Research And Application Of KNN Dust Isolated Point Detection Algorithm Based On KD-Tree

In recent years,due to the over exploitation of natural resources,dust weather has become more and more frequently appear in people's daily production and life,which has brought terrible disasters to the environment of the occurrent region and transit area.Remote sensing technology has good monitoring effect on dust due to its short period to get information and large amount of data.However,some isolated pixels will appear in the dust monitoring results obtained by using the dust pixel identification technology.Most of these isolated pixels are the results of the misjudgement of dust monitoring,which seriously affects the accuracy of the result of dust identification and increases the difficulty of the dust identification work.To solve this problem,based on the KNN outlier detection algorithm,a KD-Tree outlier detection algorithm based on KNN is proposed in this paper and effectively applied to the monitoring of dust isolated points.The research work in this paper is as follows:1.Based on remote sensing data sets have large scale and high dimensional characteristics in analysising of dust isolated point,on the basis of the traditional K nearest neighbor outlier detection algorithm,this paper uses the index structure KD-Tree(k-dimensional tree for short)to efficiently search the key data of multidimensional space,and designs and implements the KNN dust outlier detection algorithm KD-KNN based on KD-Tree for middle and high dimension data.The idea of dimensionality reduction by KD-Tree is applied to ensure the accuracy of classification results,we can improve the efficiency of algorithm classification.2.The KD-KNN outlier detection algorithm proposed in this paper is applied to the classification of information of dust isolated point in Northwest China on 1-20 March 2016.The experimental results show that the algorithm proposed in this paper can effectively improve the performance of the algorithm and reduce the time of classification by KD-Tree,which ensures the accuracy of the KNN algorithm for the classification of dust isolated points and provides an effective means for the classification of dust isolated points.3.This paper designs and develops a remote sensing application prototype system for dust monitoring,the system uses dynamic dust extraction technology to extract dust information all day;For the dust isolated pixel,this system uses methods of the combination of KNN algorithm and the KD-KNN algorithm to classify it,which is convenient for users to compare the classification results.4.By applying the algorithm in this paper and other dust distinguishing technology,The remote sensing application system of dust monitoring has realized the functions of dust information extraction,outlier detection and evaluation,which is able to provide various products for the monitoring of dust storm in quantitative and fine sevice way.