Abstract: Developing a data processing model and relevant algorithm is very important in the process of signal detection, especially when the detected signal has been distorted severely and nonlinearly by strong noise, and it is difficult to find the relational expression of the data. Describes how to use the genetic programming (GP) to solve this problem, and several improvements are given to the algorithm for practical application as follows. The Chebyshev uniformity approximation is used to evaluate the fitness of the individuals. The simulated annealing is applied to the GP to optimize the parameters of relational expression. And a piecewise function-fitting method is applied to reduce the complexity of relational expression with a group evolution strategy proposed to implement this method.