The invention relates to a mass spectrum image super-resolution reconstruction method based on deep learning, and belongs to the image processing field. The mass spectrum image super-resolution reconstruction method based on deep learning includes the following steps: using a matrix-assisted laser desorption / ionization mass spectrometry imaging instrument to perform imaging on a biological sample,according to different m / z values, obtaining multiple mass spectrum images, obtaining multiple low resolution images through down-sampling, solving the morphological information of the acquired high / low resolution images and taking the morphological information as a training sample set, designing and training deep convolutional neural networks, according to the input training sample set, predicting the missing information in the low resolution images; during the test stage, by means of the prior knowledge obtained through networks, guiding reconstruction of the morphological information of the low resolution mass spectrum images; and reconstructing the morphological information into high resolution mass spectrum images by establishing a partial differential equation. The mass spectrum image super-resolution reconstruction method based on deep learning avoids the defect that a traditional technology improves the imaging quality by improving imaging equipment or by means of repeated sampling, can reduce the cost and the experimental cycle, and can breakthrough limitation of image resolution for a hardware system.