Research On Remote Sensing Image Retrieval Based On Artificial Immune System

With the rapid development of remote sensing technology, remote sensing image data shows the trend of growth in geometric progression, large-capacity image databases appeared. How to retrieval the required images from the mass of remote sensing images quickly and accurately has become the highlight problem of current remote sensing technology. Lots of questions in image retrieval can be translated into optimization problems, such as optimal feature extraction, multi-feature weight assignment. But how to achieve and improve the automation and intelligent level have been a research focus and the difficult part. In this paper, a number of new attempts have been tried. The latest artificial intelligence technology-artificial immune system(AIS) has been introduce to the remote sensing image retrieval to solve the optimization problems, some new thought and methods have been put forward in remote sensing image retrieval. The major works that has been completed are listed as follows:(1) The history and status quo of AIS and Content Based Image Retrieval technology has been summarized, the similarity of them are compared in different levels, and the feasibility of combining them is studied.(2) A novel remote sensing image retrieval framework based on CLONALG is proposed. Based on the relevance feedback technology and immune mechanism, CLONALG is used to learn and memorize the user feedback image feature, the recognition of customers'semantic target for system is improved. Experimental results show that this approach can recognize the user feedback information efficiently and improve the retrieval accuracy.(3) Based on the clonal selection algorithm, an improved remote sensing image retrieval framework is promoted. This framework combined with fuzzy set theory, an unrelated antibody set is added into the model, and it is to remember the user's negative feedback images, so that the framework can make use of positive and negative feedback images simultaneously, the shortcomings of the original framework that only positive feedback can be studied is overcome. The experimental results show that compared with the original framework, this improved framework can improve the image retrieval accuracy.(4) A remote sensing image retrieval framework based on aiNet artificial immune network is presented. Because of the introduction of unrelated antibody set and the redundancy of antibody sets, the retrieval speed of the improved framework is decreased. In order to overcome this disadvantage, this model is proposed. Using the advantage of reducing redundancy, redundancy antibodies in the antibody set are removed, thereby the retrieval speed is enhanced.(5) A feature weight assignment method based on artificial immune recognition system (AIRS) is studied. The generalized learning and memory character of AIRS is used to learn the character of training samples, in order to determine the weight between the image features. Experimental results show that comparing to the traditional feature weight assignment method, this method can provide better feature weight, and improve the image retrieval accuracy.At last, the main work of this thesis is summarized and more research work that needs to be done is also brought forward.