Study On Content-based Cloud Image Retrieval Technology

The rapid development of meteorological satellite technology has brought about the explosive growth of meteorological cloud image data.The traditional cloud retrieval method,which relies on artificial annotation text retrieval,is unable to meet the meteorological needs.The content-based image retrieval(CBIR)technology,which is used to managing massive image data,shows a very efficient retrieval performance.Based on the analysis of the characteristics of satellite cloud images,this paper has carried out the key technology of content-based cloud image retrieval based on CBIR technology.The main works are as follows:(1)Research on Cloud Information Extraction Based on Fractional Darwinian Particle Swarm Optimization Algorithm(FODPSO)and FCM Clustering Algorithm.When the traditional FCM clustering method is used to extract the cloud feature,the convergence of the algorithm is susceptible to the initial clustering center,and the local optimal solution is often got.FODPSO uses the natural selection mechanism to find the optimal solution globally,which can avoid the local optimal value,and optimize the fuzzy C-mean initial clustering center by using the hierarchical Darwinian particle swarm optimization algorithm with very good global optimization performance.And the fuzzy C-means clustering algorithm is used to improve the information extraction of the clouds,and then,the preparation of the effective feature of the next stage is prepared.(2)Research on the Fusion of Multi-channel Satellite Cloud Image.The cloud images obtained by different channels contain different weather characteristics.The fusion of multi-channel cloud images can be used to form a cloud map with multiple weather feature information,and this cloud map can be used to search for a more accurate historical similarity cloud image.In this paper,a method that uses NSST and adaptive PCNN for the fusion of infrared and visible satellite images is proposed.The experimental results show that,the fusion cloud image has more obvious features,such as rich edge and texture detail information with higher clarity,and also contains more meteorological information.Using this fusion cloud for cloud image retrieval,a better performance than that of a single type of cloud can be obtained.(3)Research on Feature Extraction of "Content" in Accurate and Efficient Representation.The fundamental works of Content-based cloud image retrieval technology is to extract effective features,which can be expressed as the "content" of cloud image.According to the inherent characteristics of the cloud,three reliable underlying features such as gray value,texture and shape of cloud image,are used as the "content" of cloud,The gray feature is extracted by the gray histogram method of the cloud image;texture feature is extracted by the LTrP operator with properties of anti-noise and invariance of gray-scale translation;shape feature is extracted by Krawtchouk moment with invariance of scale,rotation and translation.(4)Research on Multi-feature Decision Fusion of Cloud Image Retrieval.In the process of multi-feature decision fusion for retrieval,the choice of feature weight directly affects the accuracy of image retrieval.The traditional method that manually allocates different feature weights for decision fusion requires a lot of experiments to find a better search result.With the increasing types of fusion characteristics,the determination of artificial weight will become more and more inefficient.In this paper,the weight of each feature is determined adaptively according to the area under the similarity score curve of retrieval results by different feature.Experiment results show that,the size of the feature weight is negatively correlated with the area under the similarity measure,better feature can possess smaller area of score curve,and worse feature possesses larger area of score curve.