Based Support Vector Machine Retrieval Model For Ocean Suspended Solids Remote Sensing Concentration

In recent years, with China's rapid economic development and people's livingstandards significantly improve, a large number of industrial and domestic wastewater discharged into the South China Sea, this caused serious pollution of coastalwaters in the South China,Sea marine water quality decreased. The suspended solid-s content of the water is one of the very important water quality parameters, Its co-ntent in a direct impact on water transparency, turbidity, color and other optical prope-rties, it is important for marine water quality monitoring.In this paper, I used the measured concentration of suspended solids and thesynchronous optical data (TM image) and radar data (Radarsat-2images) to obtain t-he characteristic factor of optical and radar. Then I established the support vectormachine inverse model of the suspended solids concentration, the specificsteps are as follows:(1) Data preparation and pretreatment. Before conductingresearch, I preprocessed on the TM images include geometric correction, atmosphericcorrection to eliminate atmospheric interference; I preprocessed on the Radarsat-2images include filtering and mask handling to eliminate the speckle noise.(2) I usedCorrelation analysis on TM images of each band and the band combination of spectralreflectance and the concentration of suspended solids to obtain the characteristicfactor of optical, it is TM2,TM3and(TM2+TM3)(/TM2/TM3).Then I establishedthe optical inverse model of the suspended solids concentration, it is y=0.002x~2-0.01x-0.65, x=TM3.(3) I used Radarsat-2images four polarization back scatteringcoefficient to establish the radar inverse model of the suspended solids concentration,it is y=-0.01x~2+0.03x+9.68,x=HH.(4) I get the support vector machine inverse modelof the suspended solids concentration. Its input parameter is VV, HH, TM2, TM3, and(TM2,+TM3)/(TM2,/TM3); Its kernel function is RBF kernel function; Its inte-rnal parameters is C=100,σ~2=0.1.Through this study, I get the following conclusions:(1) the concentrateo-n of suspended solids and TM images of each band spectral reflectance andRadarsat-2images four polarization back scattering coefficient is Positive correlation.(2) The optical model and radar model using a single data effect is inferior to linearregression model and SVM inversion model combination of optical and radar data.Simple optical remote sensing and radar remote sensing has its own can not overcome the shortcomings, but the combination of the two can achieve complementary advent-ages.(3) On combination of methods, the SVM is superior to a simple multiple linearregression analysis. SVM can well solve the problem of small sample, nonlinear,high dimension, etc. these features is suitable for complexity of inversion SSC.