Research On Hyperspectral Remote Sensing Image Retrieval Based On Cloud Computing

With the rapid development of 3S technology and information technology,the ability of human being's comprehensive observation to the earth's surface and the ability of processing,transmission and application on the surface information have been greatly ascend.However,with an expanding demand to hyperspectral data from the hyperspectral research department and the innovation of the acquisition method for hyperspectral data,the amount of hyperspectral data is in explosive surge.How to store the vast number of hyperspectral data reasonably and retrieve the user interested information has become a problem urgently to be solved in current remote sensing research field.Meanwhile,because the hyperspectral images have characteristics of multi-band and high spectral resolution,its processing algorithm of has the problem of high complexity and large demand for hardware resource.The cloud computing technology with scalable storage capacity can provide a better solution for hyperspectral data storage,what's more,its distributed computing ability can solve the single bottleneck problem of hyperspectral processing algorithm effectively.This paper mainly analyzes the current demand of content-based retrieval of hyperspectral image,combines with the distributed storage and parallel processing capability of cloud computing platform,and we have designed a system based on cloud computing for content retrieval of hyperspectral remote sensing images by spectral mixed pixel decomposition.The main contents include:(1)According to the background of hyperspectral image processing,this paper designs a hyperspectral image retrieval system based on cloud computing after analyzing the HDFS,Hadoop MapReduce programming model and Spark framework based on memory calculation and other related cloud computing technology.What's more,this paper gives the system architecture and the design of function module,including the massive hyperspectral remote sensing data distributed storage module based on HDFS,the remote sensing image data management module based on MySQL database,the hyperspectral image retrieval module based on mixed pixel decomposition and so on.(2)We designed a method based on mixed pixel decomposition for hyperspectral image retrieval,which can achive image content retrieval based on spectral characteristics.Comparing the characteristic by spectral angle matching method and extracting the position and component information of the target spectrum by abundance inversion algorithm,this method can extract the end member spectra in the hyperspectral image to be retrieved by the pure pixel index algorithm,as a result,the retrieval precision is high.(3)In this paper,a distributed parallel optimization method based on Spark is proposed.This method is aim for the pure pixel index algorithm and the abundance inversion algorithm of mixed pixel decomposition in the system.By taking advantage of the characteristic of the Spark,this paper optimizes the algorithm's logic and uses the broadcast mechanism of Spark to reduce the memory overhead of the executing node and optimizes the Shuffle data structure to reduce the network transmission overhead.On the basis of ensuring the correctness of mixed pixel decomposition,the parallel optimized algorithm achieves a good acceleration effect in a real hyperspectral image scene.