The full width at half maximum (FWHM) of spectra of acousto-optic tunable filter (AOTF) and non-ideal transducer structure result directly in image degradation. In order to increase the spatial resolution of hyperspectral images, the expectation-maximization algorithm is ultilized and optimized in pre-processing, and it could also be used even when degration is not accurately estimated. The original target is ultinreately approached by iterative method. Experimental results show that the algorithm does not depend on the assumption of the image spliced periodically, and thus efficiently avoids the defects of the traditional deconvolution method. This method increases the spatial resolution of the image, and improves image quality. It plays a significant role in improving image quality of AOTF hyperspectral image.