloud type identification method developed for tropical ocean areas with Nimbus-3 MRIR measurements (Medium Resolution Infrared Radiometer). Mon Wea Rev, 1976, 104(3): 284-291. doi: 10.1175/1520-0493(1976)104<0284:AMCTIM>2.0.CO;2 [2] 张承福.神经网络系统.力学进展, 1988, 18(2): 145-160. http://cdmd.cnki.com.cn/Article/CDMD-10424-2004133649.htm [3] 张承福.人工神经网络在天气预报中的应用研究.气象, 1994, 20(6): 43-47. doi: 10.7519/j.issn.1000-0526.1994.06.008 [4] 蔡煜东, 汪列, 姚林声.人工增雨有利条件的神经网络识别与估价.应用气象学报, 1994, 5(1): 125-128. http://qikan.camscma.cn/jams/ch/reader/view_abstract.aspx?file_no=19940121&flag=1 [5] 张韧, 蒋国荣, 余志豪, 等.利用神经网络计算方法建立太平洋副高活动的预报模型.应用气象学报, 2000, 11(4): 474-483. http://qikan.camscma.cn/jams/ch/reader/view_abstract.aspx?file_no=20000469&flag=1 [6] 师春香, 卢乃锰, 张文建.卫星面降水估计人工神经网络方法.气候与环境研究, 2001, 6(4): 467-472. http://www.cnki.com.cn/Article/CJFDTOTAL-QHYH200104010.htm [7] 梁益同, 胡江林. NOAA卫星图像水体信息神经网络识别方法的探讨.应用气象学报, 2001, 12(1): 85-90. http://qikan.camscma.cn/jams/ch/reader/view_abstract.aspx?file_no=20010111&flag=1 [8] 熊秋芬, 胡江林, 夏军.神经网络方法在静止卫星多通道资料估算降水中的应用.气象, 2002, 28(9): 17-21. doi: 10.7519/j.issn.1000-0526.2002.09.004 [9] 樊晓平, 彭展, 杨胜跃, 等.基于多层前馈型人工神经网络的抑郁症分类系统研究.计算机工程与应用, 2004, 40(13): 205-208. doi: 10.3321/j.issn:1002-8331.2004.13.067 [10] 金龙, 吴建生, 林开平, 等.基于遗传算法的神经网络短期气候预测模型.高原气象, 2005, 24(6): 981-987. http://cpfd.cnki.com.cn/Article/CPFDTOTAL-ZGQX200312005034.htm [11] 邵利民, 傅刚, 曹祥村, 等. BP神经网络在台风路径预报中的应用.自然灾害学报, 2009, 18(6): 104-111. http://www.cnki.com.cn/Article/CJFDTOTAL-ZRZH200906018.htm [12] 黄小刚, 费建芳, 陈佩燕.利用神经网络方法建立热带气旋强度预报模型.应用气象学报, 2009, 20(6): 699-705. doi: 10.11898/1001-7313.20090607 [13] 闵晶晶, 孙景荣, 刘还珠, 等.一种改进的BP算法及在降水预报中的应用.应用气象学报, 2010, 21(1): 55-62. doi: 10.11898/1001-7313.20100107 [14] Lee J, Weger R C, Sengupta S K, et al. A neural network approach to cloud classification. IEEE Transactions on Geoscience and Remote Sensing, 1990, 28(5): 846-855. doi: 10.1109/36.58972 [15] Bankert R L. Cloud classification of a AVHRR imagery in maritime regions using a probabilistic neural network. J Appl Meteor, 1994, 33(8): 909-918. doi: 10.1175/1520-0450(1994)033<0909:CCOAII>2.0.CO;2 [16] Baum B A, Tovinkere V, Titlow J, et al. Automated cloud classification of global AVHRR data using a fuzzy logic approach. J Appl Meteor, 1997, 36(11): 1519-1540. doi: 10.1175/1520-0450(1997)036<1519:ACCOGA>2.0.CO;2 [17] Tian B, Shaikh M A, Azimi-Sadjadi M R, et al. A study of cloud classification with neural networks using spectral and textural features. IEEE Transactions on Neural Networks, 1999, 10(1): 138-151. doi: 10.1109/72.737500 [18] Liu Y, Xia J, Shi C X, et al. An improved cloud classification algorithm for China's FY-2C multi-channel images using artificial neural network. Sensors, 2009, 9(7): 5558-5579. doi: 10.3390/s90705558 [19] 师春香, 瞿建华.用神经网络方法对NOAA-AVHRR资料进行云客观分类.气象学报, 2002, 60(2): 250-256. doi: 10.11676/qxxb2002.031 [20] 师春香, 吴蓉璋, 项续康.多阈值和神经网络卫星云图云系自动分割试验.应用气象学报, 2001, 12(1): 70-78. http://qikan.camscma.cn/jams/ch/reader/view_abstract.aspx?file_no=20010108&flag=1 [21] 国家卫星气象中心. 风云二号C卫星业务产品释用手册. 2005: 61-66. [22] 金龙.人工神经网络技术发展及在大气科学领域的应用.气象科技, 2004, 32(6): 385-392. http://www.cnki.com.cn/Article/CJFDTOTAL-QXKJ200406000.htm [23] Hsieh W W, Tang B. Applying neural network models to prediction and data analysis in meteorology and oceanography. Bull Amer Meteor Soc, 1998, 79(9): 1855-1870. doi: 10.1175/1520-0477(1998)079<1855:ANNMTP>2.0.CO;2 [24] 金龙, 况雪源, 黄海洪, 等.人工神经网络预报模型的过拟合研究.气象科技, 2004, 62(1): 62-70. http://www.cnki.com.cn/Article/CJFDTOTAL-QXXB200401006.htm [25] 俞善贤, 钟元, 滕卫平.热带气旋路径人工神经元预报方法对比试验研究.热带气象学报, 2004, 20(5): 523-529. http://cpfd.cnki.com.cn/Article/CPFDTOTAL-ZGQX200312006032.htm [26] 陈渭民, 夏浣清, 陈光宇.卫星气象学.北京:气象出版社, 2003. [27] 胡金滨, 唐旭清.人工神经网络的BP算法及其应用.信息技术, 2004, 4(4): 1-4. http://cdmd.cnki.com.cn/Article/CDMD-10220-2002073238.htm [28] 郁凡, 陈渭民.双光谱云图的云分类探讨.南京气象学院学报, 1994, 17(1):117-124. http://www.cnki.com.cn/Article/CJFDTOTAL-NJQX199201012.htm [29] 吕砚山, 赵正琦. BP神经网络的优化及应用研究.北京化工大学学报, 2001, 28(1): 67-69. http://cdmd.cnki.com.cn/Article/CDMD-10157-1012302301.htm [30] 朱乾根, 林锦瑞, 寿绍文, 等.天气学原理和方法.北京:气象出版社, 2000: 515-521.