中文摘要：

γ能谱平滑的效果直接影响寻峰、解谱的精度,为此需要在消除放射性计数的统计涨落,提高信噪比的同时最大程度地保留能谱特征信息。本文采用基于结构风险化最小准则的最小二乘支持向量机方法 LSSVM,结合交叉验证参数寻优方法对Na I（Tl）γ能谱进行平滑处理,实验结果表明：LSSVM能谱平滑在全谱上保持较好峰形而不形变,各谱峰位置与理论计算值的最大误差（即峰偏）≤0.6个道址,平均误差为0.43个道址;且该方法具备良好的适应性和推广能力。

英文摘要：

The accuracy of γ spectrum smoothing has a direct effect on the accuracy of peak searching and spectrum unscrambling. Therefore it needs to not only reduce and suppress the statistical fluctuation of radioactive counts and improve noise-signal ratio,but also persist the main character of the spectrum. This paper adopts the improved Support Vector Machine,LSSVM method which based on structural risk minimization criterion（SRM criterion） regression,and then combines with cross validation parameter optimization method,to smoothing Na I（Tl） γ spectrum. The experimental results show that the smooth result of the method based on LSSVM can keep peak form with no deformation,the maximum error of each spectral peak position is no more than 0. 5channel,the average error value is 0. 35 channel,besides,this method has good adaptability and generalization ability.