Detecting and monitoring fire-affected areas in remote sensing imagery plays a crucial role in mitigating the impact of wildfires and aiding in disaster management. This paper presents an investigation into the application of semantic segmentation for fire-affected area detection in remote sensing imagery. Our methodology involves preprocessing the remote sensing images, followed by the application of deep learning-based semantic segmentation models to extract meaningful features and classify pixels into fire and non-fire classes. Furthermore, we employ a transductive approach to further enhance the performance of the model on specific test dataset. We evaluate the performance of our approach on the ChaBuD dataset of fire-affected areas and compare it with baseline methods. The experimental results demonstrate the effectiveness and superiority of our proposed semantic segmentation approach in accurately detecting fire-affected areas, achieving high IoU. Code is available at https://github.com/woshiluren/ECML .