To solve the problem of low accuracy for mechanical crack detection in corn seeds, a digital image fusion method based on two-channel pulse coupled neural network (PCNN) model was proposed. Firstly, discrete wavelet transform (DWT) and non-subsampled contourlet transform (NSCT) were used to decompose the pretreated mechanical crack images of corn seeds, respectively, in order to obtain their high and low frequency sub-bands. Secondly, the high and low frequency sub-band coefficients are fused by using the improved spatial frequency excitation two-channel PCNN model with different link strengths. Thirdly, the final mechanical crack image of corn seed was obtained by NSCT inverse transformation. The experimental results show that the accuracy of the two-channel PCNN model is 97.2%, and the image entropy, correlation entropy, correlation coefficient and root mean square error are 0.351 1, 1.731 4, 0.983 5 and 0.526 3, respectively, which are better than the LoG, DWT, NSCT and PCNN methods. The execution time of single image by using dual-channel PCNN method is 14.900 7 s, which has the longest running time and the best effect.