Abstract: Noise elimination is one of the important step in image preprocessing. T'his paper, taking the imaging spectrometer data of Shunyi, Beijing as an example, deals with the research on noise analysis and elimination.Statis-tical met hod, combined with image processing technique, is used to analyze the existence and feature of noise. There are two types of noise in imaging spectrometer image of Shunyi.One type takes the form of speckleswhich exist in all the channels at random. The other type takes the form of strips which distribute in dark ob-jects such as river and pond. There are dfferent types noise in remote sensing imagery which influence the quali-ty of image along the scanning direction and appear only in channel 18, 19, 20 and 21. The noise is eliminated bydifferent methods according to the feature of noises.T'he speckle noise is eliminated by median filler quite well. While the second type noise is very special and can not be eliminated by conventional method. Therefore, a localneighboring algorithm is developed to eliminate it. The procedure is: automatically searching the noise along theline of image, when some pixel meets the demand of limit preset, replace its grey value by the average value of 5×5 pixel matrix in neighbor with uniform value. In this way the special noise can be eliminated effectivelywithout damaging the raw data because the algorithm don 't perform any processing on normal pixels.
lt can be seen from the processing result of study area that the proposed method is effective in analyzing andeliminating noise of imaging spectrometer data.

Key words: imaging spectrometer data, noise, elimination.