Font Size: a A A

The Application Of Cluster Analysis In Satellite Cloud Image Segmentation

Posted on:2017-02-24Degree:MasterType:Thesis
Country:ChinaCandidate:M WangFull Text:PDF
GTID:2308330482489823Subject:Computer technology
Abstract/Summary:PDF Full Text Request
Cluster analysis is one of the important research contents in the field of data mining, which absorbs the knowledge and thought of the fields of computer science, statistics, mathematics and so on. For those datasets having no obvious characteristics, clustering analysis can dig out the “dissimilarity” or “similarity” and points into several subsets in accordance with this feature, so that objects subset have little difference within the same and larger difference between different subsets. In recent years, the theory of cluster analysis has diversified development, and it has been extended to the field of image processing, meteorological analysis, medicine and on.Image processing is a new direction in the application of clustering analysis, and image segmentation is a key branch in the research of this direction. There are a huge number of satellite cloud images in the field of meteorological analysis. How to process satellite images and extract valid information of cloud by computer technology, how to analyze and forecast the weather such as thunderstorm and rainfall by using these information, are difficult issues to be resolved for meteorological and computer information processing researchers. This paper use clustering analysis in satellite image segmentation, achieve qualitative and quantitative identification of cloud segmentation, improve the efficiency and accuracy of thunderstorms weather, and expand the cluster analysis of meteorological satellite applications.The main contents of this study are:(1) Using the fuzzy c-means algorithm to execute the first satellite image segmentation and realize the separation of the cloud and land. The key of satellite image segmentation is achieving the segmentation of the cloud region and the land area, according to the “similarity” between the pixels in the multispectral image. The traditional threshold method is a “hard” method. Some reasons, such as the difficulty to obtain a suitable threshold and environmental factor, led to its decline in accuracy. Considering the complexity and fuzziness of cloud image, this paper studies an algorithm based on FCM. This method can effectively solve the problem of the uncertain segmentation and segmentation accuracy of cloud image. Experiments shows that the method proposed by this paper can better reflect the dynamic change characteristics of the cloud and the surface gray threshold, and get results which are more close to the actual effect.(2) Processing the second segmentation of the cloud image by using clustering algorithm based on density to achieve cloud recognition. Cloud recognition in cloud image means identifying the cloud system or cloud, according to the spatial distribution of the relationship between the pixels. Based on the “density connectivity” between pixels, combined with ideological image region segmentation, this paper propose a cloud identification algorithm based on the density clustering DBSCAN. Using the “high-density connectivity” within the same pixels cloud and the “low-density connectivity” between different clouds, the method extracts independent from the image. Experiments shows that this method is simple, high efficiency, get the best cloud segmentation by adjusting relevant parameters of neighborhood.(3) At last, the two algorithms are applied to the data preprocessing of thunderstorm clouds recognition, and this paper puts forward a method of thunderstorm cloud recognition based on the two algorithms. Based on cloud analysis, combining the advantages of spectrum threshold identification and texture feature recognition, this method can identify the thunderstorm clouds quickly and accurately. Experimental results show that cluster analysis of meteorological satellite cloud image processing has high value in research and application.
Keywords/Search Tags:Cluster analysis, FCM, DBSCAN, Thunderstorm clouds, Image segmentation, Satellite image
PDF Full Text Request
Related items