Research On The Technology Of Content-based Satellite Cloud Image Retrieval

Satellite cloud image is an important data for meteorological analysising. It can observe lagrearea and sampled with high frequency. There are variety weather systems contained in the historicalcloud images. It is helpful faor weather forecasting through searching the historical cloud imageswhich are similar to current cloud image. This dissertation designs the sallite cloud image retrievalsystem by the contented based image retrieval method,extracts the bottom imformation of the satellitecloud image.There are clouds,land and marine information contained in the satellite cloud image. But theeffective information is clouds part only. This dissertation used Fuzyy C-means algorithm for satelliteimage clustering to extract the clouds content from the nephograms. FCM is always converges to thelocal infinitesimal values.So PSO was introduced to FCM algorithm,PSO can find a globally optimalfuzzy segmentation,it avoids the sensitivities of basic FCM algorithm to initial values. In order toimprove the speed of the algorithm, introducing the shadow sets algorithm to FCM,which can removeboundary values and abnormal values. On the basis of extract the clouds content from the nephograms,this dissertation uses the gray histogram method to extract gray feature of the image. To avoid thismethod’s problem that it lost position information of the image, this dissertation block the satellitecloud image,respectively extract the gray feature of the sub-block. The extraction of texture featuresusing grey level co-occurrence matrix method, in order to improve the algorithm speed, adopting theimproved one-dimensional co-occurrence matrix method to calculate the texture characteristic value.This paper extracts the shape feature from the main cloud in the nephograms,then using HU shapemoment calculating the shape feature values. In order to regulate the shape of the nephograms, we useimage morphological opening operation to remove the burr of isolated point and boundary,and usemorphological closing operation to remove the holes in the cloud image.Gray feature, texture feature and shape feature reflect the content of the satellite cloud imagefrom a single aspect. This paper set the weight for each feature then added the wighted feature valuesas the comprehensive feature. According to the design of the feature extraction algorithms, this paperdesigns a content-based satellite cloud image retrieval system. The system has a friendly interface,including the user interface module, image feature extraction module, database access module andimage retrieval feedback module. Proved by the experiments that the system has good retrievalperformance.