Digital image correlation (DIC) has the advantages of non-contact, high accuracy, and full-field visibility, making it a prevalent tool for structural health monitoring of stratospheric airship skins. However, real applications need to be run on mobile and embedded devices, and the high computational resource requirement of the DIC method poses a challenge to its implementation. To this end, a lightweight digital image correlation network (LW-DIC-Net) is proposed specifically for DIC applications with limited computational resources. First, we utilize the depth separable convolution to reduce the model complexity. Then, we design an attention module named efficient convolutional block attention module, which is an enhancement of convolutional block attention module adopting the efficient channel attention instead of the general channel attention to guarantee prediction accuracy. Experiments demonstrate that our method achieves a 43.04% reduction in the number of parameters and a 44.75% decrease in floating-point operations per second compared to the original DIC-Net while maintaining the measurement accuracy. With these improvements, the LW-DIC-Net proves to be exceptionally suitable for complex stratospheric environments and resource-constrained mobile and embedded devices, providing an efficient and accurate method for stratospheric airship skin deformation measurements.