Abstract: Abstract: After validating the trend filter is the special result to the LS-SVM arithmetic, eliminating the sounding outliers by the seafloor surface which constructed by LS-SVM .In order to solve the sparseness of LS-SVM results meanwhile restrain the influence of the sample-outliers. A new method of optimize samples by part samples center distance is presented. Some practical multi-beam data is chose to verify the correctness and rationality of the new method. The example shows that on the ground of the optimized train samples, the reasonable seafloor surface could be constructed by LS-SVM arithmetic, and then the outliers of Multi-beam data could be eliminated effectively.