Abstract: Remote sensing image fusion based on detail injection scheme includes two main steps: spatial detail extraction and injection. To ensure the quality of the extracted details and determine the appropriate modulation coefficients, a remote sensing image fusion method via adaptive dictionary learning based convolutional sparse representation is presented. Firstly, this method extracts spatial details from the multispectral and panchromatic images by using guided filter and nondecimated wavelet transform, respectively. Then, the dictionary for extracting spatial details is adaptively learned and introduced into convolutional sparse representation to reconstruct the joint detail image. Finally, the joint details are injected into the upsampled multispectral image by joint discrimination coefficients to obtain the final fusion result. Experimental results indicate that the proposed method outperforms some popular fusion methods both in subjective effect and objective quantitative evaluation.