Abstract: Detecting building changes (e.g., new unauthorized constructions or construction areas) along transmission line corridors based on remote sensing data enables the early warning of external damage risks, such as insufficient line-building clearance and collision of large machinery with lines, thereby ensuring power supply safety. However, existing deep learning models for change detection using satellite images suffer from insufficient accuracy due to blurred building target features, significant intra-class scale differences, and susceptibility to complex factors including varying shooting angles and changing radiation conditions. This study proposed a multi-scale difference-sensitive siamese neural network: MSSNet. Specifically, in the multi-scale feature fusion process, a temporal difference attention module (TDAM) was introduced to enhance the weights of key areas' features based on multi-scale feature differences. This effort allows the model to more significantly focus on changed building targets, thereby improving the accuracy of change detection based on satellite images. The module effectiveness was validated through ablation experiments on the S2Looking dataset and the WH measured data. Compared to the baseline model that excludes multi-scale feature fusion and TDAM, the model with a multi-scale feature fusion network increased the accuracy by 2.25 percentage points and 0.99 percentage points on the two datasets, respectively. Furthermore, the incorporation of the TDAM into the feature fusion process enhanced the model accuracy by 3.54 percentage points and 7.76 percentage points, respectively. To validate the detection performance of the MSSNet, comparative experiments were conducted on the S2Looking dataset. The results show that the MSSNet exhibited a precision of 71.76%, a recall of 54.71%, and an F1-score of 62.08%, outperforming other advanced change detection networks prevalent in recent years. The MSSNet effectively improves the accuracy and robustness of satellite image-based building change detection under complex influencing factors, thereby providing reliable technical support for the accurate monitoring and early warning of power grid hazards.