Ground Nephogram Classification Based On Textural Feature Extraction And Analysis

The ground-based observation is an essential method to get cloud parameters. It is important for scientific research and application such as objective recognition, air traffic control and weather forecast. Now, ground nephogram observation mainly relies on investigators by observing instruments to make judgments, the results are vulnerable to human factors, lack of objectivity and accuracy. After studied the development of ground nephogram classification technology and its application, this paper set up an experimental system using texture feature extraction, minimum distance classifiers and BP Neural Network (BPNN) to achieve the goal of ground nephogram classification.The main work of this paper were the followings:After in-depth study of the ground nephogram classification theory, in this paper the minimum distance classifiers and BPNN classifiers were introduced to classify ground nephogram.Based on in-depth study of the theory of the digital image processing, we pay attention to the image preprocessing, such as grey level transformation, histogram equalization and so on. Through this work, the quality of the cloud image was improved and the cloud texture characteristics were highlighted which created favorable conditions to the cloud texture feature extraction. After in-depth study of the feature of ground nephogram, Texture analysis and description, the gray level co-occurrence matrix (GLCM) was introduced to extract cloud texture feature parameters. It is important to analyze and filter these data in order to ensure the feature parameters can reflect texture characteristics of the clouds and the clouds can be distinguished between different samples.After in-depth study of the minimum distance and BPNN classifiers, this paper designed and implemented these two kinds of classifier; through this work this paper got and analyzed the results.A ground nephogram processing system and classification system were designed and implemented to achieve man-machine interface.This paper tested well-known texture feature extraction approaches for automatically recognizing Cumulus humilis, Nimbostratus, Altocumulus translucidus and Cirrus densus. For the four-class problem using BPNN and minimum distance classifiers, recognition rates were 91.67% and 87.50% obtained on the test data. It could find that the BPNN had a better result on the ground nephogram and also showed that the proposed method was effective and feasible.