In the field of remote sensing image change detection and segmentation, accurately identifying and segmenting changed regions is critical for dynamic surface monitoring. Existing methods are significantly limited in their detection accuracy due to their inability to effectively capture spatio-temporal feature differences in complex scenarios. To address this challenge, this paper proposes an innovative network named STIKDNet, which first generates a perturbation dataset through data perturbation to enhance the model’s sensitivity and robustness to changes. Then, the original bi-temporal image dataset and the perturbation dataset are jointly fed into STIKDNet for training un-der a knowledge distillation mechanism, where the original dataset and the perturbed dataset are fed into the teacher model and the student model, respectively. By combining the original super-vised loss, the student distillation loss, the supervised loss, and a specially designed loss that is used to precisely measure the output differences between the teacher and student models, the STIKDNet’s parameters are optimized. Experimental results demonstrate that STIKDNet signifi-cantly outperforms state-of-the-art methods on three benchmark datasets LEVIR-CD, SYSU-CD, and WHU-CD, providing a high-precision solution for surface monitoring. The code is available at https://github.com/cyl1238685387/STILKDNet/ .