Thermal infrared remote sensing is widely used for wildfire monitoring, but its low spatial resolution makes it difficult to accurately delineate fire boundaries, especially in small-scale or edge cases. To address this, we constructed a wildfire image dataset for the Australian region, named WildfireForest-S2, which better aligns the model with the satellite monitoring environment. In addition, we propose WSRNet, an image segmentation approach based on U-Net, specifically designed for satellite-based wildfire detection. WSRNet integrates classification information to refine and guide segmentation features, enhancing the accuracy and efficiency of wildfire detection in satellite images. The model is capable of not only identifying the presence of wildfire events in a given image but also segmenting fire regions with high precision. Experimental validation on the WildfireForest-S2 dataset demonstrates that WSRNet achieves notable improvements in fire region segmentation. At the same time, it also demonstrates strong performance in fire classification, greatly enhancing the accuracy of wildfire monitoring.