The reflection spectrum shape of the Bragg grating sensor will be distorted even with multi-peaks when there exists great strain gradient in the axis direction of the FBG, so it's meaningful for developing strain profile reconstruction method in structure health monitoring field. With the problems that noised reflection spectrum will decline the precision of strain reconstruction, a novel method based on the support vector regressin (SVR) preprocessing of noised reflection spectrum is proposed for the non-uniformly distributed strain profile reconstruction along the axis of Bragg grating. This method combines fitness sorting improved genetic algorithm with transform-matrix reflection spectrum construction. The reflection spectrum is viewed as a time serial and SVR is developed to constrain the noise for effective regression of reflection spectrum due to its global optimization and well generalization. Subsequently, the strain distribution is treated as a piecewise constant function, the spectral response is calculated using the transform-matrix approach, and then improved genetic algorithm is employed for reconstruction simultaneously. The simulations for strain profile reconstruction are conducted in various kinds of strain distributions. Results show that the SVR reflection spectrum regression is very effective and the proposed algorithm can improve precision of non-uniform strain distribution reconstruction.