In the process of low resolution laser image processing, a sparse representation based method is used to achieve high-resolution image reconstruction. The obtained feature information is blurred, the peak signal-to-noise ratio of fuzzy features is high, resulting in poor image reconstruction effect. Therefore, a high-resolution reconstruction method for low resolution laser images under visual communication is proposed. Using sensitivity difference algorithm under visual communication technology to process low resolution laser images, the feature information contained in laser images has a high substantive effect. Using the quadtree partitioning idea to partition the preprocessed laser image, forming multiple image sub regions. Starting from each sub region image, input it into the Smooth PCANet for deep learning to extract deep level detail features of the image. Based on the sparse regularization model and combined with non local approximation prior constraints, high-resolution reconstruction of laser images is achieved. The experimental results show that using new research methods to process different types of laser images, the peak signal-to-noise ratio of the reconstructed images always exceeds 46 dB, and the structural similarity is above 0.96, achieving a significant improvement in image resolution.