Pattern Recognition Technology And Its Application In Meteorological Research

The pattern recognition as a quickly developing new theory and technique plays important roles in many areas such as image processing and remote sensing etc,. In this paper we studied the textural features extraction , remote sensing images classification and BP neural network techniques and their applications in the meteorological problems such as recognition of the cloud cluster feature , cloud-drift wind retrieval and heavy rain process analysis etc.To the question of the low precise recognition of satellite images by using spectral features, the proposed approach assumes to perform a multiple analysis based on an advisable decision-making model by first developing a mixed pixel model which was based on the textural features of images, and then improving the recognition intelligence. This scheme was finally tested by a typical meso-scale severe rainstorm process which happened on the middle - lower parts of the Yangtze River in China during the period of July 4th -5th, 2003.In the first part of this paper, some typical schemes in analyzing textual feature are introduced. Among them the Gray Level Co-occurrence Matrix (GLCM) and Gray Gradient Co-occurrence Matrix (GGCM) methods ,which attributed to the statistic textural analysis scheme were then chosen to extract the textural features of five kind areas on satellite images. In the second part the principle of classification and BP neural network were introduced. Combined with textural features, the improved BP neural network successfully performed on the classification of the satellite images. Based on the classification results, the correlation coefficient method introduced in the last part of this paper was then used to retrieve cloud-drift wind from the sequent images. By examining on the rainstorm process happened during the period of July 4th -5 , 2003, the rationality of using textural feature as the whole decision-making model's foundation was successfully testified. Furthermore, the improved cloud-drift wind retrieval technique which based on this decision-making model also showed a remarkable performance in analyzing a meso-scale weather system. The weather system on July 5th 08BST depicted by the cloud-drift winds could be distinctively seen an anticyclone with divergence in the north and convergence in the south which was much helpful in rationally explaining the real weather process combined with some traditional methods.