| With the rapid development of modern science and technology,hyperspectral imaging technology may be in some high-tech areas,or in some demanding areaes about data accuracy,detail or deeper field of information,will gradually replace the ordinary camera shooting technology.Hyper-spectral image processing technology as one of the key directions,will gradually become a popular research field,attracting more and more experts and scholars.Traditional hyper-spectral image classification method only uses the reflectivity information of hyper-spectral image,and did not take an example by some mature methods and theories of traditional image processing.So we construct a new calculating method in preprocess which use for reference of digital image process,such us morphology method and relations of relative position.Too much spectral information will not only slow down the efficiency of the entire image classification,but also will lead to "Hughes" phenomenon,which means its accuracy of classification will decrease,because of the excess band data.It is highly important to filter out the appropriate bands before we put them into the classification model to start direct classification.It will reduce the time costs,also can improve the accuracy and stability of algorithm.Therefore,this paper presents a pre-processing algorithm of hyper-spectral image,which based on band selection to solve the above-mentioned efficiency and accuracy problems in the pre-process of hyper-spectral image classification.Our experiments show that our theory has a very good effect in the real classification,it shows that even in a different complex classification environment,our algorithm accuracy is stable at a very good level.And in the final process,we introduce the distributed computing system,and put the band selection algorithm into the distributed computing environment,and the time consumed by the band selection algorithm is greatly reduced.Easily to make it in reality. |
PDF Full Text Request