A revised 2D Fourier-wavelet descriptor for the object recognition is proposed. The main revision lies in the process of transforming the object image into polar coordinate system. For the purpose of reducing calculating time and improving transforming precision, the redundant pixels in the image are deprived and the rest of them are fractionlized. Using a neural network classifier, this revised descriptor evidently reduces the calculating time and improves the rate of recognition in the experiment of plane type recognition. It is proofed to be a very effective method for feature description of object.