Remote sensing signal reflected from natural background is of important significance in the field of geography. However the signal we can get is always polluted by additive noise. Since it has been proved that the remote sensing signal reflected from natural background always has some fractal characteristics, just like the background it came from, it is possible for us to deal with it with the theory of fractal. For the perfect analytical function on both time and scale, the wavelet theory is used to analyze the remote sensing signals in this paper. Shannon entropy represents how much information in an information source, so it is possible to estimate the remote sensing signal from noise based on the radio of information entropy at different scales. In this paper, the Shannon entropy of remote sensing signals' wavelet coefficients and that of additive noise in different scales are discussed respectively. And then a method for estimating the Shannon entropy of signal's wavelet coefficients is discussed. Finally, the wavelet coefficients belonging to signal are estimated, and the signal is estimated from the added noise at last. In order to demonstrate the effectiveness of this method, some simulation studies are performed in this paper. Since it doesn't need to estimate the fractal parameter of remote sensing signal, this method is suitable in many situations.