In order to avoid the obvious drawbacks of traditional image enhancement methods in addressing the demand for multi-scale image enhancement, a laser image multi-scale enhancement based on variable structure deep learning is designed in this paper. Firstly, a variable structure deep learning model is constructed, and in the generative network module, a weighted least squares filter is used to filter and decompose the image. Then, the detail layer of images is enhanced through adaptive enhancement technology. Subsequently, the detail layer is fused with the base layer and reconstructed through the deconvolution layer of the convolutional neural network. In the discrimination network module, the PatchGAN structure is used to distinguish the authenticity of the generated image and the target image, and repeated optimization is achieved through the superposition training of the loss function to enhance the infrared laser image. The experimental results show that this method not only effectively preserves the edges and details of the image, but also achieves a global smooth and delicate effect, significantly improving the contrast and clarity of the image, and at the same time performs well in all the objective evaluation indicators.