The invention provides an image fusion method based on compressed sensing. The image fusion method includes the three steps of image collection, observed value fusion and image reconstruction. According to the image collection, an image to be fused is divided into image blocks, according to the observed value fusion, a double-channel pulse coupling neural network model is adopted for performing primary image fusion, and a weighted average method is adopted for performing fine fusion on observed values, and finally a final image fusion result is obtained through an image reconstruction algorithm. In the sampling step, characteristics of the image self to be fused are fully considered, and detail information of the result obtained through the fusion is improved; a partitioning compression method is adopted, the partitioned images are sampled and compressed at the same time at a sampling terminal, and complexity, caused by sparse processing performed in advance, of a traditional compressed sensing sampling terminal is avoided; the reconstruction algorithm is high in reconstruction speed and robustness.