Abstract: In order to realize aircraft intelligent maintenance technology by using images of airframe damage region, an airframe damage transition region extraction method based on improved spectral clustering is proposed. The characteristic of airframe damage neighboring region was analyzed, on the basis of which multi-dimension gray entropy value was used as similarity measurement. Spectra has been constructed and dimension of weights has been reduced. The critical value of cluster was the weight which was corresponding to the first minimum point of statistical group. By improving eigenvalue calculation and eigenvector selection of spectral clustering method, information loss causing by these two steps is avoided. Finally, validation verification was performed using several airframe damage images.