Click for new scientific resources and news about Corona[COVID-19]

Paper Information

Journal:   REMOTE SENSING & GIS   SUMMER 2010 , Volume 2 , Number 2; Page(s) 1 To 16.
 
Paper: 

IDENTIFICATION OF THE OPTIMUM SEGMENTATION SCALE FOR CLOUDS OBJECT-ORIENTED CLASSIFICATION, USING NOAA/AVHRR IMAGES

 
 
Author(s):  AZARI H.*, MATKAN A.A., SHAKIBA A.
 
* DEPARTMENT OF GIS AND REMOTE SENSING, SHAHID BEHESHTI UNIVERSITY, VELENJAK ST., TEHRAN, IRAN
 
Abstract: 

Precipitation rate and amount measurements are among the flood warning methods which have been suggested by remote sensing in recent years. Cloud type identification and classification, as basic principles of precipitation estimation methods, are usually performed using visual interpretation of satellite images. In these studies only cloud brightness temperature and albedo are used for cloud classification, while texture and shape of clouds are effective properties in cloud type detection as well. Textures and shapes of clouds are ignored in pixel base classifications. So object- oriented classification technique is a suitable approach. In this technique, in addition to cloud brightness temperature and albedo, textures and shapes are the major parameters. Object-orient classification method, despite its benefits, depends on segmentation accuracy. The accuracy of segmentation is scale dependent too. Therefore, optimum segmentation scale is resulting to higher accuracy of object oriented classification. In this study two NOAA/AVHRR images in two consecutive cloudy days in August 2005 are used. In the first step, additional information included brightness temperature of band 3 and 4 and cloud height produced from NOAA/AVHRR images that used in image segmentation, and bi-spectral method has been employed for training region selection. Then the negative impacts of under-segmentation errors on the potential accuracy of object-based classification were quantified by developing a new segmentation accuracy measure. In this step, scale evaluation was performed with quantifying overall effect relative to features and units in 31 scales of segmentation.
The results based on a NOAA/AVHRR satellite images were the same and indicate that: (1): cloud segmentation accuracies decrease with increasing segmentation scales, and (2) the negative impacts of under-segmentation errors in cloud segmentation become significantly large at large scales. Hence, the finest scale for cloud segmentation has been defined 50 as in this scale the overall accuracy of classification was 90.5% in cloud object oriented classification.

 
Keyword(s): IMAGE REGISTRATION, SIFT, HARRIS OPERATOR, STABILITY, DISTINCTIVENESS, SPATIAL DISTRIBUTION
 
References: 
  • ندارد
 
  Persian Abstract Yearly Visit 70
 
Latest on Blog
Enter SID Blog