The spatial resolution of a sub-pixel imaging system can be improved by increasing temporal and spatial sampling frequencies of detectors. However, the data collected by detectors are prone to aliasing and the resolution of reconstructed image is far away from the ideal value. In this paper, an algorithm of super-resolution reconstruction was proposed based on sub-pixel imaging achieved by three linear array detectors. Firstly, an interpolation model on high-resolution grid was established. Then, blur kernels in an image with high-resolution were identified in linear array and scanning directions respectively, from which the blur kernel in a frame was obtained. Finally, a gradient smoothing regularization model with Neumman boundary conditions was employed to deblur and inhibit ringing effects. Experimental results show that the system resolution of sub-pixel imaging by the proposed algorithm is 2.6 times that of non-oversampling imaging system of a single linear detector, and the Gray Mean Grade ( GMG) is improved by 7.71 as compared to that of the bilinear interpolation algorithm. The algorithm can achieve super-resolution reconstruction for sub-pixel imaging systems with more linear array detectors and can obtain a higher system resolution.