Echelle spectrometer gets full spectrum by transient direct reading because of the characteristic of cross-dispersion. The two-dimension spectra received by flat-plane detector needs to be reduced to one-dimension spectra so that the effective wavelength can be detected. Because of huge original data and few effective data, background removal plays an important role of decreasing the amount of data and improving data processing speed. The two-dimension spectrum of echelle spectrometer is analyzed and a suitable background removal algorithms is came up. The edge detection method is applied to diffuse spot detection. Selecting appropriate operator to convolute original image to get edge image and calculating global threshold to segment edge image which can be used to map original image to get the background removed image. Two-dimensional spectral images based on different elements at different integration time are used to judge the effect of different background removal algorithm and different operator are analyzed to figure out their effect of speed and accuracy for algorithm. Experimental result shows that the algorithm came up by this letter is better for image background removal than the others. The background removed image can be used in spectrum reductionand the speed of data processing is notable promoted.