Research On High-precision And Fast Detection Technology Of Red Tide Optical Remote Sensing Based On Space Spectrum Information

This paper applies the red tide hyperspectral remote sensing data of the Bayuquan Bay Research Area in Liaodong Bay.Aiming at the problem of high spectral data dimension,large data volume and slow data processing,a rapid detection model of red tide hyperspectral remote sensing based on automatic subspace division is proposed.Aiming at the unstable red tide detection results of single source algorithms such as mutual information RX(Reed-Xiaoli)and traditional correlation coefficient RX,and the possible problems of missed detection and false detection,a decision tree RX high precision red tide detection model based on mutual information and correlation coefficient is proposed.Aiming at the problem of sea surface glint affecting the accuracy of red tide detection,different filtering methods and windows were used to explore the influence of glint suppression on red tide hyperspectral remote sensing detection.The main results of the paper are as follows:1.A rapid detection model of red tide hyperspectral remote sensing based on automatic subspace division is proposed.and compared with the red tide feature band obtained by three methods such as correlation coefficient subspace division combined with maximum information entropy and mutual information subspace division combined with principal component analysis,correlation coefficient of subspace division combined with principal component analysis and original full-band red tide detection results.The results show that the feature band of red tide hyperspectral image(FBH)obtained by the algorithm has better spectral properties.On the basis of maintaining the accuracy of the original red tide detection,the detection time is reduced by nearly 8 times,Red tide detection is performed based on spectral,texture and spatial spectral information for the data 1-FBH and the data 3-FBH,respectively.The overall accuracy and Kappa coefficient of the 5*5 window spatial and spectral information of the data 1-FBH are the highest,reaching 96.91%and 0.93 respectively.The overall accuracy and Kappa coefficient of the data 3-FBH using 5*5 window spatial and spectral information are 94.66%and 0.89,respectively.2.The unsupervised decision tree RX algorithm based on mutual information and correlation coefficient is proposed,and the algorithm is compared with the four classic algorithms such as K-means,SVM,maximum likelihood method and minimum distance method.The results show that the decision tree RX algorithm has better detection effect.Data 2-FBH red tide detection overall accuracy and Kappa coefficient were the highest 95.13%and 0.89,respectively.Data 4-FBH red tide detection overall accuracy and Kappa coefficient were the highest 96.02%and 0.87 respectively.When the two groups of data are in different texture Windows,the detection accuracy is basically unchanged.The overall accuracy of the red tide detection using the 3*3 window spatial and spectral information of the data 2-FBH is up to 95.66%,and the detection accuracy is reduced by 0.15%as the window increases to 5*5.The detection result of 5*5 window spatial and spectral information of the data 4-FBH is the most ideal,and the overall accuracy reaches 96.42%.With the window increasing to 5*5,the detection accuracy is improved by 0.13%.3.Three kinds of filtering methods,such as median,Gaussian low-pass and gamma,and five filtering windows such as 3*3,5*5,7*7,9*9 and 11*11 are used to perform glint suppression,and the influence of different glint suppression methods and windows on the accuracy of red tide hyperspectral detection was discussed.The results show that data 2-FBH adopts median filtering of 9*9 window for the best glint suppression effect,data 4-FBH adopts 7*7 window gamma filter to suppress glint with the best effect.Red tide detection was performed by spectrum,3*3 window spatial and spectral and 5*5 window spatial and spectral information respectively.The overall detection accuracy of data 2-FBH was 98.74%,98.78%and 98.78%,respectively.The overall detection accuracy of data 4-FBH corresponds to 97.71%,97.71%and 97.70%,respectively,indicating that the spatial texture information is increased after glint suppression,which has little effect on the accuracy of red tide detection.