Abstract: Signal analysis has important theoretical and practical application.Non-stationary signal analysis and processing is one of the hot topics in the scientific and engineering research area.Because of the limit of linearity and stationarity assumption, the traditional methods can not be effectively used in image processing, speech processing and radar signal processing. A model suiting for non-linear and non-stationary is established. The empirical data decomposition algorithm is discussed. A suitable design criteria is established. The use of cubic spline functions to predict the parameters of the predictive filter is discussed. Making a test on spectrum image data with empirical data decomposition. The system is simulated in Matlab. The probability distribution of the samples in high-frequency subbands whose values are within the specified range and the corresponding entropy are analyzed through simulation. The results show that the high-frequency coefficients produed by empirical data decomposition algorithm is more concentrated than those of 5/3 wavelet, which is useful to image compression, and also proved empirical data decomposition is an effective analysis method for non-stationary image data.