運用機器學習辨識雲的種類

Cloud Types Identification by Using the Machine Learning

陳維楨(Wei-Zhen Chen)

指導教授 ： 黃健興

義守大學/電機資訊學院/資訊工程學系/碩士(2015年)

https://doi.org/10.6343/ISU.2015.00189

摘要

在天氣觀測中，日照、雨量、風與溫度等資訊可藉由儀器觀測獲得，但雲的種類與雲的覆蓋量則需要使用肉眼觀測，本研究之目的是藉由攝影機所拍攝的影像去估測雲量與辨識雲族。
在估測雲量的部分，取得影像後給予閥值，將其分割為雲和藍天黑白兩部分，然而，不同天氣的情況，我們所提出的分割雲的方法是最有效的，當影像的標準差在0.04到0.1之間時以R/B平均值作為閥值，其他情況使用色彩空間R與B的差值作為選取條件，當差值介於30到-30之間，判斷此像素點為雲的部分。
在辨識雲族的部分，使用區域二元編碼(LBP)取得紋理特徵，也運用R/B值正規化後之影像強度作為色彩特徵，接著使用主成分分析(PCA)找出色彩與紋理特徵分佈維度，將降低維度的特徵帶入支撐向量機(SVM)分類成高雲族、中雲族、低雲族與直展雲族。
混合R/B平均值作為閥值的方法與使用色彩空間R與B的差值判斷出雲的方法，比單一種做法所分割出來的結果更好，能有效的分割出雲的部分，在雲族分類的部分，SVM分類測試的結果是有效的，正確率為96.7%。

關鍵字

雲的種類 ； 區塊二元局部編碼 ； 主成分分析 ； 支撐向量機

並列摘要

For observation of weather, we can observe the information of sunshine, rain, temperature and wind by instruments, but the amount of cloud and the cloud types must be obtained by human eyes. The purpose of this thesis is to measure the amount of cloud and to identify the cloud types by the technique of computer vision and machine learning.
In order to estimate of amount of cloud, the image segmentation is adopted to part the cloud and the sky as white and black partition by thresholding. However, there are many situation of weather, we propose the method to choice the best criteria for segmentation. While the standard deviation of image between 0.04 and 0.1, the average of R/B values is adopted as threshold, otherwise the intensity difference of the red and blue color between 30 and -30 is judged as cloud part.
For the identification of cloud types, the Local Binary Pattern (LBP) is adopted to obtain the texture feature vector, and the ratio of red to blue intensity in image is adopted to obtain the color feature vector after normalized. Then, the algorithm of principal components analysis (PCA) is used to reduce the dimension of the feature vector and then the support vector machine is adopted to classify the types of cloud.
There are two major contributions in this thesis. Firstly, our cloud segmentation combines two criteria which is better than the binarization using R/B=0.8 as the threshold and the judgment by the difference between red and blue intensity, and secondly, we speed up the classification of cloud using support vector machine by decision tree. The accuracy of classification is up to 96.7%.

並列關鍵字

Cloud Types ； LBP ； PCA ； SVM

參考文獻

[2] C. N. Long, J. M. Sabburg, J. Calbó, and D. Pagès, “Retrieving cloud characteristics from ground-based daytime color all-sky images,” J. Atmos. Ocean. Technol., vol. 23, no. 5, pp. 633–652, May 2006.

[3]A. Heinle, A. Macke, and A. Srivastav, “Automatic cloud Classification of whole sky images,” Atmos. Meas. Technol., vol. 3, no. 1, pp. 557–567, May 2010.

[5]S. L. M. Neto, A. V. Wangenheim, E. B. Pereira, and E. Comunello, “The use of Euclidean geometric distance on RGB color space for the classification of sky and cloud patterns,” J. Atmos. Ocean. Technol., vol. 27, no. 7, pp. 1504–1517, Sep. 2010.

[6]Shuang Liu, Linbo Zhang, Zhong Zhang, Chunheng Wang, an Baihua Xiao“Automatic Cloud Detection for All-Sky Images Using Superpixel Segmentation,” IEEE Geoscience And Remote Sensing Letters, Vol. 12, No. 2, February 2015

[7]Qing Zhang and Chunxia Xiao, “Cloud Detection of RGB Color Aerial Photographs by Progressive Refinement Scheme,” IEEE Transactions On Geoscience And Remote Sensing, Vol. 52, No. 11, Pp. 7267–7269, November 2014