Based on the selection algorithm and corresponding optimization of the kernel function, we put forward a new approach for extracting mineralization information with Support Vector Machine(SVM) applied into remote sensing image in this paper. We synthesize false color using the TM remote sensing data as the test sample, and with the RGB value of the synthetic image as characteristic vector of train samples, with kernel function selection algorithm and man-selected kernel function method, the samples is classified through the SVM algorithm. The test indicated that the radial primary kernel function is the best classification method. It is considered that good recognition effect can be gained for extracting mineralization information with Support Vector Machine(SVM) applied into remote sensing image with the RGB value as the characteristic vector after pre-processing the remote sensing images. The kernel function model selected by LOO estimation can approach the best value satisfactorily.